ComfyUI Node
MOSS Layout from Qwen-VL
A ComfyUI node in MOSS-Image with 17 inputs and 5 outputs.
MOSS Layout from Qwen-VL
- image
- clip
- prompt
- preview
- raw_qwen_output
- width
- height
◄width0►
◄height0►
◄targetsperson►
◄metaphoto►
◄coord_modenormalized_0_1000►
◄max_new_tokens4096►
◄temperature0.20►
◄seed0►
◄hf_modelQwen/Qwen3-VL-8B-Instruct►
◄keep_loadedfalse►
◄change_request►
◄box_padding0.0►
◄smart_modefalse►
◄max_items40►
◄style_in_itemstrue►
CategoryMOSS-Image
Inputs (17)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | — | |
| width | INT | 00–8192 | Target MOSS canvas width. 0 = use the image's width (rounded to 16). |
| height | INT | 00–8192 | Target MOSS canvas height. 0 = use the image's height (rounded to 16). |
| targets | STRING | person | One thing to detect per line. Ignored/optional extras when smart_mode is on. For text, say so explicitly, e.g. 'title text EXPLORE'. Text is only handled when you ask for it. |
| meta | STRING | photo | Style lock. Comma-separated tags, e.g. photo, poster. The overall caption is built around them (added if Qwen forgets) and words from other styles are stripped from all captions. |
| coord_mode | COMBO | normalized_0_1000 | Qwen3-VL/3.5 use normalized 0-1000. Use absolute_pixels for Qwen2.5-VL style models. |
| max_new_tokens | INT | 4096256–16384 | — |
| temperature | FLOAT | 0.200–2 | — |
| seed | INT | 00–18446744073709550000 | — |
| clipopt | CLIP | Qwen3-VL / Qwen3.5-VL loaded as a text encoder. If unconnected, hf_model is used. | |
| hf_modelopt | STRING | Qwen/Qwen3-VL-8B-Instruct | HF repo id or local folder. Used only when clip is not connected. |
| keep_loadedopt | BOOLEAN | false | HF path only: keep Qwen in memory after running. |
| change_requestopt | STRING | Describe edits to make, one per line. E.g. 'make the dress red', 'replace the man with an elderly woman in a grey coat', 'add a dog on the left', 'change the sign text to OPEN'. Leave empty to just describe the image. | |
| box_paddingopt | FLOAT | 0.00–25 | Grow every box by this % of its size on each side (clamped to the canvas). Extra safety on top of Qwen estimating hidden/occluded parts. |
| smart_modeopt | BOOLEAN | false | Qwen decides what to detect by itself: people and their body parts, clothing, objects and background regions, to match the reference layout as closely as possible. Anything in targets is still included. |
| max_itemsopt | INT | 405–120 | Smart mode only: maximum number of regions Qwen may return. |
| style_in_itemsopt | BOOLEAN | true | Every item caption must carry the meta style (Qwen is asked to, and 'Rendered as <meta>.' is added if it doesn't). Turn off to keep style only in the overall caption. |
Outputs (5)
| Name | Type | Description |
|---|---|---|
| prompt | STRING | — |
| preview | IMAGE | — |
| raw_qwen_output | STRING | — |
| width | INT | — |
| height | INT | — |