Luma Dream Machine API prompt rules: 3 to 5000 characters

Luma's Dream Machine API checks a request before it creates a generation. A prompt shorter than 3 characters or longer than 5000 characters is rejected, a request with image keyframes needs text, and loop is refused for keyframes, extend-reverse and interpolation. Sume returns a 400 invalid_request for a bad body and publishes per-model limits in its catalog.

The pre-submission errors

The Luma errors page splits problems into two groups. The pre-submission group fails the create call. The post-submission group appears later in failure_reason. The first group is the one you can test for in your own code.

Where Sume checks the same things

Sume's /v1/videos route has prompt and model as required fields. Duration, resolution and aspect ratio are checked against the model's own lists, so a value outside supported_durations, supported_resolutions or supported_aspect_ratios is a client error. The error table lists 400 invalid_request for a bad body, query or header, and the job error category validation means fix input.

Frames and references

Two Sume rules map onto Luma's keyframe messages. First, frame_images and input_references are different modes, and when both are present frame_images wins and the request is image-to-video. Second, frame_type must be first_frame or last_frame. The Video Router docs also say to read capabilities from GET /v1/video-router/models instead of assuming one envelope for all models.

• Validate prompt length in your own code before you submit.
• Fetch GET /v1/videos/models and check the three lists for the model you pin.
• Do not send both frame images and references and expect both to apply.
• Treat a 400 as a bug in the request, not as a reason to retry.

What carries over

The Video Router page does not list a Luma model, so this is a port-the-habits comparison and not a drop-in: the habit to keep is checking the request locally, because a rejected request is the cheapest failure you can have.