Few-step video on MiniMax H3, decoded without the VAE
DiR generates a clip in 4 iterations instead of 20 steps. DWD reads the pixels straight off the transformer instead of running the VAE decoder. Every clip here is real model output, rendered on one RTX 5090.
Update, 22 November: pi-H3 is the faster route
The DWD authors released pi-H3, the official 2- and 4-step MiniMax H3 with DWD built in, by Hansheng Chen, Shengqu Cai, Jiaming Song, Gordon Wetzstein, Xinchen Yan and Leonidas Guibas. We measured it on the same RTX 5090, 2 steps with its DWD, through their ComfyUI nodes, with dense attention and with Spark-Attn as in their RTX 50-series workflow.
| pi-H3 2-step + DWD, RTX 5090 | Dense | Spark-Attn |
|---|---|---|
| 768×512, 39 frames (1.6 s) | 1.39 s | 1.43 s |
| 1344×768, 39 frames | 4.98 s | 5.09 s |
| 768×512, 124 frames (5.2 s) | 6.49 s | 4.92 s |
| 1344×768, 124 frames | 29.6 s | 12.9 s (2.3×) |
Transformer plus DWD decode, after warm-up. Spark stays dense below 12,288 tokens, so short clips gain nothing. On 5-second clips it keeps the scene and motion and changes fine detail. For comparison, DiR with our DWD takes 4.1 s at 768×512 and runs out of memory at 1344×768 × 124 frames. More on the explainer.
Time per 39-frame clip, 768×512, RTX 5090 (DiR, 20 November)
A 39-frame clip is 1.6 s of video at 24 fps, so 1.0 s is faster than real time. The last two rows make a different, softer video than full-resolution DiR, not the same video sooner. DWD's decode alone: 0.26 s against the VAE's 1.8 s.
DWD against the VAE
Same DiR output, three decoders side by side. LPIPS against the VAE on held-out clips: linear head 0.267, with the refiner 0.114.
The DiR generator
Fewer tokens: MPEG-Diffusion's two-track sampler
The v0.2 refiner also learns the two-track region (LPIPS against the VAE 0.065 → 0.041 at 768×512). Two-track stays softer than full resolution; details in issue #7.
Numbers
| 39 frames, RTX 5090 | 768×512 | 1344×768 | DWD LPIPS vs VAE |
|---|---|---|---|
| DiR + VAE | 5.49 s | 16.77 s | – |
| DiR + DWD | 4.19 s | 13.95 s | 0.100 / 0.108 |
| Two-track fast path + VAE | 3.08 s | 7.57 s | – |
| Two-track, DWD on both tracks | 1.84 s | 5.07 s | 0.219 / 0.181 |
| Half res + DWD, pixel ×2 | 1.02 s | 2.88 s | 0.223 / 0.169 |
Medians of 2 warm prompts. LPIPS compares DWD with the VAE decode of the same latent; it does not compare one path with another.
Run it yourself
Needs a 32 GB card with no block swap. DWD decodes only the published DiR checkpoint.
git clone -b feat/accel https://github.com/scrya-com/dir-diffusion && cd dir-diffusion
hf download johndpope/dir-diffusion-h3-dwd --local-dir dwd
python3 lot/accel_h3.py check --run johndpope/dir-diffusion-h3 \
--lora dwd/dwd_lora.pt --head dwd/dwd_head.pt --refiner dwd/dwd_refiner.pt