Repository navigation
Layout and Box Edits
The distilled model with the flux3 preset (Euler, FLUX.3, 16 steps, CFG 1), flux3_image_klein_sd_int8_convrot, text encoder qwen3vl_4b_int8_convrot. The coloured frames are drawn on top afterwards, to show the boxes of each prompt; the model does not draw them.
The boxes follow Black Forest Labs' guide: a JSON list at the end of the prompt, each element with an id, a bbox of [top, left, bottom, right] on a 0-1000 grid and a desc, and the caption names each element as <id>.
 the panel |
 the result |
The companion Regional Boxes extension writes the list for you, in txt2img and img2img: drag on the frame (shaped like the picture's width and height) to draw a box, give it a name and a description, and click Write to prompt. The <name> button puts each tag at the cursor in the caption. The two boxes above, a teapot and a vase, made this prompt (16 steps, CFG 1, random seed):
Full-color studio photograph of a still life on a plain off-white backdrop: a glossy red ceramic teapot <teapot_1> and a tall cobalt-blue glass vase <vase_1> on a pale oak tabletop. [{"id": "teapot_1", "bbox": [102, 101, 465, 448], "desc": "A round glossy red ceramic teapot with a curved spout."}, {"id": "vase_1", "bbox": [541, 599, 947, 932], "desc": "A tall slender cobalt-blue glass vase."}]
The vase rises above its box: boxes guide placement and scale, they are not a hard mask.
For box edits, each element's kind is add new (from: target), from reference (from: ref_image_N, drawn twice: where it is, then where it goes) or remove (where it is only); in img2img the input picture shows behind the frame.
The example of Black Forest Labs' model card:
Minimalist graphic illustration featuring a black silhouette of a person <silhouette_1> centered against a solid, vibrant chartreuse background <background_1>. The figure is captured in a dynamic, mid-stride running pose, facing toward the left side of the frame. ["id": "background_1", "bbox": [0, 0, 1000, 1000], "desc": "A flat field of neon yellow-green with a subtle paper texture and fine, organic grain.", "id": "silhouette_1", "bbox": [150, 150, 850, 850], "desc": "A black silhouette of a person in motion, made of dense stippled ink."]
1024×1024, seed 42 - about 30 seconds.
Full-color studio photograph of a still life on a plain off-white backdrop with soft, even daylight from the left: a glossy red ceramic teapot <teapot_1> and a tall cobalt-blue glass vase <vase_1> stand far apart on a pale oak tabletop <table_1>, casting soft shadows to the right. ["id": "teapot_1", "bbox": [80, 60, 420, 400], "desc": "A round glossy red ceramic teapot with a curved spout and a small lid.", "id": "vase_1", "bbox": [480, 600, 950, 940], "desc": "A tall slender cobalt-blue glass vase, translucent, with light passing through it.", "id": "table_1", "bbox": [430, 0, 1000, 1000], "desc": "A pale oak tabletop with fine straight grain."]
1024×1024, seed 42 - about 35 seconds. Red: the teapot's box; blue: the vase's; orange: the table's.
img2img with [Klein] Enable Reference on and Denoising strength 1, on the mug photo of Black Forest Labs' examples (examples/inputs/2.png). Each element says where it comes from (from: ref_image_0 is the first reference, target is something new), where it is in the reference (src_bbox) and where it goes (tgt_bbox; null removes it). Red: src_bbox; green: tgt_bbox. 1024×1024, seed 1 - about a minute each.
 input |
 add |
Black Forest Labs' own cactus example prompt:
Add a small potted cactus <obj_1> on the wooden tabletop in the lower-right area indicated by the locked placement box, with its pot resting near the lower-right edge of the white coffee mug <mug_1>. Give it a compact green cactus in a simple terracotta pot, matching the scene’s perspective, warm sunlight, and natural contact shadow. Keep the coffee mug and every other part of the image unchanged. ["id": "obj_1", "from": "target", "src_bbox": null, "tgt_bbox": [550, 620, 850, 850], "desc": "A small potted cactus with a compact upright green barrel cactus featuring several short vertical ribs and evenly spaced clusters of rounded spines, planted in dark soil inside a simple unadorned terracotta pot with a slightly flared rim.", "id": "mug_1", "from": "ref_image_0", "src_bbox": [354, 283, 768, 755], "tgt_bbox": [354, 283, 768, 755], "desc": "A large plain white ceramic coffee mug with a rounded body and broad curved handle, filled with dark black coffee and a thin ring of steam rising above the rim."]
 remove |
 move |
Remove the white coffee mug <mug_1> from the wooden table, leaving the bare honey-brown planks with their grain, the bright band of morning window light and the window behind exactly as they are, with no trace or shadow of the mug. Keep every other part of the image unchanged. ["id": "mug_1", "from": "ref_image_0", "src_bbox": [354, 283, 768, 755], "tgt_bbox": null, "desc": "A large plain white ceramic coffee mug with a rounded body and broad curved handle, filled with dark black coffee and a thin ring of steam rising above the rim."]
Move the white coffee mug <mug_1> to the left side of the wooden table, into the placement box, with its handle still facing frame-right and its pale shadow falling toward the lower right; the spot where it stood is now bare honey-brown planks. Keep the window, the light and every other part of the image unchanged. ["id": "mug_1", "from": "ref_image_0", "src_bbox": [354, 283, 768, 755], "tgt_bbox": [354, 40, 768, 512], "desc": "A large plain white ceramic coffee mug with a rounded body and broad curved handle, filled with dark black coffee and a thin ring of steam rising above the rim."]