ComfyUI Node
Automatic Whole-Image Analysis + Cache
A ComfyUI node in Smart Upscaler/Prompting with 13 inputs and 3 outputs.
Automatic Whole-Image Analysis + Cache
- clip
- image
- prompt_system
- prompt_text
- cache_status
- review_text
◄prompt—►
◄max_length1024►
◄sampling_modeManaged by Prompt Director (recommended)►
◄thinkingfalse►
◄use_default_templatetrue►
◄cache_moderead_write►
◄cache_tagqwen3vl_4b_fp8_master_scene_v22_uniform_surfaces_v13►
◄key_context►
◄analysis_max_side1344►
◄vision_model_idQwen3-VL-4B-FP8►
CategorySmart Upscaler/Prompting
Inputs (13)
| Name | Type | Default | Description |
|---|---|---|---|
| clip | CLIP | — | |
| image | IMAGE | — | |
| prompt | STRING | — | |
| max_length | INT | 10241–32768 | — |
| sampling_mode | COMBO | Managed by Prompt Director (recommended) | Leave Managed. The Prompt Director owns semantic choices; the older Consistent and Varied values remain only for loading older workflows. |
| thinking | BOOLEAN | false | Leave off for a clean caption-only response. |
| use_default_template | BOOLEAN | true | Recommended ON for Qwen3-VL. Disable only when a replacement caption model explicitly expects raw unformatted text. |
| cache_mode | COMBO | read_write | 3 options: read_write, refresh, bypass |
| cache_tag | STRING | qwen3vl_4b_fp8_master_scene_v22_uniform_surfaces_v13 | — |
| key_contextopt | STRING | — | |
| prompt_systemopt | SMART_PROMPT_SYSTEM | — | |
| analysis_max_sideopt | INT | 13440–8192 | Longest side of the image sent to the vision model for the whole-image brief and its follow-up questions. Smaller = faster and less VRAM on every question; 1344 loses nothing for a scene brief. 0 sends the full image. |
| vision_model_idopt | STRING | Qwen3-VL-4B-FP8 | Cache identity for the connected caption model. Change this when replacing or updating that model; this prevents old captions from another model being reused. |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| prompt_text | STRING | — |
| cache_status | STRING | — |
| review_text | STRING | — |