Why did my video settings change when I switched model in Sume?

When you switch the model in Sume's Agents Videos panel, the panel clamps your settings to what the new model lists: a resolution it does not offer becomes 720p (or 768p if there is no 720p), and an aspect ratio or duration it does not offer becomes the first value on that model's list. Audio is switched off only for a model with no audio support.

The reset is a panel behavior, not an API one: the docs describe each model advertising the values it accepts, and it is worth knowing before you submit a 12-second square clip that quietly became 5 seconds at 16:9.

What exactly does the panel reset?

The clamp runs on the model change and on the mode change between Create and Edit. For each of the three controls it keeps your value when the new model lists it. Resolution has its own order: keep it if listed, otherwise 720p if listed, otherwise 768p if listed, otherwise the first value. Aspect ratio and duration fall back to the first listed value of the new model.

Audio is the odd one: it is forced off only when the model has no audio support, which in the panel means Grok Imagine. For Auto and MiniMax H3 the audio toggle is hidden because audio is always on, per the video docs: the current Auto family always generates audio.

What do the fallbacks look like in practice?

These are worked from the panel's per-model lists, not from a test run, so treat the first listed value as the rule.

Why 3 seconds when I switched to Auto?

Auto lists 3, 4, 5, 6, 8 and 10 seconds, and the first value is 3, so a 15-second request from Kling 3.0 lands on 3, not on the nearest valid length of 10. The same logic gives 5 seconds on Grok Imagine and Kling 3.0. The panel does not pick the closest value; it picks the first.

The Sume docs say Auto create controls default to 720p and 8 seconds with 3 to 10 second clips at 16:9 or 9:16, so Auto's ceiling is 10 seconds even though Wan 3.0 runs to 30 and Kling 3.0 to 15 in the same panel. Kling's own comparison page says Kling 3.0 supports 3 to 15 seconds natively; the panel lists a subset.

How do I avoid surprises?

Choose the model first, then set resolution, ratio and duration. If you must change models mid-draft, re-read the three chips before you press generate. On the API, the clamp is a panel behavior, and GET /v1/videos/models tells you the allowed supported_resolutions, supported_aspect_ratios and supported_durations up front.

• Pick the model before the settings.
• Re-check the ratio chip after any switch, because the fallback is usually 16:9.
• Do not assume a long duration survives a switch to Auto, Grok Imagine or Kling 3.0.
• Keep a copy of the settings you wanted in your prompt notes, not in the panel.

What should I check before a batch?

After any model switch, the three chips and the audio toggle are the whole state, so reading them once is cheap insurance.

Submit one short clip first, open the finished job, and compare what you asked for with what came back. Use an Idempotency-Key on each attempt, because a replay with the same key returns the original job instead of creating and billing a second one. Only then queue the rest.

Sume reserves provider list times 1.25 when a job is submitted, and the poll response's usage.cost is the billable amount. Treat that field, not a panel estimate, as the number to budget with.

What does Sume not do?

The panel code clamps to the first listed value and keeps no history of your earlier values, so after a switch back the clamped value is what you see. Confirm the chips after every switch.