Sam Kumar
894706d782
LM Studio endpoint usage and remove base_url parameter from input definitions for consistency ♻️ (llm_studio_structured.py): refactor code to remove base_url parameter from describe_image and describe_video methods, relying on global LM Studio endpoint configuration ✨ (swiss-army-knife.js): add debug logging for model dropdown interactions to enhance traceability and debugging capabilities in the UI
25 lines
1.7 KiB
Markdown
25 lines
1.7 KiB
Markdown
# 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
|
|
- LM Studio endpoint: configured globally in Swiss Army Knife settings (all nodes automatically consume it).
|
|
- `model_name`: Any loaded LM Studio multimodal model (defaults to `qwen3-vl-8b-thinking-mlx`).
|
|
- `image`: IMAGE tensor to analyze.
|
|
- `schema_preset`: Choose `simple_description` or `character_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 using `character_analysis`).
|
|
|
|
# Usage Tips
|
|
- Keep `schema_preset` on `character_analysis` to leverage the Visionary Image Architect default; switch to `simple_description` for lightweight captions.
|
|
- Use `verbose=True` while 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.
|
|
|
|
# Documentation
|
|
- Full guide: [docs/nodes/lm-studio-describe/STRUCTURED_OUTPUT_IMPLEMENTATION.md](../../docs/nodes/lm-studio-describe/STRUCTURED_OUTPUT_IMPLEMENTATION.md)
|