FramePack: what the repository publishes
FramePack states its memory floor together with what the floor buys: to generate a one-minute video at 30fps, 1800 frames, using the 13B model, the minimal required GPU memory is 6GB. It names the card generations that support the run and the ones nobody tested. As of 2026-09-22.
| Field | What is stated | Detail |
|---|---|---|
| Licence | Apache 2.0 | Stated on the repository |
| Size | 13B models | The size the memory floor is stated for |
| Hardware | 6GB minimum for 1800 frames at 30fps | RTX 30XX, 40XX, 50XX with fp16 and bf16 |
| Resolution | No output size recorded | The repository states length and memory instead |
| Length | 1 minute, 1800 frames at 30fps | Seconds and frames, tied to the memory floor |
Inclusion rule. The same five fields for every family here. A field the repository does not state is recorded as not stated rather than estimated from the parameter count. Order. Fixed field order, identical on every family page.
1A floor attached to an output, which is the form every other entry here lacks
Every other hardware figure in this register answers whether the weights will load. This one answers what they produce at that memory: one minute, 30fps, 1800 frames, from a 13B model, at 6GB. The two halves of the sentence are inseparable, which is why the row quotes it whole.
It also demolishes the intuition the size column keeps inviting. Thirteen billion parameters sits beside a 60GB figure in one entry here and a 6GB figure in this one, so the parameter count is not what decided the memory. What decided it is how the work is broken up over time, and a parameter count cannot see that.
2Naming the card generations, and naming the ones nobody checked
The requirement names Nvidia GPUs in the RTX 30XX, 40XX and 50XX series that support fp16 and bf16, and adds that the GTX 10XX and 20XX are not tested. Publishing what was not tested is rarer than publishing what was, and for anybody holding older hardware it is the more useful half of the sentence.
The numeric formats are part of the requirement rather than a footnote to it. A card with enough memory and no bf16 support is not covered by this figure at all, which is precisely why this register quotes hardware statements instead of reducing them to a number of gigabytes.
3Speed given as a comparison between named laptops
The repository puts a per-frame time against an RTX 4090 and then says that on the maintainer's own 3070ti and 3060 laptops it runs about four to eight times slower. That is a range rather than a benchmark, and it is expressed in the unit that matters at the low end: how much longer the cheap card takes.
Nothing else in this register makes that comparison. Where other repositories publish timings at all, they are measured on data-centre cards, so the question of what a laptop does is normally unanswerable from published material.
4The same field, across every family
| Family | Licence |
|---|---|
| Allegro | apache-2.0 |
| CogVideoX | Split by model size |
| Cosmos Predict 2.5 | NVIDIA Open Model License |
| FramePack | Apache 2.0 |
| HunyuanVideo | tencent-hunyuan-community |
| LongCat-Video | MIT License |
| LTX-Video | Apache 2.0 |
| MAGI-1 | Apache License 2.0 |
| Mochi 1 | Apache 2.0 |
| Open-Sora | Apache 2.0 |
| Pyramid Flow | stabilityai-ai-community |
| SkyReels-V2 | skywork-license |
| Stable Video Diffusion | stable-video-diffusion-community |
| Step-Video-T2V | MIT |
| Wan | Apache 2.0 |
Inclusion rule. Families whose vendor has published weights openly. A family whose repository does not state this particular field keeps its row, marked hollow. Order. Alphabetical by family name.
| Family | Size |
|---|---|
| Allegro | VAE 175M and DiT 2.8B |
| CogVideoX | 2B and 5B, carried in the model names |
| Cosmos Predict 2.5 | 2,059,174,912 parameters |
| FramePack | 13B models |
| HunyuanVideo | Over 13 billion parameters |
| LongCat-Video | 13.6B parameters |
| LTX-Video | A 13B model |
| MAGI-1 | 24B and 4.5B models |
| Mochi 1 | 10 billion parameters |
| Open-Sora | An 11B model |
| Pyramid Flow | No parameter count recorded |
| SkyReels-V2 | 1.3B, 5B and 14B variants |
| Stable Video Diffusion | 2B params |
| Step-Video-T2V | 30 billion parameters |
| Wan | A 5B text-to-video and image-to-video model |
Inclusion rule. Families whose vendor has published weights openly. A family whose repository does not state this particular field keeps its row, marked hollow. Order. Alphabetical by family name.
| Family | Hardware |
|---|---|
| Allegro | 9.3G BF16 with cpu_offload |
| CogVideoX | From 4GB in FP16 on the 2B |
| Cosmos Predict 2.5 | 32.54 GB of GPU VRAM |
| FramePack | 6GB minimum for 1800 frames at 30fps |
| HunyuanVideo | 60GB peak at 720px1280px, 129 frames |
| LongCat-Video | No memory figure recorded |
| LTX-Video | No minimum stated |
| MAGI-1 | H100 or H800 x 8 for the 24B |
| Mochi 1 | About 60GB VRAM on a single GPU |
| Open-Sora | 60.3GB peak at 768px on one GPU |
| Pyramid Flow | No memory figure recorded |
| SkyReels-V2 | About 14.7GB peak on the 1.3B at 540P |
| Stable Video Diffusion | About 180s on an A100 80GB card |
| Step-Video-T2V | 78.55 GB peak at 768px768px204f |
| Wan | A consumer 4090 on the 5B release |
Inclusion rule. Families whose vendor has published weights openly. A family whose repository does not state this particular field keeps its row, marked hollow. Order. Alphabetical by family name.
- LicenceApache 2.0stated on the repository
- Model size13B modelsthe size the memory floor is stated for
- HardwareTo generate a 1-minute video at 30fps, 1800 frames, using the 13B model, the minimal required GPU memory is 6GBa floor stated together with the output it buys
- HardwareNvidia GPU in RTX 30XX, 40XX, 50XX series that supports fp16 and bf16; the GTX 10XX/20XX are not testedcard generations and numeric formats
5Sources
Read from github.com/lllyasviel on 2026-09-22. Hardware figures across every family are collected on requirements; the licence column is set out on licence and what counts as a published weight is on how read. Previous family: Stable Video Diffusion. Next family: Pyramid Flow.