Appearance
Choose a MiniMax H3 or FastH3 variant, then set the resolution, video length and reference media. See the estimated generation time and VRAM requirements on Olares One.
Model variant
Generate from a text prompt, with optional first and last frames.
Weight precision
Sizes shown are model weights.
Output resolution
Video length 5 s
124 frames after ComfyUI rounding
First and last frames
Estimated generation time
1.6 min
Sampling 77 s, video decoding 11 s, encoding and loading 6.0 s
VRAM needed
18.5 GiB
To keep the video generation model on the GPU
Memory placement
Model fits in VRAM
Video model weights stay on the GPU
Token and VRAM breakdown
| Segment | Tokens | GiB |
|---|---|---|
| Text and vision tokens | 150 | 0.02 |
| Output video | 37,296 | 4.96 |
| Output audio | 414 | 0.06 |
| Total + 0.04 constant | 37,860 | 5.03 |
| Each | Tokens | GiB |
|---|---|---|
| 480p output, per s | 2,915 | 0.39 |
| 720p output, per s | 7,136 | 0.95 |
| 720p frame or image | 2,024 | 0.27 |
| 720p ref video, per s | 8,072 | 1.05 |
| Audio, per s | 80 | 0.01 |