HunyuanVideo: what the repository publishes
HunyuanVideo publishes weights for a model of over 13 billion parameters under a licence the vendor wrote itself rather than a standard permissive one. The repository states 60GB of peak GPU memory at its larger setting, measured on a single 80G card. As of 2026-09-12.
| Field | What is stated | Detail |
|---|---|---|
| Licence | tencent-hunyuan-community | A vendor-written licence, not a standard permissive one |
| Size | Over 13 billion parameters | Stated in the repository |
| Hardware | 60GB peak at 720px1280px, 129 frames | 45GB at 544px960px, on a single 80G GPU |
| Resolution | No output size recorded | Its pixel dimensions size the memory figures instead |
| Length | 129 frames, as a measurement setting | Frames, sizing a memory figure rather than a ceiling |
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 vendor-written licence is a different thing from Apache 2.0
Nine of the 15 families in this register carry a licence whose text is standard and whose terms a reader has cleared once and for all. This one carries a licence named after the company that wrote it, which means the terms have to be read rather than assumed.
This register records the licence name and does not summarise its terms. A one-line paraphrase of a bespoke licence is exactly the kind of helpfulness that turns into somebody relying on a wrong reading, and the licence file is one click from the row.

2The memory figures are stated per setting, which is unusually clear
Sixty gigabytes at the larger resolution and frame count, forty-five at the smaller. Most repositories publish one number or none; publishing two with the settings attached lets a reader work out which side of a hardware boundary their intended output falls on.
Both figures sit well above anything a desktop card offers, so this family is a rented-server proposition rather than a local one on the published evidence.
3What the register does not carry for this family
No hosted service in this register names this model on a priced page, so there is no seconds rate to set against the hardware. That leaves the cost comparison open in one direction: the hardware cost is knowable and the alternative is not.
The row will appear if a service starts naming it. Until then the family sits in this register as published weights with no commercial counterpart to compare against.
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.
- Licencetencent-hunyuan-community, a vendor-written licence rather than a standard permissive oneshown on the model card
- Model sizeOver 13 billion parametersstated in the repository
- HardwarePeak GPU memory of 60GB at 720px1280px and 129 frames, and 45GB at 544px960pxthe vendor's own figures
- HardwareTested on a single 80G GPUwhat the figures were measured on
5Open weights, hosted
Instead of local hardware for “HunyuanVideo”, SceneMixer runs hosted models; its guides describe the result:
Typically 60–120 seconds, always 9:16 vertical, ending on a hook. SceneMixer's episode outlines are generated with that pacing, and setting a project to 9:16 makes every segment vertical (the default is 16:9).
SceneMixer, How to Make ReelShort-Style Dramas with AI — “How long is a ReelShort-style episode?” (read 2026-10-13)
Doubao Seed 2.1 Pro is the family’s full-size tier at 100% of the base text rate, twice the default. Same three jobs — reading, writing, storyboards — with more headroom on long, crowded episodes: it holds the storyboard’s strict format more reliably and keeps track of many characters and locations in one pass.
SceneMixer, Doubao Seed 2.1 Pro (read 2026-10-13)
6Sources
Read from github.com/Tencent-Hunyuan on 2026-09-12. 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 the method page. Previous family: LongCat-Video. Next family: CogVideoX.