What problem does it solve? Publishing a demo for a LoRA on Hugging Face Spaces requires picking the correct diffusers pipeline class, designing a UI matched to the LoRA's specific task, configuring ZeroGPU hardware, and debugging build failures — a multi-step process where small mistakes produce silently broken Spaces. ## Core Features & Use Cases - Pipeline Selection and Verification: Reads the LoRA's model card and the base model's own card to pick the correct diffusers pipeline class (e.g. QwenImageEditPipeline vs QwenImageEditPlusPipeline), avoiding wrong-class failures. - Task-Adapted UI Design: Designs Gradio interfaces from the LoRA's actual inputs — pose extraction previews, aspect-ratio pickers, brush canvases, or custom HTML/JS widgets — rather than generic templates. - ZeroGPU Publishing and Smoke-Testing: Writes app.py, requirements.txt, and README.md with correct ZeroGPU configuration, publishes a private Space, and verifies it end-to-end with gradio info and gradio predict. - Use Case: A user trained a relighting LoRA on Qwen-Image-Edit and wants a shareable browser demo; the skill reads the model card, builds a Space with an image upload and lighting controls, publishes it privately, and smoke-tests inference before sharing the URL. ## Quick Start Build and publish a Hugging Face Space demo for my LoRA at username/my-lora.
