Sam Kumar
db4f667203
📝 (docs): update documentation to reflect changes in base_url input and its usage ♻️ (config_api, media_describe): refactor code to resolve LM Studio base URL from settings or environment variables for better maintainability
1.8 KiB
1.8 KiB
Summary
Generates JSON-structured descriptions for a single image using LM Studio's schema-enforced responses. Default behavior targets photorealistic character analysis with the Visionary Image Architect system prompt that locks appearance, expression, pose, and clothing outputs for Wan 2.2 workflows.
Inputs
base_url(optional override): Leave blank to use the Swiss Army Knife LM Studio Base URL setting; fill only to temporarily target a different endpoint.model_name: Any loaded LM Studio multimodal model (defaults toqwen3-vl-8b-thinking-mlx).image: IMAGE tensor to analyze.schema_preset: Choosesimple_descriptionorcharacter_analysis(default). Character mode uses the new architect prompt described above.system_prompt: Override instructions if you need a custom style.user_prompt: Free-form request text.temperature,top_p,max_tokens,verbose: Standard generation controls.
Outputs
json_output: The entire LM Studio response string.field_1…field_5: Individual schema fields mapped in order (appearance, expression, pose, clothing when usingcharacter_analysis).
Usage Tips
- Keep
schema_presetoncharacter_analysisto leverage the Visionary Image Architect default; switch tosimple_descriptionfor lightweight captions. - Use
verbose=Truewhile troubleshooting to inspect payloads and LM Studio errors in the console. - The system prompt enforces decisive wording—omit hedging phrases in your own overrides to maintain consistency.
- Downstream nodes can parse the JSON string or read the individual field outputs without extra string handling.