Background
As the library of user models expands, the base recommendations may not exactly match the demand. Copilot can be trained to provide more accurate model recommendations through the following methods.
Core Solutions
- feedback training: Implemented after each recommendation:
- Rate the results of the recommendation (1-5 stars)
- Enter "why this score" to explain why.
- Use "better alternative" to mark more suitable models
- Feature labeling: Enhance model understanding with the following commands:
- "tag model [name] as [style]" (e.g. "tag model portraitLoRA as photorealistic")
- "set model priority [name] [1-10]" sets personal preferences
- "hide model [name]" hides irrelevant models
- Scene Optimization: Improving Matching Accuracy Using Scenario Prefixes:
- "commercial ad:" (commercial advertising requirements)
- "scientific visualization:"
- "mobile app UI:"
practical skill
- Periodically synchronize the community with new models using the "update model database".
- Create a personal model collection: "create model collection [name]"
- For specialized areas, use the "train custom recommender" to upload sample data.
Summary points
With continuous feedback and scenario labeling, it usually takes 2-3 weeks to improve recommendation accuracy by more than 40%. It is recommended to deeply label the models involved in the core workflow, and Copilot will prioritize the labeled models.
This answer comes from the articleComfyUI-Copilot: an AI assistant for text description generation ComfyUI workflowsThe
May not be reproduced without permission:AI productivity tools " How can I improve ComfyUI-Copilot's model recommendation accuracy?
