quota issues (#2)

drop HF token (#1)

Give enhance 90s budget, cap PE generation at 1024 tokens

Fix default node layout spacing; duration=60 so xlarge 2x request stays under free-tier cap

Keep PE models resident on GPU, cap durations for free tier, shorter enhance budget

Install transformers from git main (like diffusers)

Use xlarge ZeroGPU; require transformers>=5.15 for guard config format

Add PE-T2I/PE-I2I prompt enhancement nodes and NCII guard for image-input requests

Remove sample condition image; users upload their own

Prefill reference values via node data, wire condition image, handle canvas file URLs in edit_image

Prefill prompt/instruction/steps defaults; chain edit node after generation

Return canvas-renderable {path,url,is_file} dicts from bound functions

Handle None for unconnected number ports

Add torchvision (required by Qwen3VLVideoProcessor)

Rely on implicit HF_TOKEN env auth, fail fast with clear message

Pass access_token per Hub auth docs

Pass HF_TOKEN for gated model repo

Add Qwen-Image 2.1 ZeroGPU workflow app

initial commit

Duplicate from gradio-templates/workflow