ComfyUI Node
Media Describe
Analyzes an image or video (via Media Selection) with Gemini/LLM Studio, then outputs the full structured JSON, flattened prompts, and inferred dimensions for downstream control-panel and prompt-building nodes.
Media Describe
- llm_studio_options
- overrides
- all_media_describe_data
- raw_llm_json
- positive_prompt_json
- positive_prompt
- prompt_request
- height
- width
◄media_processed_path►
◄use_custom_promptsfalse►
◄custom_system_promptGenerate a Wan 2.2 optimized text to image prompt. You are an expert assistant specialized in analyzing and verbalizing input media for instagram-quality posts using the Wan 2.2 Text to Image workflow.
DECISIVENESS REQUIREMENT: Always provide definitive, certain descriptions. When you see something that could be described multiple ways, make a confident choice and state it as fact. Never use uncertain language like "appears to be", "seems to be", "might be", "possibly", "likely", or "or". Never mention watermarks, logos, branding, or any textual overlays.
Return **only** a single valid JSON object (no code fences, no extra text) with **exactly five** string fields in this exact order:
1. "subject" - Detailed description of the main subject
2. "clothing" - Clothing and style details
3. "movement" - Pose, gesture, or implied motion
4. "scene" - Setting, environment, and background elements
5. "visual_style" - Combined lighting, camera details, rendering cues, mood/genre descriptors, and overall aesthetic direction
Each field's value is one fully formed paragraph (a single string) for that category.►
◄custom_user_promptPlease analyze this image and provide a detailed description in the JSON format specified in the system prompt.►
CategorySwiss Army Knife 🔪/Media Caption
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| media_processed_path | STRING | Path to the processed media file from Media Selection node | |
| llm_studio_optionsopt | LLM_STUDIO_OPTIONS | Configuration options from LLM Studio - Options node | |
| overridesopt | OVERRIDES | Paragraph overrides from Media Describe - Overrides node (optional) | |
| use_custom_promptsopt | BOOLEAN | false | Toggle to send manual system/user prompts to LM Studio instead of the auto-generated templates |
| custom_system_promptopt | STRING | Generate a Wan 2.2 optimized text to image prompt. You are an expert assistant specialized in analyzing and verbalizing input media for instagram-quality posts using the Wan 2.2 Text to Image workflow. DECISIVENESS REQUIREMENT: Always provide definitive, certain descriptions. When you see something that could be described multiple ways, make a confident choice and state it as fact. Never use uncertain language like "appears to be", "seems to be", "might be", "possibly", "likely", or "or". Never mention watermarks, logos, branding, or any textual overlays. Return **only** a single valid JSON object (no code fences, no extra text) with **exactly five** string fields in this exact order: 1. "subject" - Detailed description of the main subject 2. "clothing" - Clothing and style details 3. "movement" - Pose, gesture, or implied motion 4. "scene" - Setting, environment, and background elements 5. "visual_style" - Combined lighting, camera details, rendering cues, mood/genre descriptors, and overall aesthetic direction Each field's value is one fully formed paragraph (a single string) for that category. | System prompt to send when Use Custom Prompts is enabled |
| custom_user_promptopt | STRING | Please analyze this image and provide a detailed description in the JSON format specified in the system prompt. | User prompt to send when Use Custom Prompts is enabled |
Outputs (7)
| Name | Type | Description |
|---|---|---|
| all_media_describe_data | STRING | — |
| raw_llm_json | STRING | — |
| positive_prompt_json | STRING | — |
| positive_prompt | STRING | — |
| prompt_request | STRING | — |
| height | INT | — |
| width | INT | — |