Who owns an AI video from open weights? Output clauses compared
H3 says you own outputs, Hunyuan, LTX and Wan 2.2 claim no rights in them. What each clause adds or withholds, read from the vendor files.
Under all four open video models, the vendor says it does not claim ownership of what you generate. MiniMax H3's FAQ says you own the outputs you create and the derivative models you train. The Tencent Hunyuan license and the LTX-2.x license each say the licensor claims no rights in the Output. The Wan 2.2 README says it claims no rights over your generated contents. What differs is what each adds after that sentence, and those additions decide what you can do with a clip. Read 2026-10-08.
The four clauses
The ownership sentence is the easy part. The conditions are in the rest of each file.
| Model | Ownership wording | Condition that follows |
|---|---|---|
| MiniMax H3 | You own the outputs you create and the derivative models you train | Outputs and weights may not be used to train or improve other AI models; follow the Acceptable Use Policy |
| HunyuanVideo | Section 6(d): Tencent claims no rights in Outputs | Section 5(c) bars use or display of Output outside the Territory; 5(b) bars using Output to improve other AI models |
| LTX-2.5 | Section 5: Licensor claims no rights in the Output | Use of Output cannot contravene the Agreement; provenance and disclosure duties in Section 6 |
| Wan 2.2 | README: no rights claimed over your generated contents | Use must comply with the Apache 2.0 license and not be harmful or unlawful, per the README |
Reading the conditions
Ownership without freedom to reuse is a narrow gift. Three of the four files restrict using clips to train other models, so a studio that wants to build its own video model from open-weights output has a clean path only through Wan 2.2's license text, and should still read the README's harm and misinformation language.
Geography matters for Hunyuan, as the Section 5(c) post on this site explains. Provenance matters for LTX-2.x: the August 2026 text asks you not to remove watermarking or disclosure features.
What this says about a hosted job
These clauses are about weights you run. A clip from a hosted API is governed by that service's terms. Sume's docs say completed jobs return Sume-hosted video artifacts at media.sume.com URLs, and that artifact URLs are opaque. They do not state an output-ownership clause in the pages cited here, so read Sume's terms of service for that, and read the provider terms for the underlying model.
If you need a clip you can use as training data for your own model, check the terms for the specific route before you generate, not after.
A checklist for the legal review
Bring these four questions to counsel:
- May we train or fine-tune another model on these clips?
- Where may the clips be shown? (Hunyuan names territories.)
- Must we keep watermarks or disclosures on the clips? (LTX-2.x says yes.)
- Must our interface say which model made the clip? (H3 asks for a visible 'MiniMax H3' label in commercial products.)
Sources
Related posts
More in Models
- 1:1 square AI video on Sume: eight rows take it, Omni does not
Eight Sume video rows list 1:1 as an aspect ratio. Gemini Omni Flash 1.1 allows only 16:9 and 9:16. Cheapest square clip: 5 s at 480p is $0.31.
- Edit a video with Gemini Omni Flash on Sume: fields allowed or refused
Omni Flash 1.1 edit mode on Sume takes one video_url plus a prompt, defaults to 720p, and refuses aspect_ratio, duration and other media fields.
- Omni Flash 1.1 is billed in tokens: 5,792 per 720p second, in dollars
Google bills Gemini Omni Flash 1.1 at 5,792 output tokens per 720p second and $17.50 per million. The math to dollars, and the Sume per-second rate.
- Omni Flash 1.1 takes no audio reference: where a Short's voice goes
Gemini Omni Flash 1.1 accepts image and video references but not audio_url. For a voiced Short use native audio or a Timeline spine; 60 s is $7.60 at 720p.
Written by Sume