Krea2EditModelPatch.patch.<locals>.wrapper() takes from 4 to 6 positional arguments but 7 were given

[Jandown]

Error reported after updating to v1.2

[suurtollxi]

Also have this problem after updating ComfyUI and nodes yeasterday.

[yaz152]

I had this too and put it into AI and it said change this line in ComfyUI\custom_nodes\comfyui-krea2edit\ __init__.py
def wrapper(executor, x, timesteps, context, attention_mask=None, transformer_options={}, **kwargs):

to

def wrapper(executor, x, timesteps, context=None, attention_mask=None, transformer_options=None, *args, **kwargs):
if transformer_options is None:
transformer_options = {}

Originally it was just add the *args, but that caused a second error further in the workflow. This replacement solved the problem for me and everything is working great.

[akashzeno]

This is from the ComfyUI nightly (not v1.2): upstream commit c9602625 added a new ref_latents positional argument to SingleStreamDiT.forward, inserted between attention_mask and transformer_options. The wrapper is called positionally, so it now receives 7 positional args instead of 6 — hence takes from 4 to 6 positional arguments but 7 were given.

Fix is in #9. The minimal correct change is to add ref_latents=None in that exact position:

def wrapper(executor, x, timesteps, context, attention_mask=None, ref_latents=None, transformer_options={}, **kwargs):

⚠️ Heads-up on the *args / transformer_options=None workarounds: because ref_latents now comes before transformer_options, those signatures bind the wrapper's transformer_options to ref_latents and push the real transformer_options into *args, where it gets dropped — so the node then runs with an empty transformer_options. It looks fine on a basic graph, but it silently loses anything carried on transformer_options (attention/patch wrappers, other active model patches). Putting ref_latents in the correct slot keeps it intact.

[yaz152]

Thanks. I corrected my file to match your edit and everything works.

[fritzfantom23]

could someone please explain which file needs to be edited? I also have the same error with Krea2T Enhancer

[KarterDesign]

could someone please explain which file needs to be edited? I also have the same error with Krea2T Enhancer

Replace the entire ComfyUI\custom_nodes\comfyui-krea2edit\__init__.py file with this one:

"""ComfyUI-Krea2Edit — in-context edit forward for the Krea2 model.

ComfyUI's native Krea2 `_forward` is text-to-image only: it builds the sequence
`[text | target]`. The krea2_edit LoRA (trained in ai-toolkit) needs the *appearance
path*: the VAE-encoded SOURCE latent prepended as a block of clean tokens, distinguished
from the (noisy) target purely by the 3-axis RoPE frame index (source=1, target=0, h/w
aligned). This node adds that by wrapping the model's DIFFUSION_MODEL forward and rebuilding
the sequence as `[text | source(frame=1) | target(frame=0)]`, keeping only the target tokens
out — mirroring ai-toolkit's `predict_velocity_edit` exactly, using the model's own submodules.

Wiring:  LoadImage -> VAEEncode(source) --\
                                            Krea2EditModelPatch(model, source_latent) -> KSampler
         UNETLoader -> LoraLoaderModelOnly -/
KSampler.latent_image <- EmptySD3LatentImage (noise). Text: NATIVE krea2 CLIP + CLIPTextEncode.
"""
import math

import torch
import torch.nn.functional as F
from einops import rearrange

import comfy.patcher_extension
import comfy.utils
import comfy.ldm.common_dit
from comfy.ldm.flux.layers import timestep_embedding
from comfy.ldm.flux.math import apply_rope
from comfy.ldm.modules.attention import optimized_attention_masked


def _imgids(bs, frame, h_, w_, device):
    ids = torch.zeros(h_, w_, 3, device=device, dtype=torch.float32)
    ids[..., 0] = frame
    ids[..., 1] = torch.arange(h_, device=device, dtype=torch.float32)[:, None]
    ids[..., 2] = torch.arange(w_, device=device, dtype=torch.float32)[None, :]
    return ids.reshape(1, h_ * w_, 3).repeat(bs, 1, 1)


def _imgids_offset(bs, frame, gh, gw, th, tw, device):
    """Stride-1 integer positions at a centered integer offset. For `fit` refs the
    pixels are already resampled to target grid density, so the position grid is
    stride-1 BY CONSTRUCTION — scaling it again only manufactures skip/collision
    artifacts. Requires gh<=th, gw<=tw (guaranteed by the floor+cap in fit)."""
    off_h, off_w = max(0, (th - gh) // 2), max(0, (tw - gw) // 2)
    ids = torch.zeros(gh, gw, 3, device=device, dtype=torch.float32)
    ids[..., 0] = frame
    ids[..., 1] = (torch.arange(gh, device=device, dtype=torch.float32) + off_h)[:, None]
    ids[..., 2] = (torch.arange(gw, device=device, dtype=torch.float32) + off_w)[None, :]
    return ids.reshape(1, gh * gw, 3).repeat(bs, 1, 1)


def _to_4d(v):
    """(B,C,T,H,W) -> (B*T,C,H,W); pass 4D through. Images use T=1."""
    if v.ndim == 5:
        b, c, t, h, w = v.shape
        return v.reshape(b * t, c, h, w)
    return v


def _fit_src(src, H, W):
    """Fit a source latent to the target grid the way TRAINING did: center-crop to
    the target aspect ratio, then resize. A plain interpolate (the pre-fix behavior)
    STRETCHES mixed-AR sources — users saw stretched people whenever their input AR
    differed from the output resolution."""
    sh, sw = src.shape[-2:]
    if (sh, sw) == (H, W):
        return src
    s = max(H / sh, W / sw)
    ch, cw = min(sh, int(round(H / s))), min(sw, int(round(W / s)))
    y0, x0 = (sh - ch) // 2, (sw - cw) // 2
    src = src[..., y0:y0 + ch, x0:x0 + cw]
    return F.interpolate(src.float(), size=(H, W), mode="bilinear")


def _fit_encode_image(image, vae, H, W, cache, key, fit_mode="crop"):
    """Pixel-space source prep (blur-proof path): center-crop the IMAGE to the
    target AR, resize to the exact target pixel grid, VAE-encode. Latent-space
    resizing (the old fallback) softens VAE latents — this path never resizes
    latents at all. Cached per target resolution (encode once, not per step)."""
    key = key + (fit_mode,)
    if key in cache:
        return cache[key]
    print(f"[krea2edit] _fit_encode_image: mode={fit_mode} in={tuple(image.shape)} target_latent={H}x{W}", flush=True)
    px_h, px_w = H * 8, W * 8
    img = image.movedim(-1, 1)  # B,H,W,C -> B,C,H,W
    ih, iw = img.shape[-2:]
    if fit_mode == "fit":
        # "bilinear" answer to scale mismatch: resample CONTENT (pixel space, bicubic)
        # to the target's grid density instead of moving positions. AR-preserving
        # fit-inside, no crop, no grey canvas — the forward places it at an integer
        # centered offset (scaled-pos with s=1 -> stride 1, no rounding artifacts).
        sc = min(px_h / ih, px_w / iw)
        # NEAR-MATCHED AR: fill the target grid EXACTLY via a minimal center-crop.
        # Fit-inside margins of 1-2 tokens are not harmless: target edge columns
        # with no ref correspondence get filled by repeating adjacent ref content
        # (2026-07-14 edge-duplication bug: ref (74,54) vs target (74,56)).
        # This also restores the design promise fit == crop at matched AR.
        CROP_TOL = 0.08
        if ih * sc >= px_h * (1 - CROP_TOL) and iw * sc >= px_w * (1 - CROP_TOL):
            s = max(px_h / ih, px_w / iw)
            ch, cw = min(ih, int(round(px_h / s))), min(iw, int(round(px_w / s)))
            y0, x0 = (ih - ch) // 2, (iw - cw) // 2
            img = img[..., y0:y0 + ch, x0:x0 + cw]
            nh, nw = px_h, px_w
        else:
            # genuine AR mismatch: MUST match the trainer's _fit_prep EXACTLY
            # (krea2_edit.py) — /16 floor snap capped at the target's /16 floor.
            # The model is trained on this geometry; a /8-round node grid would
            # produce a different ref latent size -> different centered offset ->
            # a visible margin-boundary seam even from a well-trained model
            # (train/infer geometry must be byte-identical). 2026-07-15 alignment.
            nh = min(max(16, int(ih * sc) // 16 * 16), max(16, px_h // 16 * 16))
            nw = min(max(16, int(iw * sc) // 16 * 16), max(16, px_w // 16 * 16))
        img = F.interpolate(img.float(), size=(nh, nw), mode="bicubic", antialias=True)
        lat = vae.encode(img.movedim(1, -1)[..., :3].clamp(0, 1))
        cache[key] = lat
        return lat
    # crop (default / "v1 legacy"): center-crop to the target AR, then resize.
    s = max(px_h / ih, px_w / iw)
    ch, cw = min(ih, int(round(px_h / s))), min(iw, int(round(px_w / s)))
    y0, x0 = (ih - ch) // 2, (iw - cw) // 2
    img = img[..., y0:y0 + ch, x0:x0 + cw]
    img = F.interpolate(img.float(), size=(px_h, px_w), mode="bicubic", antialias=True)
    lat = vae.encode(img.movedim(1, -1)[..., :3].clamp(0, 1))
    cache[key] = lat
    return lat


def _ref_attn_bias(boosts, boost_mask, txtlen, slens, tgtlen, mask_hw, device, dtype):
    """Additive attention-logit bias on the [text | refs... | target] sequence.

    boosts: per-ref factor on target->ref attention, aligned with the source blocks
    (last entry = last ref = the subject by workflow convention). Equivalent to
    multiplying those keys' post-softmax attention weight before renormalization.
    boost_mask (ComfyUI MASK, ref-image pixel space) restricts the LAST ref's boost
    to a region (e.g. the face).
    """
    nsrc = len(slens)
    offs = [txtlen]
    for sl in slens:
        offs.append(offs[-1] + sl)
    rows0 = offs[-1]
    L = rows0 + tgtlen
    bias = torch.zeros(1, 1, L, L, device=device, dtype=dtype)
    for i, b in enumerate(boosts):
        if b == 1.0:
            continue
        off, sl = offs[i], slens[i]
        if boost_mask is not None and i == nsrc - 1 and mask_hw is not None:
            mask = boost_mask[:1]
            if mask.ndim == 2:
                mask = mask[None]
            mask = F.interpolate(mask[None].float(), size=mask_hw[i], mode="area")[0, 0]
            cols = off + torch.nonzero(mask.reshape(-1) > 0.5, as_tuple=True)[0].to(device)
        else:
            cols = torch.arange(off, off + sl, device=device)
        bias[:, :, rows0:, cols] = math.log(max(b, 1e-4))
    return bias


def krea2_edit_forward(m, x, timesteps, context, src_latent, transformer_options,
                       ref_boost=1.0, ref_boost_a=1.0, ref_boost_mask=None,
                       ref_native=False, pos_mode="anchor"):
    """Krea2 SingleStreamDiT._forward, but with source block(s) prepended.

    m           : the SingleStreamDiT (LoRA-patched at sample time)
    x           : (B,C,H,W) or (B,C,T,H,W) noisy TARGET latent
    src_latent  : clean SOURCE latent (VAE-encoded), 4D/5D — or a LIST of them
                  (multi-ref: [scene, subject], frames 1..N, training-matched)
    context     : (B, seq, txtlayers*txtdim) — the 12-layer Qwen3-VL stack
    """
    patch = m.patch

    # Mirror ComfyUI _forward: latents may arrive 5D (B,C,T,H,W) for this model.
    temporal = x.ndim == 5
    if temporal:
        b5, c5, t5, h5, w5 = x.shape
    x = _to_4d(x)
    bs, c, H_orig, W_orig = x.shape

    x = comfy.ldm.common_dit.pad_to_patch_size(x, (patch, patch), padding_mode="replicate")
    H, W = x.shape[-2], x.shape[-1]
    h_, w_ = H // patch, W // patch

    # source(s) -> (bs, C, H, W): flatten temporal, match batch, fit to the target grid
    # (center-crop to target AR then resize — training-matched; never stretch).
    src_list = src_latent if isinstance(src_latent, (list, tuple)) else [src_latent]
    srcs = []
    for sl in src_list:
        src = _to_4d(sl).to(x.device, x.dtype)
        if src.shape[0] != bs:
            src = src[:1].expand(bs, *src.shape[1:])
        if not ref_native and src.shape[-2:] != (H, W):
            print(f"[krea2edit] LATENT-PATH fit_src (crop): src={tuple(src.shape[-2:])} -> {H}x{W}", flush=True)
            src = _fit_src(src, H, W).to(x.dtype)
        srcs.append(comfy.ldm.common_dit.pad_to_patch_size(src, (patch, patch), padding_mode="replicate"))
    src_grids = [(s_.shape[-2] // patch, s_.shape[-1] // patch) for s_ in srcs]

    context = m._unpack_context(context)                       # (B, seq, 12, 2560)

    tgt_img = m.first(rearrange(x, "b c (h ph) (w pw) -> b (h w) (c ph pw)", ph=patch, pw=patch))
    src_imgs = [m.first(rearrange(s_, "b c (h ph) (w pw) -> b (h w) (c ph pw)", ph=patch, pw=patch))
                for s_ in srcs]

    t = m.tmlp(timestep_embedding(timesteps, m.tdim).unsqueeze(1).to(tgt_img.dtype))
    tvec = m.tproj(t)

    context = m.txtfusion(context, mask=None, transformer_options=transformer_options)
    context = m.txtmlp(context)

    txtlen, tgtlen = context.shape[1], tgt_img.shape[1]
    srclen = sum(si.shape[1] for si in src_imgs)
    combined = torch.cat([context] + src_imgs + [tgt_img], dim=1)  # [text | refs... | target]

    device = combined.device
    if pos_mode == "stride1" and ref_native:
        print(f"[krea2edit] STRIDE1-POS fit: ref grids {src_grids} centered in ({h_},{w_})", flush=True)
        if any(h_ - gh > 2 or w_ - gw > 2 for gh, gw in src_grids):
            print("[krea2edit] NOTE: fit margins >2 tokens (large source/output aspect-ratio "
                  "gap). fit is trained for matched/near-matched AR; for a big AR change "
                  "prefer 'crop', or set the output AR closer to the source.", flush=True)
        ref_ids = [_imgids_offset(bs, i + 1, gh, gw, h_, w_, device)
                   for i, (gh, gw) in enumerate(src_grids)]
    else:
        ref_ids = [_imgids(bs, i + 1, gh, gw, device) for i, (gh, gw) in enumerate(src_grids)]
    pos = torch.cat([
        torch.zeros(bs, txtlen, 3, device=device, dtype=torch.float32)]   # text @ 0
        + ref_ids
        + [_imgids(bs, 0, h_, w_, device)],                                    # target frame=0
        dim=1)
    freqs = m.pe_embedder(pos)

    attn_bias = None
    if ref_boost != 1.0 or ref_boost_a != 1.0:
        # last ref = subject (single-ref: the only ref); earlier refs (scene) get ref_boost_a
        boosts = [ref_boost_a] * (len(src_imgs) - 1) + [ref_boost]
        attn_bias = _ref_attn_bias(boosts, ref_boost_mask, txtlen,
                                   [si.shape[1] for si in src_imgs], tgtlen,
                                   src_grids, combined.device, combined.dtype)

    for block in m.blocks:
        combined = block(combined, tvec, freqs, attn_bias, transformer_options=transformer_options)

    final = m.last(combined, t)
    out = final[:, txtlen + srclen: txtlen + srclen + tgtlen, :]         # target tokens only
    out = rearrange(out, "b (h w) (c ph pw) -> b c (h ph) (w pw)",
                    h=h_, w=w_, ph=patch, pw=patch, c=m.channels)
    out = out[:, :, :H_orig, :W_orig]
    if temporal:
        out = out.reshape(b5, t5, m.channels, H_orig, W_orig).movedim(1, 2)
    return out


class Krea2EditModelPatch:
    @classmethod
    def INPUT_TYPES(cls):
        return {"required": {
            "model": ("MODEL",),
            "source_latent": ("LATENT",),
        }, "optional": {
            "source_latent_b": ("LATENT", {"tooltip": "2nd reference (subject photo) for multi-ref LoRAs -> RoPE frame=2, training-matched order: scene first, subject second"}),
            "ref_boost": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1000.0, "step": 0.01, "round": 0.001,
                                     "tooltip": "reference-fidelity dial: multiplies target->reference attention. Applies to the LAST ref (= the subject in two-ref workflows, the only ref in single-ref). 1.0 = off, >1 pulls harder toward the reference's appearance, <1 loosens. Optimal value is model-specific"}),
            "ref_boost_a": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1000.0, "step": 0.01, "round": 0.001,
                                       "tooltip": "same dial for the FIRST ref (= the scene in two-ref workflows). No effect in single-ref workflows. 1.0 = off"}),
            "fit_mode": (["fit", "crop (legacy)"], {"default": "fit",
                          "tooltip": "how an image source fits a mismatched output aspect ratio (needs vae + source_image connected): fit = resample the source to the target grid at a centered offset — matches how this model was trained (default, use this); crop (legacy) = center-crop to the target AR then resize (v1/v1.1 geometry, only for older weights)"}),
            "ref_boost_mask": ("MASK", {"tooltip": "optional region on the (last) reference to boost, e.g. the face; empty = whole reference"}),
            "vae": ("VAE", {"tooltip": "RECOMMENDED with source_image: enables the blur-proof pixel-space path (crop+resize in pixels, encode internally) — immune to input/output resolution mismatches"}),
            "source_image": ("IMAGE", {"tooltip": "source as IMAGE (with vae connected): overrides source_latent with exact pixel-space fitting — fixes blurry results from mismatched resolutions"}),
            "source_image_b": ("IMAGE", {"tooltip": "2nd reference as IMAGE (with vae)"}),
        }}

    RETURN_TYPES = ("MODEL",)
    FUNCTION = "patch"
    CATEGORY = "krea2edit"
    DESCRIPTION = "Adds the krea2_edit in-context source-preservation path (source latent as frame=1 tokens) to a Krea2 model."

    def patch(self, model, source_latent, source_latent_b=None, ref_boost=1.0, ref_boost_a=1.0,
              ref_boost_mask=None, vae=None, source_image=None,
              source_image_b=None, fit_mode="fit"):
        m = model.clone()
        # The target latent reaches the diffusion model already scaled (process_latent_in);
        # scale the source(s) the same way so all share one latent space.
        src_samples = model.model.process_latent_in(source_latent["samples"])
        if source_latent_b is not None:
            src_samples = [src_samples, model.model.process_latent_in(source_latent_b["samples"])]

        px_cache = {}   # pixel-path encoded sources, keyed per target resolution
        mm = model.model  # for process_latent_in on the pixel path

        if fit_mode == "fit" and (vae is None or source_image is None):
            print(f"[krea2edit] WARNING: fit_mode='fit' has NO EFFECT — it needs both "
                  f"'vae' and 'source_image' connected (the pixel path). Falling back to the "
                  f"latent crop path.", flush=True)

        def wrapper(executor, x, timesteps, context, *rest, **kwargs):
            # ComfyUI calls DIFFUSION_MODEL wrappers with a positional arg list whose
            # length/order has changed across versions (e.g. an extra `y` slot was
            # added before `transformer_options` in newer releases: x, timestep,
            # context, y, control, transformer_options, ...). Rather than hard-coding
            # a fixed number of positional slots (which breaks with a TypeError like
            # "takes from 4 to 6 positional arguments but 7 were given" whenever
            # ComfyUI's internal signature changes), pull transformer_options out by
            # keyword if present, else take it as the last dict found in the extra
            # positional args — everything else in `rest` (y, guidance, control...)
            # is irrelevant to this node's logic.
            if "transformer_options" in kwargs:
                transformer_options = kwargs["transformer_options"]
            else:
                transformer_options = next((a for a in reversed(rest) if isinstance(a, dict)), {})
            dm = executor.class_obj  # the SingleStreamDiT instance
            src = src_samples
            if vae is not None and source_image is not None:
                if not px_cache:
                    print(f"[krea2edit] pixel path ACTIVE (fit_mode={fit_mode})", flush=True)
                xx = _to_4d(x)
                Hh, Ww = xx.shape[-2], xx.shape[-1]
                lat = mm.process_latent_in(_fit_encode_image(source_image, vae, Hh, Ww, px_cache, ("a", Hh, Ww), fit_mode))
                if source_image_b is not None:
                    lat = [lat, mm.process_latent_in(_fit_encode_image(source_image_b, vae, Hh, Ww, px_cache, ("b", Hh, Ww), fit_mode))]
                src = lat
            v = krea2_edit_forward(dm, x, timesteps, context, src, transformer_options,
                                   ref_boost=ref_boost, ref_boost_a=ref_boost_a,
                                   ref_boost_mask=ref_boost_mask,
                                   ref_native=(fit_mode == "fit" and vae is not None
                                               and source_image is not None),
                                   pos_mode=("stride1" if fit_mode == "fit" else "anchor"))
            return v

        to = m.model_options.setdefault("transformer_options", {})
        comfy.patcher_extension.add_wrapper_with_key(
            comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, "krea2_edit", wrapper, to
        )
        return (m,)


class Krea2EditGroundedEncode:
    """Image-grounded instruction encode — the SEMANTIC path of krea2_edit.

    Training always encodes the instruction TOGETHER with the source image through
    Qwen3-VL (user turn = <vision tokens: source> + instruction) and taps 12 layers.
    Stock CLIPTextEncode is text-only, so inference was running with the grounding
    half of the recipe missing (the VAE source tokens carry appearance; THIS carries
    scene semantics: "the man on the left", "the sign in the back").

    Requires a qwen3vl TE checkpoint WITH the vision tower (all local ones have it).
    grounding_px caps the longest side fed to the VLM — the 2026-07-02 LoRA trained
    with 384-768px jitter, so 640-768 is in-distribution; 0 = native res (the jitter
    training makes that tolerable too). For CFG, ground the NEGATIVE too: second node,
    empty prompt, same image (matches training's unconditional).
    """
    DEFAULT_SYSTEM = (
        "Describe the image by detailing the color, shape, size, "
        "texture, quantity, text, spatial relationships of the objects and background:"
    )

    KREA2_EDIT_TEMPLATE = (
        "<|im_start|>system\nDescribe the image by detailing the color, shape, size, "
        "texture, quantity, text, spatial relationships of the objects and background:"
        "<|im_end|>\n<|im_start|>user\n<|vision_start|><|image_pad|><|vision_end|>"
        "{}<|im_end|>\n<|im_start|>assistant\n"
    )

    @classmethod
    def _template(cls, nimg, system_prompt=""):
        sp = system_prompt.strip() or cls.DEFAULT_SYSTEM
        vis = "<|vision_start|><|image_pad|><|vision_end|>" * nimg
        return ("<|im_start|>system\n" + sp + "<|im_end|>\n<|im_start|>user\n"
                + vis + "{}<|im_end|>\n<|im_start|>assistant\n")

    @classmethod
    def INPUT_TYPES(cls):
        return {
            "required": {
                "clip": ("CLIP",),
                "prompt": ("STRING", {"multiline": True, "default": ""}),
            },
            "optional": {
                "image": ("IMAGE",),
                "image_b": ("IMAGE", {"tooltip": "2nd reference (subject) for multi-ref LoRAs; vision blocks in training order: scene, subject"}),
                "grounding_px": ("INT", {"default": 768, "min": 0, "max": 4096, "step": 64,
                                          "tooltip": "cap longest side fed to Qwen3-VL; 0 = native"}),
                "system_prompt": ("STRING", {"multiline": True, "default": "",
                                              "tooltip": "advanced (optional): override the grounding system prompt (empty = training default). Steers what the vision encoder attends to, e.g. facial identity detail."}),
            },
        }

    RETURN_TYPES = ("CONDITIONING",)
    FUNCTION = "encode"
    CATEGORY = "krea2edit"
    DESCRIPTION = "Encodes the edit instruction grounded on the source image (training-matched semantic path)."

    KREA2_EDIT_TEMPLATE_2REF = (
        "<|im_start|>system\nDescribe the image by detailing the color, shape, size, "
        "texture, quantity, text, spatial relationships of the objects and background:"
        "<|im_end|>\n<|im_start|>user\n<|vision_start|><|image_pad|><|vision_end|>"
        "<|vision_start|><|image_pad|><|vision_end|>"
        "{}<|im_end|>\n<|im_start|>assistant\n"
    )

    def _prep(self, image, grounding_px):
        samples = image.movedim(-1, 1)  # B,H,W,C -> B,C,H,W
        h, w = samples.shape[2], samples.shape[3]
        if grounding_px and max(h, w) > grounding_px:
            s = grounding_px / max(h, w)
            samples = comfy.utils.common_upscale(samples, round(w * s), round(h * s), "area", "disabled")
        return samples.movedim(1, -1)[:, :, :, :3]

    def encode(self, clip, prompt, image=None, image_b=None, grounding_px=768, system_prompt=""):
        if image is None:  # text-only fallback = old behavior
            tokens = clip.tokenize(prompt)
            return (clip.encode_from_tokens_scheduled(tokens),)
        imgs = [self._prep(image, grounding_px)]
        if image_b is not None:
            imgs.append(self._prep(image_b, grounding_px))
        template = self._template(len(imgs), system_prompt)
        tokens = clip.tokenize(prompt, images=imgs, llama_template=template)
        return (clip.encode_from_tokens_scheduled(tokens),)


NODE_CLASS_MAPPINGS = {
    "Krea2EditModelPatch": Krea2EditModelPatch,
    "Krea2EditGroundedEncode": Krea2EditGroundedEncode,
}
NODE_DISPLAY_NAME_MAPPINGS = {
    "Krea2EditModelPatch": "Krea2 Edit (source patch)",
    "Krea2EditGroundedEncode": "Krea2 Edit (grounded encode)",
}

[lbouaraba]

Fixed in v1.2.1 (and v1.2.2 is out now with a version banner so this kind of thing is easier to spot). This was ComfyUI upstream commit c9602625 adding a new ref_latents argument that shifted the call signature the node wraps.

Update the node pack (ComfyUI Manager -> Update, or git pull in custom_nodes/comfyui-krea2edit) and restart ComfyUI fully. After the update you should see "[krea2edit] nodes v1.2.2 loaded" in the console at startup, which confirms the new version is active.

Thanks everyone for the reports. Closing, but reopen if you still hit it after updating to 1.2.2.

[ameen-roayan]

I am actually on 1.2.2 and i am still getting the error
comfyui Updated as well

`Node threw an error during execution.

ComfyUI Error Report

Error Details

• Node ID: 85
• Node Type: Krea2EditGroundedEncode
• Exception Type: TypeError
• Exception Message: TypeError: scaled_dot_product_attention() takes from 3 to 6 positional arguments but 7 were given

Stack Trace

  File "D:\Comfyui\ComfyUI\execution.py", line 543, in execute
    output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
                                                              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\execution.py", line 342, in get_output_data
    return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-lora-manager\py\metadata_collector\metadata_hook.py", line 171, in async_map_node_over_list_with_metadata
    results = await original_map_node_over_list(
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\execution.py", line 316, in _async_map_node_over_list
    await process_inputs(input_dict, i)

  File "D:\Comfyui\ComfyUI\execution.py", line 304, in process_inputs
    result = f(**inputs)
             ^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-krea2edit\__init__.py", line 416, in encode
    return (clip.encode_from_tokens_scheduled(tokens),)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\comfy\sd.py", line 331, in encode_from_tokens_scheduled
    pooled_dict = self.encode_from_tokens(tokens, return_pooled=return_pooled, return_dict=True)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\comfy\sd.py", line 400, in encode_from_tokens
    o = self.cond_stage_model.encode_token_weights(tokens)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\comfy\text_encoders\krea2.py", line 44, in encode_token_weights
    out, pooled, extra = super().encode_token_weights(token_weight_pairs)  # out: (B, 12, seq, 2560)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 743, in encode_token_weights
    out = getattr(self, self.clip).encode_token_weights(token_weight_pairs)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 45, in encode_token_weights
    o = self.encode(to_encode)
        ^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 306, in encode
    return self(tokens)
           ^^^^^^^^^^^^

  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 266, in forward
    embeds, attention_mask, num_tokens, embeds_info = self.process_tokens(tokens, device)
                                                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 228, in process_tokens
    emb, extra = self.transformer.preprocess_embed(emb, device=device)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen3vl.py", line 65, in preprocess_embed
    merged, deepstack = self.visual(image.to(device, dtype=torch.float32), grid)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen35.py", line 666, in forward
    x = blk(x, cu_seqlens=cu_seqlens, position_embeddings=position_embeddings, optimized_attention=optimized_attention)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen35.py", line 522, in forward
    x = x + self.attn(self.norm1(x), cu_seqlens=cu_seqlens, position_embeddings=position_embeddings, optimized_attention=optimized_attention)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen35.py", line 505, in forward
    attn_outputs.append(optimized_attention(q, k, v, self.num_heads, skip_reshape=True))
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\comfy\ldm\modules\attention.py", line 178, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\comfy\ldm\modules\attention.py", line 544, in attention_pytorch
    out = comfy.ops.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False, **sdpa_extra)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-sage-guard\__init__.py", line 135, in __call__
    return self._current_sdpa(
           ^^^^^^^^^^^^^^^^^^^

System Information

• ComfyUI Version: 0.28.0
• Arguments: ComfyUI\main.py --windows-standalone-build --listen --use-sage-attention --preview-method auto
• OS: win32
• Python Version: 3.12.10 (tags/v3.12.10:0cc8128, Apr 8 2025, 12:21:36) [MSC v.1943 64 bit (AMD64)]
• Embedded Python: true
• PyTorch Version: 2.10.0+cu130

Devices

• Name: cuda:0 NVIDIA GeForce RTX 5090 : cudaMallocAsync

  • Type: cuda
  • VRAM Total: 34190458880
  • VRAM Free: 32462733312
  • Torch VRAM Total: 33554432
  • Torch VRAM Free: 25034752

• Name: cuda:1 NVIDIA GeForce RTX 5090 : cudaMallocAsync

  • Type: cuda
  • VRAM Total: 34190458880
  • VRAM Free: 32496418816
  • Torch VRAM Total: 0
  • Torch VRAM Free: 0

• Name: cuda:2 NVIDIA GeForce RTX 5090 : cudaMallocAsync

  • Type: cuda
  • VRAM Total: 34190458880
  • VRAM Free: 32496418816
  • Torch VRAM Total: 0
  • Torch VRAM Free: 0

Logs

2026-07-20T22:45:55.400824 - 
2026-07-20T22:45:55.418843 - �[1m�[31m[ERROR]�[0m !!! Exception during processing !!! Krea2 expects conditioning with 12x2560=30720 features (a 12-layer Qwen3-VL stack) but got 2560. Load the text encoder with CLIPLoader type 'krea2'.
2026-07-20T22:45:55.473904 - �[1m�[31m[ERROR]�[0m Traceback (most recent call last):
  File "D:\Comfyui\ComfyUI\execution.py", line 543, in execute
    output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
                                                              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\execution.py", line 342, in get_output_data
    return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-lora-manager\py\metadata_collector\metadata_hook.py", line 171, in async_map_node_over_list_with_metadata
    results = await original_map_node_over_list(
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\execution.py", line 316, in _async_map_node_over_list
    await process_inputs(input_dict, i)
  File "D:\Comfyui\ComfyUI\execution.py", line 304, in process_inputs
    result = f(**inputs)
             ^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\nodes.py", line 1607, in sample
    return common_ksampler(model, seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, denoise=denoise)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\nodes.py", line 1571, in common_ksampler
    samples = comfy.sample.sample(model, noise, steps, cfg, sampler_name, scheduler, positive, negative, latent_image,
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sample.py", line 74, in sample
    samples = sampler.sample(noise, positive, negative, cfg=cfg, latent_image=latent_image, start_step=start_step, last_step=last_step, force_full_denoise=force_full_denoise, denoise_mask=noise_mask, sigmas=sigmas, callback=callback, disable_pbar=disable_pbar, seed=seed)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\samplers.py", line 1444, in sample
    return sample(self.model, noise, positive, negative, cfg, self.device, sampler, sigmas, self.model_options, latent_image=latent_image, denoise_mask=denoise_mask, callback=callback, disable_pbar=disable_pbar, seed=seed)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\samplers.py", line 1334, in sample
    return cfg_guider.sample(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\samplers.py", line 1316, in sample
    output = executor.execute(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\patcher_extension.py", line 113, in execute
    return self.original(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\samplers.py", line 1254, in outer_sample
    output = self.inner_sample(noise, latent_image, device, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\samplers.py", line 1229, in inner_sample
    samples = executor.execute(self, sigmas, extra_args, callback, noise, latent_image, denoise_mask, disable_pbar)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\patcher_extension.py", line 113, in execute
    return self.original(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\samplers.py", line 999, in sample
    samples = self.sampler_function(model_k, noise, sigmas, extra_args=extra_args, callback=k_callback, disable=disable_pbar, **self.extra_options)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\utils\_contextlib.py", line 124, in decorate_context
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\k_diffusion\sampling.py", line 205, in sample_euler
    denoised = model(x, sigma_hat * s_in, **extra_args)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\samplers.py", line 639, in __call__
    out = self.inner_model(x, sigma, model_options=model_options, seed=seed)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\samplers.py", line 1202, in __call__
    return self.outer_predict_noise(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\samplers.py", line 1209, in outer_predict_noise
    ).execute(x, timestep, model_options, seed)
      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\patcher_extension.py", line 113, in execute
    return self.original(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\samplers.py", line 1212, in predict_noise
    return sampling_function(self.inner_model, x, timestep, self.conds.get("negative", None), self.conds.get("positive", None), self.cfg, model_options=model_options, seed=seed)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\samplers.py", line 619, in sampling_function
    out = calc_cond_batch(model, conds, x, timestep, model_options)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\samplers.py", line 210, in calc_cond_batch
    return _calc_cond_batch_outer(model, conds, x_in, timestep, model_options)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\samplers.py", line 218, in _calc_cond_batch_outer
    return executor.execute(model, conds, x_in, timestep, model_options)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\patcher_extension.py", line 113, in execute
    return self.original(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\samplers.py", line 334, in _calc_cond_batch
    output = model.apply_model(input_x, timestep_, **c).chunk(batch_chunks)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\model_base.py", line 193, in apply_model
    return comfy.patcher_extension.WrapperExecutor.new_class_executor(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\patcher_extension.py", line 113, in execute
    return self.original(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\model_base.py", line 237, in _apply_model
    model_output = self.diffusion_model(xc, t, context=context, control=control, transformer_options=transformer_options, **extra_conds)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\ldm\krea2\model.py", line 277, in forward
    return comfy.patcher_extension.WrapperExecutor.new_class_executor(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\patcher_extension.py", line 114, in execute
    return self.wrappers[self.idx](self, *args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-krea2edit\__init__.py", line 320, in wrapper
    v = krea2_edit_forward(dm, x, timesteps, context, src, transformer_options,
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-krea2edit\__init__.py", line 198, in krea2_edit_forward
    context = m._unpack_context(context)                       # (B, seq, 12, 2560)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\ldm\krea2\model.py", line 386, in _unpack_context
    raise ValueError(
ValueError: Krea2 expects conditioning with 12x2560=30720 features (a 12-layer Qwen3-VL stack) but got 2560. Load the text encoder with CLIPLoader type 'krea2'.

2026-07-20T22:45:55.476407 - �[32m[INFO]�[0m �[32mPrompt executed in 1.91 seconds�[0m
2026-07-20T22:45:57.321565 - FETCH ComfyRegistry Data: 85/1632026-07-20T22:45:57.321565 - 
2026-07-20T22:46:03.364142 - FETCH ComfyRegistry Data: 90/1632026-07-20T22:46:03.364142 - 
2026-07-20T22:46:06.273202 - [UmiTextBypass DEBUG] /umi/bypass_preview called2026-07-20T22:46:06.273202 - 
2026-07-20T22:46:06.273202 - [UmiTextBypass DEBUG] prompt_payload type: <class 'dict'>2026-07-20T22:46:06.273202 - 
2026-07-20T22:46:06.273202 - [UmiTextBypass DEBUG] prompt_payload keys: ['workflow', 'output']2026-07-20T22:46:06.273202 - 
2026-07-20T22:46:06.273202 - 
[UmiTextBypass DEBUG] ===== ANALYZING PROMPT GRAPH =====2026-07-20T22:46:06.273202 - 
2026-07-20T22:46:06.273202 - [UmiTextBypass DEBUG] Total nodes in graph: 22026-07-20T22:46:06.273202 - 
2026-07-20T22:46:06.274201 - [UmiTextBypass DEBUG] Node types:2026-07-20T22:46:06.274201 - 
2026-07-20T22:46:06.274201 -   Node workflow: UNKNOWN2026-07-20T22:46:06.274201 - 
2026-07-20T22:46:06.274201 -   Node output: UNKNOWN2026-07-20T22:46:06.274201 - 
2026-07-20T22:46:06.278817 - �[32m[INFO]�[0m got prompt
2026-07-20T22:46:06.279436 - 
[UmiTextBypass DEBUG] ===== ANALYZING PROMPT GRAPH =====2026-07-20T22:46:06.279436 - 
2026-07-20T22:46:06.279436 - [UmiTextBypass DEBUG] Total nodes in graph: 142026-07-20T22:46:06.279436 - 
2026-07-20T22:46:06.279436 - [UmiTextBypass DEBUG] Node types:2026-07-20T22:46:06.279436 - 
2026-07-20T22:46:06.279436 -   Node 29: SaveImage2026-07-20T22:46:06.279974 - 
2026-07-20T22:46:06.279974 -   Node 53: KSampler2026-07-20T22:46:06.279974 - 
2026-07-20T22:46:06.279974 -   Node 54: VAEDecode2026-07-20T22:46:06.279974 - 
2026-07-20T22:46:06.279974 -   Node 55: UNETLoader2026-07-20T22:46:06.279974 - 
2026-07-20T22:46:06.279974 -   Node 56: CLIPLoader2026-07-20T22:46:06.279974 - 
2026-07-20T22:46:06.279974 -   Node 57: VAELoader2026-07-20T22:46:06.279974 - 
2026-07-20T22:46:06.279974 -   Node 71: LoraLoaderModelOnly2026-07-20T22:46:06.279974 - 
2026-07-20T22:46:06.280500 -   Node 72: LoadImage2026-07-20T22:46:06.280500 - 
2026-07-20T22:46:06.280500 -   Node 73: VAEEncode2026-07-20T22:46:06.280500 - 
2026-07-20T22:46:06.280500 -   Node 79: Krea2EditModelPatch2026-07-20T22:46:06.280500 - 
2026-07-20T22:46:06.280500 -   Node 82: EmptySD3LatentImage2026-07-20T22:46:06.280500 - 
2026-07-20T22:46:06.280500 -   Node 83: ResolutionSelector2026-07-20T22:46:06.280500 - 
2026-07-20T22:46:06.280500 -   Node 84: Krea2EditGroundedEncode2026-07-20T22:46:06.280500 - 
2026-07-20T22:46:06.280500 -   Node 85: Krea2EditGroundedEncode2026-07-20T22:46:06.280500 - 
2026-07-20T22:46:06.670719 - �[32m[INFO]�[0m CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
2026-07-20T22:46:06.674721 - �[32m[INFO]�[0m Requested to load Krea2TEModel_
2026-07-20T22:46:06.728777 - �[32m[INFO]�[0m Model Krea2TEModel_ prepared for dynamic VRAM loading. 8463MB Staged. 0 patches attached. Force pre-loaded 249 weights: 627 KB.
2026-07-20T22:46:06.913671 - �[1m�[31m[ERROR]�[0m !!! Exception during processing !!! scaled_dot_product_attention() takes from 3 to 6 positional arguments but 7 were given
2026-07-20T22:46:06.914672 - �[1m�[31m[ERROR]�[0m Traceback (most recent call last):
  File "D:\Comfyui\ComfyUI\execution.py", line 543, in execute
    output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
                                                              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\execution.py", line 342, in get_output_data
    return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-lora-manager\py\metadata_collector\metadata_hook.py", line 171, in async_map_node_over_list_with_metadata
    results = await original_map_node_over_list(
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\execution.py", line 316, in _async_map_node_over_list
    await process_inputs(input_dict, i)
  File "D:\Comfyui\ComfyUI\execution.py", line 304, in process_inputs
    result = f(**inputs)
             ^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-krea2edit\__init__.py", line 416, in encode
    return (clip.encode_from_tokens_scheduled(tokens),)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd.py", line 331, in encode_from_tokens_scheduled
    pooled_dict = self.encode_from_tokens(tokens, return_pooled=return_pooled, return_dict=True)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd.py", line 400, in encode_from_tokens
    o = self.cond_stage_model.encode_token_weights(tokens)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\krea2.py", line 44, in encode_token_weights
    out, pooled, extra = super().encode_token_weights(token_weight_pairs)  # out: (B, 12, seq, 2560)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 743, in encode_token_weights
    out = getattr(self, self.clip).encode_token_weights(token_weight_pairs)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 45, in encode_token_weights
    o = self.encode(to_encode)
        ^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 306, in encode
    return self(tokens)
           ^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 266, in forward
    embeds, attention_mask, num_tokens, embeds_info = self.process_tokens(tokens, device)
                                                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 228, in process_tokens
    emb, extra = self.transformer.preprocess_embed(emb, device=device)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen3vl.py", line 65, in preprocess_embed
    merged, deepstack = self.visual(image.to(device, dtype=torch.float32), grid)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen35.py", line 666, in forward
    x = blk(x, cu_seqlens=cu_seqlens, position_embeddings=position_embeddings, optimized_attention=optimized_attention)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen35.py", line 522, in forward
    x = x + self.attn(self.norm1(x), cu_seqlens=cu_seqlens, position_embeddings=position_embeddings, optimized_attention=optimized_attention)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen35.py", line 505, in forward
    attn_outputs.append(optimized_attention(q, k, v, self.num_heads, skip_reshape=True))
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\ldm\modules\attention.py", line 178, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\ldm\modules\attention.py", line 544, in attention_pytorch
    out = comfy.ops.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False, **sdpa_extra)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-sage-guard\__init__.py", line 135, in __call__
    return self._current_sdpa(
           ^^^^^^^^^^^^^^^^^^^
TypeError: scaled_dot_product_attention() takes from 3 to 6 positional arguments but 7 were given

2026-07-20T22:46:06.916181 - �[32m[INFO]�[0m �[32mPrompt executed in 0.55 seconds�[0m
2026-07-20T22:46:10.199463 - FETCH ComfyRegistry Data: 95/1632026-07-20T22:46:10.199463 - 
2026-07-20T22:46:12.208769 - [UmiTextBypass DEBUG] /umi/bypass_preview called2026-07-20T22:46:12.208769 - 
2026-07-20T22:46:12.208769 - [UmiTextBypass DEBUG] prompt_payload type: <class 'dict'>2026-07-20T22:46:12.208769 - 
2026-07-20T22:46:12.208769 - [UmiTextBypass DEBUG] prompt_payload keys: ['workflow', 'output']2026-07-20T22:46:12.208769 - 
2026-07-20T22:46:12.208769 - 
[UmiTextBypass DEBUG] ===== ANALYZING PROMPT GRAPH =====2026-07-20T22:46:12.208769 - 
2026-07-20T22:46:12.208769 - [UmiTextBypass DEBUG] Total nodes in graph: 22026-07-20T22:46:12.208769 - 
2026-07-20T22:46:12.208769 - [UmiTextBypass DEBUG] Node types:2026-07-20T22:46:12.208769 - 
2026-07-20T22:46:12.209768 -   Node workflow: UNKNOWN2026-07-20T22:46:12.209768 - 
2026-07-20T22:46:12.209768 -   Node output: UNKNOWN2026-07-20T22:46:12.209768 - 
2026-07-20T22:46:12.224285 - �[32m[INFO]�[0m got prompt
2026-07-20T22:46:12.224790 - 
[UmiTextBypass DEBUG] ===== ANALYZING PROMPT GRAPH =====2026-07-20T22:46:12.224790 - 
2026-07-20T22:46:12.224790 - [UmiTextBypass DEBUG] Total nodes in graph: 132026-07-20T22:46:12.224790 - 
2026-07-20T22:46:12.224790 - [UmiTextBypass DEBUG] Node types:2026-07-20T22:46:12.224790 - 
2026-07-20T22:46:12.225341 -   Node 29: SaveImage2026-07-20T22:46:12.225341 - 
2026-07-20T22:46:12.225341 -   Node 53: KSampler2026-07-20T22:46:12.225341 - 
2026-07-20T22:46:12.225341 -   Node 54: VAEDecode2026-07-20T22:46:12.225341 - 
2026-07-20T22:46:12.225341 -   Node 55: UNETLoader2026-07-20T22:46:12.225341 - 
2026-07-20T22:46:12.225341 -   Node 56: CLIPLoader2026-07-20T22:46:12.225341 - 
2026-07-20T22:46:12.225341 -   Node 57: VAELoader2026-07-20T22:46:12.225341 - 
2026-07-20T22:46:12.225863 -   Node 71: LoraLoaderModelOnly2026-07-20T22:46:12.225863 - 
2026-07-20T22:46:12.225863 -   Node 72: LoadImage2026-07-20T22:46:12.225863 - 
2026-07-20T22:46:12.225863 -   Node 73: VAEEncode2026-07-20T22:46:12.225863 - 
2026-07-20T22:46:12.225863 -   Node 79: Krea2EditModelPatch2026-07-20T22:46:12.225863 - 
2026-07-20T22:46:12.225863 -   Node 82: EmptySD3LatentImage2026-07-20T22:46:12.225863 - 
2026-07-20T22:46:12.226420 -   Node 83: ResolutionSelector2026-07-20T22:46:12.226420 - 
2026-07-20T22:46:12.226420 -   Node 84: Krea2EditGroundedEncode2026-07-20T22:46:12.226420 - 
2026-07-20T22:46:12.308725 - �[1m�[31m[ERROR]�[0m Failed to validate prompt for output 29:
2026-07-20T22:46:12.308725 - �[1m�[31m[ERROR]�[0m * KSampler 53:
2026-07-20T22:46:12.308725 - �[1m�[31m[ERROR]�[0m   - Required input is missing: negative
2026-07-20T22:46:12.308725 - �[1m�[31m[ERROR]�[0m Output will be ignored
2026-07-20T22:46:12.308725 - �[1m�[33m[WARNING]�[0m invalid prompt: {'type': 'prompt_outputs_failed_validation', 'message': 'Prompt outputs failed validation', 'details': '', 'extra_info': {}}
2026-07-20T22:46:17.107363 - FETCH ComfyRegistry Data: 100/1632026-07-20T22:46:17.107363 - 
2026-07-20T22:46:19.073081 - [UmiTextBypass DEBUG] /umi/bypass_preview called2026-07-20T22:46:19.073081 - 
2026-07-20T22:46:19.073081 - [UmiTextBypass DEBUG] prompt_payload type: <class 'dict'>2026-07-20T22:46:19.073081 - 
2026-07-20T22:46:19.074081 - [UmiTextBypass DEBUG] prompt_payload keys: ['workflow', 'output']2026-07-20T22:46:19.074081 - 
2026-07-20T22:46:19.074081 - 
[UmiTextBypass DEBUG] ===== ANALYZING PROMPT GRAPH =====2026-07-20T22:46:19.074081 - 
2026-07-20T22:46:19.074081 - [UmiTextBypass DEBUG] Total nodes in graph: 22026-07-20T22:46:19.074081 - 
2026-07-20T22:46:19.074081 - [UmiTextBypass DEBUG] Node types:2026-07-20T22:46:19.074081 - 
2026-07-20T22:46:19.074081 -   Node workflow: UNKNOWN2026-07-20T22:46:19.074081 - 
2026-07-20T22:46:19.074081 -   Node output: UNKNOWN2026-07-20T22:46:19.074081 - 
2026-07-20T22:46:19.091613 - �[32m[INFO]�[0m got prompt
2026-07-20T22:46:19.092612 - 
[UmiTextBypass DEBUG] ===== ANALYZING PROMPT GRAPH =====2026-07-20T22:46:19.092612 - 
2026-07-20T22:46:19.092612 - [UmiTextBypass DEBUG] Total nodes in graph: 142026-07-20T22:46:19.092612 - 
2026-07-20T22:46:19.092612 - [UmiTextBypass DEBUG] Node types:2026-07-20T22:46:19.093612 - 
2026-07-20T22:46:19.093612 -   Node 29: SaveImage2026-07-20T22:46:19.093612 - 
2026-07-20T22:46:19.093612 -   Node 53: KSampler2026-07-20T22:46:19.093612 - 
2026-07-20T22:46:19.093612 -   Node 54: VAEDecode2026-07-20T22:46:19.093612 - 
2026-07-20T22:46:19.093612 -   Node 55: UNETLoader2026-07-20T22:46:19.093612 - 
2026-07-20T22:46:19.093612 -   Node 56: CLIPLoader2026-07-20T22:46:19.093612 - 
2026-07-20T22:46:19.093612 -   Node 57: VAELoader2026-07-20T22:46:19.093612 - 
2026-07-20T22:46:19.093612 -   Node 71: LoraLoaderModelOnly2026-07-20T22:46:19.094611 - 
2026-07-20T22:46:19.094611 -   Node 72: LoadImage2026-07-20T22:46:19.094611 - 
2026-07-20T22:46:19.094611 -   Node 73: VAEEncode2026-07-20T22:46:19.094611 - 
2026-07-20T22:46:19.094611 -   Node 79: Krea2EditModelPatch2026-07-20T22:46:19.094611 - 
2026-07-20T22:46:19.094611 -   Node 82: EmptySD3LatentImage2026-07-20T22:46:19.094611 - 
2026-07-20T22:46:19.095612 -   Node 83: ResolutionSelector2026-07-20T22:46:19.095612 - 
2026-07-20T22:46:19.095612 -   Node 84: Krea2EditGroundedEncode2026-07-20T22:46:19.095612 - 
2026-07-20T22:46:19.095612 -   Node 85: Krea2EditGroundedEncode2026-07-20T22:46:19.095612 - 
2026-07-20T22:46:19.300423 - �[32m[INFO]�[0m Model Krea2TEModel_ prepared for dynamic VRAM loading. 8463MB Staged. 0 patches attached. Force pre-loaded 249 weights: 627 KB.
2026-07-20T22:46:19.333026 - �[1m�[31m[ERROR]�[0m !!! Exception during processing !!! scaled_dot_product_attention() takes from 3 to 6 positional arguments but 7 were given
2026-07-20T22:46:19.334026 - �[1m�[31m[ERROR]�[0m Traceback (most recent call last):
  File "D:\Comfyui\ComfyUI\execution.py", line 543, in execute
    output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
                                                              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\execution.py", line 342, in get_output_data
    return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-lora-manager\py\metadata_collector\metadata_hook.py", line 171, in async_map_node_over_list_with_metadata
    results = await original_map_node_over_list(
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\execution.py", line 316, in _async_map_node_over_list
    await process_inputs(input_dict, i)
  File "D:\Comfyui\ComfyUI\execution.py", line 304, in process_inputs
    result = f(**inputs)
             ^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-krea2edit\__init__.py", line 416, in encode
    return (clip.encode_from_tokens_scheduled(tokens),)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd.py", line 331, in encode_from_tokens_scheduled
    pooled_dict = self.encode_from_tokens(tokens, return_pooled=return_pooled, return_dict=True)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd.py", line 400, in encode_from_tokens
    o = self.cond_stage_model.encode_token_weights(tokens)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\krea2.py", line 44, in encode_token_weights
    out, pooled, extra = super().encode_token_weights(token_weight_pairs)  # out: (B, 12, seq, 2560)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 743, in encode_token_weights
    out = getattr(self, self.clip).encode_token_weights(token_weight_pairs)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 45, in encode_token_weights
    o = self.encode(to_encode)
        ^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 306, in encode
    return self(tokens)
           ^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 266, in forward
    embeds, attention_mask, num_tokens, embeds_info = self.process_tokens(tokens, device)
                                                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 228, in process_tokens
    emb, extra = self.transformer.preprocess_embed(emb, device=device)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen3vl.py", line 65, in preprocess_embed
    merged, deepstack = self.visual(image.to(device, dtype=torch.float32), grid)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen35.py", line 666, in forward
    x = blk(x, cu_seqlens=cu_seqlens, position_embeddings=position_embeddings, optimized_attention=optimized_attention)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen35.py", line 522, in forward
    x = x + self.attn(self.norm1(x), cu_seqlens=cu_seqlens, position_embeddings=position_embeddings, optimized_attention=optimized_attention)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen35.py", line 505, in forward
    attn_outputs.append(optimized_attention(q, k, v, self.num_heads, skip_reshape=True))
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\ldm\modules\attention.py", line 178, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\ldm\modules\attention.py", line 544, in attention_pytorch
    out = comfy.ops.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False, **sdpa_extra)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-sage-guard\__init__.py", line 135, in __call__
    return self._current_sdpa(
           ^^^^^^^^^^^^^^^^^^^
TypeError: scaled_dot_product_attention() takes from 3 to 6 positional arguments but 7 were given

2026-07-20T22:46:19.335529 - �[32m[INFO]�[0m �[32mPrompt executed in 0.15 seconds�[0m
2026-07-20T22:46:23.489455 - FETCH ComfyRegistry Data: 105/1632026-07-20T22:46:23.489455 - 
2026-07-20T22:46:28.496260 - [UmiTextBypass DEBUG] /umi/bypass_preview called2026-07-20T22:46:28.496260 - 
2026-07-20T22:46:28.496260 - [UmiTextBypass DEBUG] prompt_payload type: <class 'dict'>2026-07-20T22:46:28.496260 - 
2026-07-20T22:46:28.496260 - [UmiTextBypass DEBUG] prompt_payload keys: ['workflow', 'output']2026-07-20T22:46:28.496260 - 
2026-07-20T22:46:28.496260 - 
[UmiTextBypass DEBUG] ===== ANALYZING PROMPT GRAPH =====2026-07-20T22:46:28.496260 - 
2026-07-20T22:46:28.497265 - [UmiTextBypass DEBUG] Total nodes in graph: 22026-07-20T22:46:28.497265 - 
2026-07-20T22:46:28.497265 - [UmiTextBypass DEBUG] Node types:2026-07-20T22:46:28.497265 - 
2026-07-20T22:46:28.497265 -   Node workflow: UNKNOWN2026-07-20T22:46:28.497265 - 
2026-07-20T22:46:28.497265 -   Node output: UNKNOWN2026-07-20T22:46:28.497265 - 
2026-07-20T22:46:28.513775 - �[32m[INFO]�[0m got prompt
2026-07-20T22:46:28.513775 - 
[UmiTextBypass DEBUG] ===== ANALYZING PROMPT GRAPH =====2026-07-20T22:46:28.513775 - 
2026-07-20T22:46:28.513775 - [UmiTextBypass DEBUG] Total nodes in graph: 152026-07-20T22:46:28.513775 - 
2026-07-20T22:46:28.513775 - [UmiTextBypass DEBUG] Node types:2026-07-20T22:46:28.513775 - 
2026-07-20T22:46:28.513775 -   Node 29: SaveImage2026-07-20T22:46:28.513775 - 
2026-07-20T22:46:28.513775 -   Node 53: KSampler2026-07-20T22:46:28.513775 - 
2026-07-20T22:46:28.513775 -   Node 54: VAEDecode2026-07-20T22:46:28.513775 - 
2026-07-20T22:46:28.515280 -   Node 55: UNETLoader2026-07-20T22:46:28.515280 - 
2026-07-20T22:46:28.515280 -   Node 56: CLIPLoader2026-07-20T22:46:28.515280 - 
2026-07-20T22:46:28.515280 -   Node 57: VAELoader2026-07-20T22:46:28.515280 - 
2026-07-20T22:46:28.515280 -   Node 71: LoraLoaderModelOnly2026-07-20T22:46:28.515280 - 
2026-07-20T22:46:28.515280 -   Node 72: LoadImage2026-07-20T22:46:28.515280 - 
2026-07-20T22:46:28.515280 -   Node 73: VAEEncode2026-07-20T22:46:28.515280 - 
2026-07-20T22:46:28.515280 -   Node 79: Krea2EditModelPatch2026-07-20T22:46:28.515280 - 
2026-07-20T22:46:28.515280 -   Node 82: EmptySD3LatentImage2026-07-20T22:46:28.515280 - 
2026-07-20T22:46:28.515280 -   Node 83: ResolutionSelector2026-07-20T22:46:28.515280 - 
2026-07-20T22:46:28.515280 -   Node 84: Krea2EditGroundedEncode2026-07-20T22:46:28.515280 - 
2026-07-20T22:46:28.515280 -   Node 85: Krea2EditGroundedEncode2026-07-20T22:46:28.515280 - 
2026-07-20T22:46:28.516285 -   Node 90: LoadImage2026-07-20T22:46:28.516285 - 
2026-07-20T22:46:28.879200 - �[32m[INFO]�[0m Model Krea2TEModel_ prepared for dynamic VRAM loading. 8463MB Staged. 0 patches attached. Force pre-loaded 249 weights: 627 KB.
2026-07-20T22:46:28.912269 - �[1m�[31m[ERROR]�[0m !!! Exception during processing !!! scaled_dot_product_attention() takes from 3 to 6 positional arguments but 7 were given
2026-07-20T22:46:28.913269 - �[1m�[31m[ERROR]�[0m Traceback (most recent call last):
  File "D:\Comfyui\ComfyUI\execution.py", line 543, in execute
    output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
                                                              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\execution.py", line 342, in get_output_data
    return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-lora-manager\py\metadata_collector\metadata_hook.py", line 171, in async_map_node_over_list_with_metadata
    results = await original_map_node_over_list(
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\execution.py", line 316, in _async_map_node_over_list
    await process_inputs(input_dict, i)
  File "D:\Comfyui\ComfyUI\execution.py", line 304, in process_inputs
    result = f(**inputs)
             ^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-krea2edit\__init__.py", line 416, in encode
    return (clip.encode_from_tokens_scheduled(tokens),)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd.py", line 331, in encode_from_tokens_scheduled
    pooled_dict = self.encode_from_tokens(tokens, return_pooled=return_pooled, return_dict=True)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd.py", line 400, in encode_from_tokens
    o = self.cond_stage_model.encode_token_weights(tokens)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\krea2.py", line 44, in encode_token_weights
    out, pooled, extra = super().encode_token_weights(token_weight_pairs)  # out: (B, 12, seq, 2560)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 743, in encode_token_weights
    out = getattr(self, self.clip).encode_token_weights(token_weight_pairs)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 45, in encode_token_weights
    o = self.encode(to_encode)
        ^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 306, in encode
    return self(tokens)
           ^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 266, in forward
    embeds, attention_mask, num_tokens, embeds_info = self.process_tokens(tokens, device)
                                                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 228, in process_tokens
    emb, extra = self.transformer.preprocess_embed(emb, device=device)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen3vl.py", line 65, in preprocess_embed
    merged, deepstack = self.visual(image.to(device, dtype=torch.float32), grid)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen35.py", line 666, in forward
    x = blk(x, cu_seqlens=cu_seqlens, position_embeddings=position_embeddings, optimized_attention=optimized_attention)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen35.py", line 522, in forward
    x = x + self.attn(self.norm1(x), cu_seqlens=cu_seqlens, position_embeddings=position_embeddings, optimized_attention=optimized_attention)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen35.py", line 505, in forward
    attn_outputs.append(optimized_attention(q, k, v, self.num_heads, skip_reshape=True))
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\ldm\modules\attention.py", line 178, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\ldm\modules\attention.py", line 544, in attention_pytorch
    out = comfy.ops.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False, **sdpa_extra)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-sage-guard\__init__.py", line 135, in __call__
    return self._current_sdpa(
           ^^^^^^^^^^^^^^^^^^^
TypeError: scaled_dot_product_attention() takes from 3 to 6 positional arguments but 7 were given

2026-07-20T22:46:28.914269 - �[32m[INFO]�[0m �[32mPrompt executed in 0.23 seconds�[0m
2026-07-20T22:46:30.040383 - FETCH ComfyRegistry Data: 110/1632026-07-20T22:46:30.041383 - 
2026-07-20T22:46:36.592648 - FETCH ComfyRegistry Data: 115/1632026-07-20T22:46:36.592648 - 
2026-07-20T22:46:43.494807 - FETCH ComfyRegistry Data: 120/1632026-07-20T22:46:43.494807 - 
2026-07-20T22:46:45.984953 - [UmiTextBypass DEBUG] /umi/bypass_preview called2026-07-20T22:46:45.984953 - 
2026-07-20T22:46:45.984953 - [UmiTextBypass DEBUG] prompt_payload type: <class 'dict'>2026-07-20T22:46:45.984953 - 
2026-07-20T22:46:45.984953 - [UmiTextBypass DEBUG] prompt_payload keys: ['workflow', 'output']2026-07-20T22:46:45.984953 - 
2026-07-20T22:46:45.984953 - 
[UmiTextBypass DEBUG] ===== ANALYZING PROMPT GRAPH =====2026-07-20T22:46:45.984953 - 
2026-07-20T22:46:45.984953 - [UmiTextBypass DEBUG] Total nodes in graph: 22026-07-20T22:46:45.984953 - 
2026-07-20T22:46:45.984953 - [UmiTextBypass DEBUG] Node types:2026-07-20T22:46:45.984953 - 
2026-07-20T22:46:45.984953 -   Node workflow: UNKNOWN2026-07-20T22:46:45.984953 - 
2026-07-20T22:46:45.984953 -   Node output: UNKNOWN2026-07-20T22:46:45.984953 - 
2026-07-20T22:46:45.989241 - �[32m[INFO]�[0m got prompt
2026-07-20T22:46:45.989241 - 
[UmiTextBypass DEBUG] ===== ANALYZING PROMPT GRAPH =====2026-07-20T22:46:45.989241 - 
2026-07-20T22:46:45.989241 - [UmiTextBypass DEBUG] Total nodes in graph: 142026-07-20T22:46:45.990243 - 
2026-07-20T22:46:45.990243 - [UmiTextBypass DEBUG] Node types:2026-07-20T22:46:45.990243 - 
2026-07-20T22:46:45.990243 -   Node 29: SaveImage2026-07-20T22:46:45.990243 - 
2026-07-20T22:46:45.990243 -   Node 53: KSampler2026-07-20T22:46:45.990243 - 
2026-07-20T22:46:45.990243 -   Node 54: VAEDecode2026-07-20T22:46:45.990243 - 
2026-07-20T22:46:45.990243 -   Node 55: UNETLoader2026-07-20T22:46:45.990243 - 
2026-07-20T22:46:45.990243 -   Node 56: CLIPLoader2026-07-20T22:46:45.990243 - 
2026-07-20T22:46:45.990243 -   Node 57: VAELoader2026-07-20T22:46:45.990243 - 
2026-07-20T22:46:45.990243 -   Node 71: LoraLoaderModelOnly2026-07-20T22:46:45.990243 - 
2026-07-20T22:46:45.990243 -   Node 72: LoadImage2026-07-20T22:46:45.990243 - 
2026-07-20T22:46:45.990243 -   Node 73: VAEEncode2026-07-20T22:46:45.990243 - 
2026-07-20T22:46:45.990243 -   Node 79: Krea2EditModelPatch2026-07-20T22:46:45.990243 - 
2026-07-20T22:46:45.990243 -   Node 82: EmptySD3LatentImage2026-07-20T22:46:45.990243 - 
2026-07-20T22:46:45.991243 -   Node 83: ResolutionSelector2026-07-20T22:46:45.991243 - 
2026-07-20T22:46:45.991243 -   Node 84: Krea2EditGroundedEncode2026-07-20T22:46:45.991243 - 
2026-07-20T22:46:45.991243 -   Node 85: Krea2EditGroundedEncode2026-07-20T22:46:45.991243 - 
2026-07-20T22:46:46.172534 - �[32m[INFO]�[0m Model Krea2TEModel_ prepared for dynamic VRAM loading. 8463MB Staged. 0 patches attached. Force pre-loaded 249 weights: 627 KB.
2026-07-20T22:46:46.202568 - �[1m�[31m[ERROR]�[0m !!! Exception during processing !!! scaled_dot_product_attention() takes from 3 to 6 positional arguments but 7 were given
2026-07-20T22:46:46.203568 - �[1m�[31m[ERROR]�[0m Traceback (most recent call last):
  File "D:\Comfyui\ComfyUI\execution.py", line 543, in execute
    output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
                                                              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\execution.py", line 342, in get_output_data
    return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-lora-manager\py\metadata_collector\metadata_hook.py", line 171, in async_map_node_over_list_with_metadata
    results = await original_map_node_over_list(
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\execution.py", line 316, in _async_map_node_over_list
    await process_inputs(input_dict, i)
  File "D:\Comfyui\ComfyUI\execution.py", line 304, in process_inputs
    result = f(**inputs)
             ^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-krea2edit\__init__.py", line 416, in encode
    return (clip.encode_from_tokens_scheduled(tokens),)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd.py", line 331, in encode_from_tokens_scheduled
    pooled_dict = self.encode_from_tokens(tokens, return_pooled=return_pooled, return_dict=True)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd.py", line 400, in encode_from_tokens
    o = self.cond_stage_model.encode_token_weights(tokens)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\krea2.py", line 44, in encode_token_weights
    out, pooled, extra = super().encode_token_weights(token_weight_pairs)  # out: (B, 12, seq, 2560)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 743, in encode_token_weights
    out = getattr(self, self.clip).encode_token_weights(token_weight_pairs)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 45, in encode_token_weights
    o = self.encode(to_encode)
        ^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 306, in encode
    return self(tokens)
           ^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 266, in forward
    embeds, attention_mask, num_tokens, embeds_info = self.process_tokens(tokens, device)
                                                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\sd1_clip.py", line 228, in process_tokens
    emb, extra = self.transformer.preprocess_embed(emb, device=device)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen3vl.py", line 65, in preprocess_embed
    merged, deepstack = self.visual(image.to(device, dtype=torch.float32), grid)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen35.py", line 666, in forward
    x = blk(x, cu_seqlens=cu_seqlens, position_embeddings=position_embeddings, optimized_attention=optimized_attention)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen35.py", line 522, in forward
    x = x + self.attn(self.norm1(x), cu_seqlens=cu_seqlens, position_embeddings=position_embeddings, optimized_attention=optimized_attention)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\text_encoders\qwen35.py", line 505, in forward
    attn_outputs.append(optimized_attention(q, k, v, self.num_heads, skip_reshape=True))
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\ldm\modules\attention.py", line 178, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\comfy\ldm\modules\attention.py", line 544, in attention_pytorch
    out = comfy.ops.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False, **sdpa_extra)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\Comfyui\ComfyUI\custom_nodes\comfyui-sage-guard\__init__.py", line 135, in __call__
    return self._current_sdpa(
           ^^^^^^^^^^^^^^^^^^^
TypeError: scaled_dot_product_attention() takes from 3 to 6 positional arguments but 7 were given

2026-07-20T22:46:46.205070 - �[32m[INFO]�[0m �[32mPrompt executed in 0.13 seconds�[0m
2026-07-20T22:46:50.237373 - FETCH ComfyRegistry Data: 125/1632026-07-20T22:46:50.237373 - 

Attached Workflow

Please make sure that workflow does not contain any sensitive information such as API keys or passwords.

Workflow too large. Please manually upload the workflow from local file system.

Additional Context

(Please add any additional context or steps to reproduce the error here)
`

[lbouaraba]

Hi @ameen-roayan, this is actually a different error from the one this issue was about, and it is not coming from this node. Looking at your stack trace:

File "...\custom_nodes\comfyui-sage-guard\__init__.py", line 135, in __call__
    return self._current_sdpa(
TypeError: scaled_dot_product_attention() takes from 3 to 6 positional arguments but 7 were given

The crash happens inside the Qwen3-VL text encoder attention (comfy/text_encoders/qwen35.py), which is ComfyUI core code, and the actual failing function belongs to comfyui-sage-guard, not to this node. You are running with --use-sage-attention, and on ComfyUI 0.28.0 the core attention now passes an extra argument to scaled_dot_product_attention that the sage-guard shim does not accept yet, so it throws before anything of mine runs. This node only makes a standard text-encode call; ComfyUI stock Krea2 text encode would hit the same error on your setup.

Three ways to fix it, any one works:

1. Update comfyui-sage-guard (the author needs to handle the new argument), or
2. Disable/remove comfyui-sage-guard, or
3. Launch without --use-sage-attention.

The original wrapper crash from this issue is fixed in v1.2.2 (you should see [krea2edit] nodes v1.2.2 loaded at startup, which confirms it is active). If disabling sage-guard does not resolve your error, please open a new issue with the fresh log and I will take a look. Thanks!