Amazon Sponsored Brands video: 1 Mbps and 96 kbps audio minimums
Amazon Sponsored Brands video needs at least 1 Mbps video, 96 kbps audio at 44.1 kHz, and 16:9 only. What to check on an AI-generated clip before upload.

Amazon's Sponsored Brands video spec sets a floor of 1 Mbps for video (4 Mbps or more recommended) and 96 kbps audio at 44.1 kHz, in 16:9 only. Those are encode settings, so you verify them on the exported file, not in the prompt. The numbers come from Amazon's Sponsored Brands video specs, read 2026-10-03.
The full spec on one page
Bitrate and audio are the least visible items on the spec, which is why clips fail on them. The rest is dimensions and length.
| Item | Requirement |
|---|---|
| Length | 6-45 s (20 s or less recommended) |
| Dimensions | 1280x720, 1920x1080 or 3840x2160; 16:9 only |
| File size | 500 MB or less |
| Codec and container | H.264 or H.265; MP4 or MOV |
| Frame rate | 23.976 to 30 |
| Video bitrate | 1 Mbps minimum, 4 Mbps or more recommended |
| Audio | 96 kbps minimum, 44.1 kHz |
| Framing | avoid letterboxing; keep text out of the lower right |
What Sume controls and what it does not
Sume's timeline and trim endpoints do not take codec or bitrate fields. The timeline docs state that the server compiles ffmpeg and that keys such as filtergraph, ffmpeg_args, codec and crf are rejected with a 400. So you cannot dial a bitrate through Sume. You can choose output size and frame rate, and then you must measure the result.
Frame rate is controllable: output.fps accepts 24, 25, 30 or 60, so pick 24, 25 or 30 for this placement. Size is controllable through output.width and output.height (even integers from 256 to 2160), but the table shows 3840x2160 as an accepted Amazon size, and a timeline render is capped at 2160 per side, so a 4K timeline render is not available. Use 1920x1080.
Check the exported file
Run ffprobe on the downloaded MP4 before you upload. It reports video bitrate, audio bitrate and sample rate. If the clip is a short generated shot, it can come in near the 1 Mbps floor, and the audio track may be missing or at a different rate, so confirm the audio stream exists.
A small script makes this repeatable for a batch:
ffprobe -v error -select_streams v:0 -show_entries stream=codec_name,width,height,r_frame_rate,bit_rate -of default=nw=1 clip.mp4
ffprobe -v error -select_streams a:0 -show_entries stream=codec_name,sample_rate,bit_rate -of default=nw=1 clip.mp4If the bitrate is under 1 Mbps
Re-encode locally with your own encoder at 4 Mbps or more for H.264, then re-run the probe. Do the re-encode as the last step, after trims and assembly, so you only do it once. Keep the Sume job id in your batch log so a failed upload can be traced back to the render that produced it.
Sources
Related posts
More in Use cases
- Article 50(2): the provider duty to mark AI outputs machine-readably
Article 50(2) makes providers mark synthetic outputs in machine-readable form. What the wording says, what is exempt, and what to ask a generator vendor.
- Article 50(2) assistive editing: trim and crop vs generate
Article 50(2) exempts assistive editing that does not substantially alter input. Sort your Sume outputs into generated and file-processed before you rely on it.
- Clean a transcript with a 2,000-character instruction, then caption
ElevenLabs speech to text can edit a transcript from a natural-language instruction up to 2,000 characters. A cleanup then captions workflow with Sume captions.
- Article 50: creator duties vs provider duties, side by side
Article 50(4) puts deepfake and certain text disclosure on deployers, while 50(2) puts marking on providers. A two-column split for teams that publish AI video.
Written by Sume