VidoGridOpen source video generation models
Open source video generation models you can download, run on your own GPU and fine-tune: releases from their GitHub repositories, the tools people run them with, and what is trending on Hugging Face this week.
Repository releases
64 frames- 206NewLightX2V 0.6.0ModelTC/LightX2V
- 205NewInvokeAI 7.0.0 (alpha 2)InvokeAI
- 204NewSana: security: harden workflows, pin actions, add SECURITY.md (#528)NVlabs/Sana
- 197NewHunyuanVideo: fix(load): load checkpoints with weights_only and pin torch 2.7.1Tencent-Hunyuan/HunyuanVideo
- 193Newmochi: Add Microsoft LoRA copyright and MIT notice to lora.py (#162)genmoai/mochi
- 121NewDiffusers 0.40.0: New pipelines, tensor-parallel support, improved CLI, and moreDiffusers
- 114NewLTX-2 v1.2.0Lightricks/LTX-2
- 113NewLTX-2 readme-demo-windowLightricks/LTX-2
- 104NewTurboDiffusion: Add vendored TurboT2AV implementation (#137)thu-ml/TurboDiffusion
- 100NewDiffusers 0.39.0: New image and video pipelines, core library improvements, and moreDiffusers
- 91NewFastVideo Release 0.2.0hao-ai-lab/FastVideo
- 89NewDiffusers 0.38.0: New image and audio pipelines, Core library improvements, and moreDiffusers
- 86NewTurboDiffusion: add discrete-time CMthu-ml/TurboDiffusion
- 85NewHunyuanVideo-1.5: fix(train): apply selective gradient checkpointing to single-stream b…Tencent-Hunyuan/HunyuanVideo-1.5
- 84NewDiffusers Fixes for AutoModel type hints in Modular Pipelines and Flux Klein LoRA loadingDiffusers
- 83NewWan2.2: Add Prompt Relay to community worksWan-Video/Wan2.2
- 82NewDiffusers 0.37.0: Modular Diffusers, New image and video pipelines, multiple core library improvements, and more 🔥Diffusers
- 77Newflux2: Update license information for FLUX.2 [dev] modelblack-forest-labs/flux2
- 76NewHunyuanImage-3.0: Add support for hunyuanimage-3.0-Instruct & hunyuanimage-3.0-Instruct…Tencent-Hunyuan/HunyuanImage-3.0
Open source video generation models trending on Hugging Face
| Repository | Task | Base model | License | Downloads | Likes |
|---|---|---|---|---|---|
| LTX-2.5Lightricks | image to video | — | other | 2.6M | 11k |
| Pika-H3-FlexPika-Labs · LoRA | image to video | MiniMaxAI/MiniMax-H3 | other | 863 | 476 |
| Pika-H3-Flex-Turbo-LoRAPika-Labs · LoRA | image to video | Iwannapose/minimax_h3_pdmd_2nfe_comfyui | other | 56k | 156 |
| MiniMax-H3-Turbo-Loralarryvrh · LoRA | text to video | Comfy-Org/MiniMax-H3 | apache-2.0 | 206k | 1.2k |
| DMADZhengmingYu · LoRA | text to video | MiniMaxAI/MiniMax-H3 | other | 108k | 162 |
| Minimax-h3_SingularityWarmBloodAban | image to video | MiniMaxAI/MiniMax-H3 | apache-2.0 | 1.1M | 961 |
| MiniMax-H3-X2-Detail-VAEspeach1sdef178 | image to video | MiniMaxAI/MiniMax-H3 | other | 43k | 120 |
| Wan2.2-TI2V-5BWan-AI | text to video | — | apache-2.0 | 19k | 928 |
| WaveForcing-1.3BSJTU-DENG-Lab | text to video | Wan-AI/Wan2.1-T2V-1.3B | apache-2.0 | 0 | 15 |
| LTX-VideoLightricks | image to video | — | other | 1.2M | 2.4k |
| Kandinsky-S2V-1.0-Pro-distill-Diffuserskandinskylab | image to video | — | mit | 88 | 14 |
| Wan2.1-T2V-1.3BWan-AI | text to video | — | apache-2.0 | 89k | 663 |
| Wan-2.2Comfy-Org | image to video | Wan-AI/Wan2.2-Animate-14B | apache-2.0 | 9.7M | 1.0k |
| Cobijada_Minimax-H3_Hybrid_Pruned_ComfyUIWinnougan | image to video | — | apache-2.0 | 548k | 101 |
| Minimax-H3-Hybrid-DMAD-PDMDOliviaRossi | text to video | Comfy-Org/MiniMax-H3 | other | 753 | 11 |
Running open source video generation models
Open weights change what you can do with a model. You can run it without per-clip fees, keep footage on your own machines, train a LoRA on a character or a style, and chain it with other tools in ComfyUI or the Diffusers library. The cost is hardware and setup: the largest open source video generation models want a GPU with plenty of memory, while distilled and quantised versions trade some quality for speed and smaller cards.
Most of the trending repositories above are not new base models but LoRAs, distillations and quantisations built on a few families such as Wan, HunyuanVideo, LTX-2 and MiniMax H3. The base model column shows what each one was trained from, and the license column tells you whether commercial use is allowed. Read the license on the model card before you ship anything.
For prompting open models, start with the wan 2.2 prompts guide; for closed models compare specs on the AI video model comparison.
Open image models trending on Hugging Face
| Repository | Task | Base model | License | Downloads | Likes |
|---|---|---|---|---|---|
| flux-3-image-klein-comfyblack-forest-labs | image to image | black-forest-labs/flux-3-image-klein | other | 1.1M | 2.6k |
| Qwen-Image-2.1-Turbo-Quantizedabenzerps | text to image | Qwen/Qwen-Image-2.1-Turbo | other | 233k | 5.7k |
| P-H3-Image-ComfyPrunaAI · LoRA | text to image | PrunaAI/P-H3-Image | other | 219k | 202 |
| P-H3-ImagePrunaAI | text to image | MiniMaxAI/MiniMax-H3 | other | 444 | 136 |
| MOSS-Image-SFTOpenMOSS-Team | text to image | OpenMOSS-Team/MOSS-Image-Base | apache-2.0 | 20k | 919 |
| FLUX.1-devblack-forest-labs | text to image | — | other | 518k | 16k |
| Qwen-Image-2.1Qwen | text to image | — | other | 238k | 3.6k |
| flux-3-image-kleinblack-forest-labs | image to image | — | other | 167k | 843 |
How this list of open source video generation models is built
VidoGrid reads the release and commit feeds of the main open model repositories every twenty minutes and keeps the ones that ship something: new weights, new modes, speed-ups, quantised builds. The Hugging Face tables are the most-trending repositories in the text-to-video, image-to-video, text-to-image and image-to-image tasks, refreshed daily.