Wan2.1-T2V-14B — ONNX
ONNX export of Wan2.1-T2V-14B for ONNX Runtime (text-to-video). The 4-bit DiT runs 81 frames at 480×832 within 24 GB of VRAM.
Setup, usage, and conversion: https://github.com/AkaneTendo25/wannx
Files
| File | Description | Size |
|---|---|---|
dit.onnx (+ .data) |
DiT, 4-bit MatMulNBits, fp16 attention |
~7.8 GB |
dit_fp16.onnx (+ .data) |
DiT, fp16 | ~27 GB |
t5_encoder_fp16.onnx (+ .data) |
umt5-xxl text encoder | ~13 GB |
vae_decoder.onnx (+ .meta.json) |
VAE decoder, streaming step graph (latents 1+) — any resolution, fits 24 GB | ~0.3 GB |
vae_decoder_init.onnx |
VAE decoder, streaming init graph (latent 0) — bit-exact first frame | ~0.3 GB |
google/umt5-xxl/ |
tokenizer | ~20 MB |
License: Apache-2.0 (inherited from Wan2.1).
Model tree for AkaneTendo25/wan21-onnx
Base model
Wan-AI/Wan2.1-T2V-14B