Walmart's three-flashes-a-second limit and Sume's 24-frame check

Sume has no flash or strobe detector, so it cannot certify a clip against Walmart's flashing rule. What it does have is POST /v1/video-frames, which pulls up to 24 stills at the times you name, enough to look at one second of a suspect moment at about 24 samples a second. Walmart's rule, from its Rich media: Technical requirements PDF (updated Sep 17, 2026, read 2026-10-08), is that videos should not use strobe or flashing lights and that any flashing visual effect must not flash more than three times in one second.

The Walmart rule

On its accessibility page the PDF says not to use visual effects such as strobe or flashing lights, which may cause some viewers to have seizures. It adds that other light effects may not trigger seizures but can be distracting, especially for people with certain cognitive disabilities. If you do use flashing effects, they must not flash more than three times in one second.

AI video models are the risk here. Fast cuts, lens flares, neon signs and camera strobes can appear in a render even when the prompt did not ask for them, so the check belongs after generation.

What the frame tools can do

The frames API takes one clip stored on media.sume.com and either an at array of 1 to 24 times or an fps value. The fps option is capped at 2, so a sampled sweep cannot see a three-per-second pattern. The at array is the useful form: each time is any second value that is at least 0 and below the clip duration, so 24 times 1/24 s apart cover one second. The source must be 300 seconds or shorter, and each call returns at most 24 frames.

Sume bills the job by its Modal compute at list times 1.25 plus the platform fee, and the job does not use an admission seat. The docs give no fixed price, so read the reserved amount from the submit response. For a first look at a whole clip, POST /v1/video-inspect returns 8 mid-bin stills by default.

curl -X POST https://api.sume.com/v1/video-frames \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: walmart-flash-window-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/clip.mp4",
    "at": [4.0, 4.042, 4.083, 4.125, 4.167, 4.208, 4.25, 4.292, 4.333, 4.375, 4.417, 4.458, 4.5, 4.542, 4.583, 4.625, 4.667, 4.708, 4.75, 4.792, 4.833, 4.875, 4.917, 4.958],
    "max_edge": 360
  }'

A review routine

Pick the moments with the most motion, usually cuts and light changes, and request one 24-frame window for each. Look at the stills in order. If brightness swings from frame to frame more than three times inside the second, regenerate that scene or trim it out with video trim.

This is a human check, not a guarantee. Walmart does not publish a measurement method in this PDF, so a clip that looks stable to you is still judged by Walmart's own review.

• Frames come back as durable media.sume.com images; a frame that fails to extract has a null url and does not fail the job.
• Import the clip first with POST /v1/media-imports; the API does not fetch other hosts.
• Scene transitions are the most common place for a brightness swing, so include each cut in a window.