Luma Ray3.2 modify: what is confirmed, and Sume's edit route

Luma's Ray3.2 notes list keyframes, 1080p, HDR and reframe but no quotable modify details. Sume's video edit route is gemini-omni-flash-1.1 with video_url.

4 min readSume
All posts

The Ray3.2 announcement on Luma's site, dated June 9, 2026, lists up to 16 keyframes, 1080p, up to 20 seconds, HDR with 16-bit EXR export, facial performance tracking for up to 8 faces, enhanced reframe and API availability. It does not describe a modify mode we can quote, so this page claims nothing about it. Sume's edit route for an existing clip is gemini-omni-flash-1.1 with a video_url.

What does Luma actually list for Ray3.2?

From Luma's announcement, read 2026-09-29: keyframes, resolution, clip length, HDR, face tracking, reframe and an API. If "modify" means changing footage you already have, check Luma's documentation for how that request is shaped, what it costs, and how long a source it takes. None of that is stated here.

Does Sume list a Luma model to modify video with?

Not in the docs read for this page. The video-to-video edit that Sume documents runs on one model, gemini-omni-flash-1.1, through the Video Router: send a prompt describing the edit plus video_url as the source.

What are the rules of Sume's edit request?

The edit shares an endpoint with the model's other modes; Sume routes by the shape of the request. These are the documented rules.

Video edit on gemini-omni-flash-1.1, from the Sume docs read 2026-09-29.
FieldRule
video_urlThe edit source, not a reference
Combined withCannot be combined with image_url, end_image_url or reference_*_urls
resolutionOptional, default 720p
aspect_ratio and durationNot sent on an edit
AudioNative audio is always on

What does the request look like?

This is the documented edit example. It sends an Idempotency-Key header and mode set to async.

curl -X POST https://api.sume.com/v1/video-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: video-router-edit-001" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "Replace the bottle with an apple. Keep everything else the same.",
    "video_url": "https://example.com/clip.mp4",
    "resolution": "720p",
    "mode": "async"
  }'

How long a source can I edit?

The docs do not give an edit source limit in the passages read, and the 3 to 10 second range they state belongs to text, image and reference generation on that model. For a longer piece, cut the source into short clips first and edit each; video trim does the cutting. See also editing a video with a prompt.

What can Sume's edit not do?

It cannot take an aspect ratio or a duration, because the edit keeps the source's own. It cannot be mixed with image or reference inputs, and native audio stays on: generate_audio: false is rejected. It is also a generative edit, not a pixel pass. For a non-generative change, such as dimming or cropping a clip you already hold, Sume's video filter returns a new MP4 at $0.02 per encode job, with no model involved.

What should I check on Luma's side?

Because the announcement text saved for this page is short, confirm these four points in Luma's own documentation before you build on a modify request.

  • The request field that carries the source clip, and whether it takes a URL or an upload.
  • The longest source and output length it accepts.
  • Whether an edit is billed per output second and at which resolution.
  • Whether the result keeps the source's audio.

Sources

Related posts

More in Models

All Models posts

Written by Sume