ComfyUI Node
RedNode Describe To Boxes
Describes an image into labelled sections and hands them to the Prompt Frame boxes. Wire subject/surroundings/light_and_colour into the matching inputs on RedNode Prompt Frame.
RedNode Describe To Boxes
- image
- subject
- surroundings
- light_and_colour
- raw
- notice
◄engineOllama►
◄model▾►
◄instructionSort what you see in this image into exactly five labelled sections, in this order, each on its own line:
SUBJECT: the main thing only. What it is, what it is wearing or made of, its pose, its expression, its own colours and materials.
SURROUNDINGS: the place only. The setting, the background, the objects near it.
LIGHT: where the light comes from, its direction and how hard or soft it is.
COLOUR: the overall palette in a few words.
MOOD: the feeling, in a few words.
Rules. Every label appears exactly once and is never empty. If the main thing is lit or shadowed, that belongs in LIGHT, not SUBJECT. Do not name the place in SUBJECT. Write nothing before SUBJECT and nothing after the MOOD line. No bullets, no markdown, no preamble.►
◄max_tokens420►
◄seed1►
CategoryRedNode/Prompt
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | The picture to describe. | |
| engine | COMBO | Ollama | Which captioner does the describing. Ollama is the only one verified so far; the ComfyUI QwenVL nodes currently cannot execute (their process() wants an argument their INPUT_TYPES never declares). |
| model | COMBO | A vision model. Anything that follows instructions will do; tested on qwen3-vl 8b instruct at Q4. | |
| instruction | STRING | Sort what you see in this image into exactly five labelled sections, in this order, each on its own line: SUBJECT: the main thing only. What it is, what it is wearing or made of, its pose, its expression, its own colours and materials. SURROUNDINGS: the place only. The setting, the background, the objects near it. LIGHT: where the light comes from, its direction and how hard or soft it is. COLOUR: the overall palette in a few words. MOOD: the feeling, in a few words. Rules. Every label appears exactly once and is never empty. If the main thing is lit or shadowed, that belongs in LIGHT, not SUBJECT. Do not name the place in SUBJECT. Write nothing before SUBJECT and nothing after the MOOD line. No bullets, no markdown, no preamble. | What the captioner is asked for. The default is the wording that scored 100% on format across 30 captions. Change the section names here and the parser will not find them. |
| max_tokens | INT | 42064–2048 | Upper bound on the reply. Too low truncates MOOD. |
| seed | INT | 10–18446744073709550000 | Same seed gives the same description. |
Outputs (5)
| Name | Type | Description |
|---|---|---|
| subject | STRING | — |
| surroundings | STRING | — |
| light_and_colour | STRING | — |
| raw | STRING | — |
| notice | STRING | — |