Which AI video model should I use? A checklist by input and output

Choose an AI video model by the constraint you cannot bend, not by a leaderboard: whether you have a first or last image, whether you need reference images, video or audio, how long the clip must be, the resolution and whether sound is built in. Each row below names the Sume catalog ids that list that need. This is a how-to-choose list, not a ranking; run one short clip on two candidates before committing.

Everything here is from the Video generation docs and catalog code, read 2026-09-29. Ask GET /v1/videos/models for the live list.

Which model fits which need?

What should I check first?

• Length: durations are whole seconds and outside the list the request is refused.
• Inputs: input_references are honored only by models that list them; kling-3 lists none.
• Resolution: list values differ per model (grok-imagine-video-1.5 stops at 720p).
• Audio: some models always add sound, grok-imagine-video-1.5 adds none, others follow generate_audio.

How do I read the live list?

One call returns each model's supported_durations, supported_aspect_ratios, supported_input_references and generate_audio.

curl "https://api.sume.com/v1/videos/models" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  | jq '.data[] | {id, supported_durations, supported_input_references, generate_audio}'

What about price?

Sume reserves provider list × 1.25 on submit, plus a 5.5% default agent fee, and the amounts differ a lot per model. Price a clip with your real length and resolution, not a per-second headline.