Grok Imagine video edit on xAI vs the video_url error on Sume

Sume cannot edit a video with a Grok Imagine model: video edit on Sume runs only on gemini-omni-flash-1.1, and any other model that receives a video_url gets a 400 validation error. xAI's own guide says editing an existing clip with a text prompt is done on the grok-imagine-video model, while grok-imagine-video-1.5 is the generation model (xAI video guide, read 2026-10-10).

That split matters in launch week, because Grok Imagine Video 1.5 Lite arrived on top of an existing lineup and people are porting scripts that mix both ids. A script that sends a source clip to a Grok id will fail on Sume before any money is reserved.

What xAI documents for editing

The xAI guide lists three modes for the Grok Imagine family: text-to-video, image-to-video (the still becomes the starting point) and video editing, where you modify an existing video with a text prompt on grok-imagine-video. It also states that grok-imagine-video-1.5 takes up to 14 reference images and up to 3 voice references, and that requests are asynchronous: you start a request, poll with the returned id, and fetch the finished video URL.

The guide does not give an input-length cap for the edit model, so this post makes no claim about one. For prices, xAI's models page shows $0.020 per second for grok-imagine-video-1.5-lite and $0.080 per second for grok-imagine-video-1.5 (read 2026-10-10); it does not give an edit price in the text we read.

What Sume does with a video_url

On Sume the Video Router schema checks the model before it reserves any balance. If the body carries video_url and the model is not gemini-omni-flash-1.1, validation fails with this message:

video_url (edit) is supported only by model gemini-omni-flash-1.1.

The Grok row in Sume's catalog is grok-imagine-video-1.5, and it is an image-to-video row only. It needs image_url, first_frame_url or one reference_image_urls entry, and it rejects end_image_url, last_frame_url, reference_video_urls, reference_audio_urls, bitrate_mode, aspect_ratio and generate_audio. So a Grok edit request fails twice over: wrong model for video_url, and no text-to-video input on that row at all.

The Sume edit route that does exist

Gemini Omni Flash 1.1 edit takes a prompt plus video_url. Resolution is optional and defaults to 720p. Do not send aspect_ratio or duration: the output keeps the source framing, and a duration is at most a reserve-estimate hint. You also cannot combine video_url with image_url, end_image_url or reference_*_urls; the schema answers with Use either video_url (edit) or frame/reference_*_urls fields, not both.

Native audio is always on for that row, and the API rejects generate_audio: false. Sume bills the provider list times 1.25 per output second by resolution; the 720p list rate in the repo is $0.10 per second, so $0.125 billed (check /v1/videos/models for the live rate).

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: edit-demo-001" \
  -d '{"model":"gemini-omni-flash-1.1","prompt":"Make the jacket red. Keep everything else the same.","video_url":"https://example.com/clip.mp4","resolution":"720p","mode":"async"}'

Porting checklist

Before you move an xAI edit job, run through these checks:

• Swap the model id to gemini-omni-flash-1.1 and keep the instruction short and specific.
• Remove aspect_ratio and duration from edit requests.
• Keep the source clip reachable over public HTTPS, as Sume's errors page asks for input media.
• Send an Idempotency-Key so a retry returns the original job.