Luma Scenes: approve keyframes, then render, on Sume
Luma Scenes renders only after you approve each keyframe. On Sume, make and approve a still first, then submit it as first_frame so you pay for video once.

Luma Scenes turns a brief into a storyboard of editable keyframes, lets you approve each scene, and renders only after that. You can copy the order on Sume by hand: generate a still for each scene, approve it, and submit the approved still as a first_frame image-to-video job. The docs I read describe no storyboard editor, so the approval step is yours.
Luma's flow is from its launch post. Sume details are from Video generation and Generation admission, read 2026-10-01.
What does Luma say about paying?
The post names the problem: ten or more generations for one usable spot, and paying full price for a render before you know it is usable. Its answer is that nothing renders before approval, and fixing one scene costs one scene.
How do I send an approved frame?
frame_images specifies first or last frame images for image-to-video; each entry needs a frame_type of first_frame or last_frame. If you also send input_references, frame_images takes precedence and the request is treated as image-to-video.
| Step | Sume action |
|---|---|
| Draft the look | Generate stills with the image API |
| Approve | Your review, outside the API |
| Render | Video job with frame_images, frame_type: first_frame |
| Fix one scene | New job for that scene only |
When does the cost land?
At submit. Sume reserves the estimated amount when the request is accepted, captures it on completion, and releases or refunds it for failed jobs. Stills are cheap to redo before that point, which is why approving them first matters. See first-frame previews for avatar videos for a similar pattern.
How do I keep the look consistent?
Reuse the same style words in every still prompt, and send the previous approved still as a reference when generating the next one. Luma generates its scenes together in one context; Sume jobs are separate, so consistency comes from your prompts and references. More in storyboard to video.
Sources
Related posts
More in Use cases
- Luma uni-1 multi-panel storyboards vs Sume n images
Luma uni-1 can draw a multi-panel storyboard in one image when the prompt describes panels. Sume n returns up to 10 separate images, per-model limits lower.
- Luma Variants placements: which ratios a Sume image can hit
Luma Variants outputs 9:16, 4:5, 1:1, 6:5 and 16:9. Sume's image ratio list has four of those five; 6:5 is not in it, so plan a crop for that slot.
- Size-by-language ad matrix: one Sume job per cell
Luma Variants returns every size and language pair from one master. On Sume each pair is a job, so 2 sizes and 2 languages is 4 submits against plan capacity.
- Mac App Store screenshot size: which of the 4 Sume can make
Mac App Store screenshots are 16:10: 1280x800, 1440x900, 2560x1600 or 2880x1800. Sume's image API covers two; Timeline covers 1440x900.
Written by Sume