The names in any AI video catalog (Veo, Seedance, Kling, Wan, Hailuo, Grok Imagine) come from big technology companies, each with a different strategy. Knowing who makes what explains why some models exist only as a service while others can be downloaded and run on your own GPU.
Who is who
| Model | Company | How you access it (public information at the time of writing) |
|---|---|---|
| Veo | Google DeepMind | service, no downloadable weights |
| Seedance | ByteDance | service, no downloadable weights |
| Kling | Kuaishou | service, no downloadable weights |
| Hailuo | MiniMax | service, no downloadable weights |
| Grok Imagine | xAI | service, no downloadable weights |
| Wan | Alibaba | service and open-weight versions in the family |
The table shows the pattern: the leading models are offered as a service, and the only family in the group with open versions is Wan. That is not a technical detail but a business decision: whoever does not publish the weights controls access, price and content policy.
Closed or open: what changes for you
Closed model (Veo, Seedance, Kling, Hailuo, Grok Imagine)
- You use it through a service. No installation, no GPU, no setup.
- Top quality, updated by the maker.
- You don't pick the weights, can't train your own style and follow the maker's content policy.
Open-weight model (Wan family and others)
- You download it and run it on a rented GPU, with your own workflow (ComfyUI, for example).
- You can train a LoRA for a character or style and chain post-production steps.
- It takes technical knowledge, a GPU with enough VRAM and setup time.
The GPUBrazil template catalog has a Wan 2.2 template that brings up a ready environment for running the open version.
How these models reach GenBrasil
GenBrasil brings the eight test models into one account, with a Portuguese interface, credits and payment in reais by Pix or card. Wan 3.0, on GenBrasil, is used as a service like the others; whoever wants the open line of the family runs it on their own GPU.
How to read the next release
When a new model ships, four questions settle most of the doubt:
- Are the weights open? If yes, you can run it on your GPU and the bill becomes machine-hours.
- What duration and resolution does it really deliver? Trust what comes out, not what the announcement promises.
- Does it generate audio? It changes the post-production flow and the cost.
- How much does it cost and how long does it take? Compare against the same prompt in another model, as in the eight-model comparison.
Frequently asked questions
Who makes Seedance?
ByteDance. Version 2.5 is a closed service with no weights to run locally.
Is there an open video model I can run on my GPU?
Yes. Alibaba's Wan family has open-weight versions, and there are other open video models. They need GPUs with a lot of VRAM.
Can I use closed and open models in the same project?
Yes. A common setup is to generate key images on a GPU with an open model and animate on GenBrasil with a closed model.
Make your video on GenBrasil
Eight AI video models in one account, with the cost shown before you generate and payment in reais by Pix or card. The interface is in Portuguese. The Free plan needs no card.
Go to GenBrasil ↗Rather build your own workflow? See the hourly GPU templates.
Read next: character LoRA on GPU, animate on GenBrasil · how to benchmark video models