Which AI video model gives 1080p on Sume, and which stop at 768p?

On Sume, Seedance 2 and 2.5, Wan 3.0, Kling 3.0 and Auto list 1080p; MiniMax H3 is native 480p or 768p, with 768p treated as first-class rather than as 720p. MiniMax H3 Max reaches 1080p on the API as a latent refinement from native 768p, while the Videos panel lists it at 480p and 768p only.

Resolution is the control most likely to change your bill, so read the model's own list before you choose. Sume's video generation docs publish the vocabulary: 480p, 720p, 768p, 1080p, 1K, 2K and 4K, and each model advertises the subset it accepts.

Which models list which resolution?

The table merges the API docs and the panel. Where they differ, the cell says so.

What is 768p and why is it not 720p?

MiniMax H3 renders natively at 768p, and Sume says 768p is first-class, not 720p. The panel treats this carefully: when you switch to a model without your resolution, it prefers 720p and then 768p, so a switch to H3 from a 1080p model lands on 768p.

Alibaba's Wan3.0 announcement lists 480P, 720P and 1080P output and prices them at $0.05, $0.10 and $0.20 per second on Alibaba Cloud Model Studio. Sume bills its own way, at provider list times 1.25 reserved on submit, so those are Alibaba's numbers, not Sume's price; read usage.cost on the job.

What about 4K?

Kling's comparison page says Kling 3.0 and 4.0 both support 4K. In Sume's panel, though, a 4K choice is mapped down to 1080p on submit, covered in Kling 3.0 4K in the Videos panel. On the API, Gemini Omni Flash 1.1 is the row whose docs list 4K.

How do I choose a resolution?

Draft low, finish high. A 480p pass tells you whether the motion works before you pay for 1080p, and the existing posts on a Seedance 2.5 draft at 480p then 1080p and a Wan 3.0 480p draft then 1080p final work through it. To check any model from the API:

curl "https://api.sume.com/v1/videos/models" \
  -H "Authorization: Bearer $SUME_API_KEY"
# read supported_resolutions for the model you will pin

What should I check before a batch?

Resolution lists are the part of a model entry most likely to change between catalog versions, so read them from GET /v1/videos/models the day you build a batch rather than from a blog post, including this one.

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?

Sume does not upscale for free: 2K and 4K upscales are priced if you request them. And it does not return a resolution the model does not list, so a request for an unlisted value fails validation instead of falling back.