Отладка моделей в SD.Next и Diffusers
★ 6.5 · testing
debug-model is a Claude Code skill that diagnoses and fixes broken model integrations in SD.Next and Diffusers. It covers six failure layers: model misdetection and routing, pipeline loader errors, raw checkpoint issues including key remapping and shape inference, prompt encoding failures, forward pass crashes during sampling, and silent wrong outputs with no exception raised. Debugging follows an outside-in order — starting with `modules/sd_detect.py`, `sd_models.py`, and `modeldata.py`, then narrowing to pipeline-specific files. The skill is built for developers porting new architectures or
- #model-debugging
- #diffusers-integration
- #pipeline-diagnosis
- #checkpoint-loading
- #inference-path