data-archetype/canter-t2i-demo

Canter ZeroGPU demo

This Space runs the full Canter Gradio interface with eager dense inference on Hugging Face ZeroGPU. It uses the numbered checkpoint release selected by the bundled Canter package and Hugging Face's managed PyTorch runtime.

The ZeroGPU profile generates one image per request and does not offer latent previews because generation runs in an isolated GPU worker. The ordinary standalone Gradio launcher retains its multi-image, preview, compilation, and model-unload behavior.

The hosted limits match the original Hugging Face deployment: 180 seconds of GPU allocation, 10–100 solver updates, and image dimensions from 256 through 2048 pixels. Canter retains its 32-pixel web-UI size lattice inside those bounds.

Empty and whitespace-only prompts are rejected with Please enter a prompt.

The model and package remain subject to the licenses and responsible-use terms published with `data-archetype/canter`.