Release 1.3.0 · RedNodeAI/ComfyUI-RedNodeStudio

Watching the picture form, the Detailer as a proper list, and a handful of
things that only made one image when they should have made several.

RedNode Live Preview, a new node: wire the workspace's image output in and
the picture forms on it step by step, with a bar and a "rendering 4 / 27"
tag, then the finished frame. The frames are decoded by the pack itself with
the small VAE from models/vae_approx (lighttaew2_1 for Krea 2), so it works
whatever ComfyUI's own preview setting is. Wire the Detailer's output instead
to watch its passes; unwired, it shows whatever is rendering
The same frames land over the Paint tab's result pane while a paint run
samples, and come off the moment the result does
Every frame says which pass, round or shot it belongs to, and the Paint tab
picks its own frame size, 512 up to the decoder's full output

A full screen room: the corner glyph, a double-click or the right-click
menu opens it, Esc closes it. The wheel zooms the picture about the pointer,
a drag pans, the wheel over the strip scrolls the strip, arrows walk the
history
A batch column: a run that made several pictures shows them down the left
edge, click to view, the count reads "2 of 4", and the right-click actions
follow the picture that is up

Each pass card is a header and three boxes: Sampling, Strength and Prompt.
Scale is a bar beside Denoise, and a duplicate button sits by the delete on
passes and group titles
Repeat above one offers a denoise and a scale per round with a Ramp, the
same rule the Img2Img tab's passes follow
A pass picks which Prompts-tab row it reads; the rig's row stays the default
A SeedVR2 upscale pass: Size 720p, 1080p, 2K, 1440p or 4K as a pixel budget
with the short edge worked out from the frame's aspect, the loader dials on
the card, the pack's own nodes doing the work. Needs ComfyUI-SeedVR2 installed
The SAM file and precision are picked on the node, and the per-pass SAM list
now reads the V3 loader's choices, which had left it empty

Img2Img RE-ANGLE: a switch to skip the i2i pass, so the re-shot picture is
the output and the rig never enters VRAM beside the edit model
Img2Img PASS: a full-width denoise bar in hundredths, a denoise per pass and
a scale per pass with a Ramp, the decode tiled when it gets big
A camera path on an engine rig, the NovelAI rig, renders one image per shot
and batches them; before, it rendered shot 1 and said nothing. On the
external sampler a path still renders shot 1 only, and the console, the
Camera tab and the sampler-mode tip now say so
Post FX: the depth estimator is a Depth card on the post panel, both the
standalone node and the Post tab: which estimator, which Depth Anything V2
checkpoint, and the working resolution
Latent tab: refine passes on a blank canvas. Pass 1 generates at the full
denoise and every pass after it refines what pass 1 made, at a Refine dial
or a denoise and scale per pass, so the tab with no source image gets the
draft-and-climb run the Img2Img tab has
A rig can name a second sampler and scheduler that take over on image to
image runs; blank means the main pair, so nothing moves unasked

Video Review: the frame rate is a dial on the panel with the same presets
as Save Video, a wired fps socket wins over it, and a workflow saved before
the panel keeps the rate and loop it had. Sound and loop flip the player on
screen instead of rebuilding it and dropping back to the start
Save Video announces its clip as animated, so the file shows in ComfyUI's
outputs instead of only on the console

RedNodeStudio_V1.3 replaces the 1.2 Ultima file: the whole rig with the
Detailer's passes and the SeedVR2 upscale on the node, the Post FX chain,
three Image Reviews and two Live Previews watching the render and the passes
RedNode_MultiAngle replaces the earlier multi-angle file: the Camera Studio
drives a Qwen-Image-Edit-2511 re-shoot, and Krea 2 finishes the frame

The socket tuck sits on its own row above the tabs; it used to wrap onto a
stray line between the tabs and the box once the Camera tab filled the width
The README named the wrong folder for the SAM3 file; it is models/sam3
A pass latent resizes by its rank rather than an assumed four dims

Navigation Menu