Veo seed doesn't make output repeatable; Sume rejects seed
Google says Veo's seed only slightly improves determinism. Sume's v1 video models report seed false and reject the field; keep the output file to get a repeat.

Setting a seed on Veo does not give you the same video twice. Google's Veo guide says the seed parameter is available for Veo 3 models and that it "doesn't guarantee determinism, but slightly improves it." On Sume the question does not come up: its video differences table says no v1 model accepts seed, each reports seed: false, and the field is rejected.
Google's statement is from its Gemini API: Generate videos with Veo, read 2026-10-02; Sume's from Video generation and Video Router. Note the wording on Google's page: it names Veo 3 models, and the parameter table lists no seed row for Veo 3.1 or 3.1 Lite.
What does seed buy you on Veo?
Google's text is the whole claim: a slight improvement, no guarantee. Do not build a pipeline that assumes a fixed seed reproduces an approved clip, and do not budget for regenerating one on demand.
What does Sume do with a seed?
The Video generation page's differences table lists seed as accepted by many models on OpenRouter and, for Sume, "no v1 model accepts seed; each reports seed: false and rejects the field." You can see the flag per model in GET /v1/videos/models. The Video Router request schema also has no seed property and does not accept extra properties, and its model_params object is reserved and empty in v1.
| Question | Google (Veo) | Sume (v1 video) |
|---|---|---|
| Accepts a seed | Veo 3 models | No; seed: false per model |
| Guarantees repeat output | No, slightly improves | Not applicable |
| Extra provider knobs | Parameter table per model | model_params empty in v1 |
How do I get the same clip again, then?
Keep the file. Sume results come back as media URLs on the completed job, so store the approved clip and its job id rather than the prompt alone. If you retry after a network failure, send the same Idempotency-Key with the same payload: the docs say a replay returns the original job, not a new generation.
If you need a variation of an approved clip, edit it. Sume's video_url field takes a source clip for Omni with a prompt describing the change, instead of regenerating the whole scene.
What should I record for each approved clip?
Store these with every approved clip.
- The job id and the result URL.
- The full request body.
- The date, since model catalogs change.
Sources
Related posts
More in Models
- Which AI video model gives 1080p on Sume, and which stop at 768p?
Sume lists 1080p for Seedance, Wan 3.0, Kling 3.0 and Auto; MiniMax H3 is native 480p or 768p in the panel. Resolution table by model, with the API check.
- Voxtral TTS: open weights under CC BY-NC versus the paid API
Mistral's Voxtral TTS has open weights under CC BY-NC 4.0 and a paid API. What the licence means for a product, and what Sume offers instead.
- Wan 2.2 weights: Apache 2.0, 5-second 480p/720p; Sume hosts wan-3.0
Wan 2.2 T2V-A14B is Apache 2.0 with open weights and makes 5-second 480P or 720P clips. Sume does not host Wan 2.2; its Wan row is wan-3.0, 2 to 30 seconds.
- Wan 3.0 allows 20,000 prompt characters, MiniMax H3 7,000
Alibaba allows 20,000 characters in a Wan 3.0 prompt; MiniMax caps each H3 text item at 7,000. How to write one prompt that fits both on Sume.
Written by Sume