Spaces:
Running
Apply for a GPU community grant: Personal project
Description of the app: A Gradio demo of MeiGen-AI/InfiniteTalk (Apache 2.0) — upload a portrait image and a speech clip, get a 480p lip-synced talking-head video. It uses the quantized fp8 T5 + quantized DiT so the whole pipeline fits a 48 GB ZeroGPU slot; audio is capped at 20 s per call. The app is fully written for ZeroGPU (import spaces, module-scope .to('cuda'), @spaces.GPU with a duration estimator) and only needs the hardware attached.
Justification: Public open-source demo focused on evaluating audio-driven lip sync for low-resource languages (Kazakh and Russian voice tracks) — there is currently no public hosted InfiniteTalk demo that runs the model itself rather than proxying a paid API. The account is new, so the free ZeroGPU slots are not yet available, and I can't cover dedicated hardware. Happy to provide more context if helpful.