Switching model on the Sume Videos page: why 1080p became 768p or 2 s
When you change models on the Videos page, resolution, ratio, duration and sound are clamped to what the new model allows. Examples: 1080p to 768p, 3 s to 2 s.
On the Sume Videos page, changing the model clamps your settings to the new model's list. An unsupported resolution falls back to 720p, then 768p, then the first allowed value. An unsupported duration or ratio falls back to the first allowed value. So 1080p on MiniMax H3 becomes 768p, and 3 s on Wan 3.0 becomes 2 s.
The clamp rules
clampVideoControlsToModel in the Agents videos controls applies four rules every time the model, mode or stored preference changes. It looks up the model's option set, then:
- Resolution: keep it if allowed; else 720p if allowed; else 768p if allowed; else the first allowed value.
- Aspect ratio: keep it if allowed; else the first allowed ratio.
- Duration: keep it if allowed; else the first allowed duration.
- Audio: forced off when the model has no audio (Grok Imagine in this picker).
What it does in practice
The option sets are the ones in VIDEO_MODEL_OPTIONS. The first allowed duration is the smallest one, which is why a switch can drop a clip to its shortest length.
| Switch to | Resolutions | Durations (s) | From 1080p / 3 s you land on |
|---|---|---|---|
| Auto | 720p, 1080p | 3, 4, 5, 6, 8, 10 | 1080p / 3 s |
| Wan 3.0 | 480p, 720p, 1080p | 2, 5, 6, 8, 10, 12, 15, 20, 30 | 1080p / 2 s |
| MiniMax H3 | 480p, 768p | 5, 6, 8, 10, 12, 15 | 768p / 5 s |
| Grok Imagine | 480p, 720p | 5, 6, 8, 10 | 720p / 5 s, audio off |
| Kling 3.0 | 720p, 1080p, 4K | 5, 6, 8, 10, 15 | 1080p / 5 s |
Check before you click Generate
Re-read the resolution and duration chips after every model change, because the clamp is silent. A 10-second plan that moved to the first allowed length costs a fraction of what you expected, and a clip that moved up from 3 s to 5 s costs more.
MiniMax H3 is native 768p. 720p is not a value for that model, which is why the page lands there. The API says the same thing with an error that names 768p.
The page is a subset of the API
Picker durations are presets. The catalog lists every whole second a model accepts (Wan 3.0 takes 2 to 30), so use the API when you need 7 s or 13 s. Read supported_durations from GET /v1/videos/models, or see the reference-limits table for the input side.
Defaults to remember
A fresh page starts at Auto, 16:9, 8 s, 720p. If you pick Grok Imagine from there, the page needs a start-frame image, because the live path for that model requires one (requiresLiveStartFrame). Audio is forced off because the Grok option set declares no audio.
Preferences persist in the browser under a versioned key. A code comment notes the key was bumped so old 1080p preferences reset to 720p; if your resolution looks different after an update, that is why.
Related posts
More in Models
- Wan 3.0 box-selection editing: Sume offers Omni video_url edit instead
Alibaba lists box-selection editing for Wan 3.0. Sume documents no box field for it; its edit route is Gemini Omni Flash 1.1 video_url at $0.125 a second.
- Wan 3.0 is a closed beta and API, not open weights: a Sume call
Alibaba's Wan 3.0 README points to wan.video and QwenCloud, not downloadable weights. Sume lists wan-3.0, 2 to 30 s. The request shape on Sume.
- Wan 3.0 dialogue prompts: says, Lip sync, Voiceover, No dialogue
How to write spoken lines for Wan 3.0 clips: the prompt grammar Alibaba documents, plus what the Sume /v1/videos request and price do with audio.
- 15 cuts in 30 seconds: Wan 3.0 2 s shots vs Seedance 2.5 4 s shots
A fast-cut 30 s ad on Sume: Wan 3.0 allows 2 s shots (15 cuts, $3.75 at 720p) while Seedance 2.5 and Kling 3 stop at 4 s (7 cuts). Prices side by side.
Written by Sume