StreamDiffusion V2 - 256x256 ONNX export (TiXL)
Ready-to-run ONNX export of the StreamDiffusionV2 causal Wan2.1 1.3B video-to-video model: causal DMD DiT (fp16, 256x256, KV-cache batch profile 1-4), UMT5-XXL text encoder, TAEHV VAE and tokenizer.
Produced with the export scripts bundled with TiXL (Operators/StreamDiffusion/tools/streamdiffusionv2). The TiXL model browser downloads these files directly - no Python needed on the target machine.
License: the underlying weights are Apache-2.0 (StreamDiffusionV2 checkpoint, Wan2.1-T2V-1.3B base) and the TAEHV autoencoder is MIT (madebyollin/taehv).
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support