TikTok Shop wants 3 seconds of dynamic content: test your AI clip

To check an AI clip against TikTok Shop's "at least 3 seconds of dynamic content" line, pull stills at 0, 1, 2 and just under 3 seconds and see whether the picture really changes. Sume's video-frames route does that in one call; deciding whether motion counts as dynamic is still your job.

The rule text comes from TikTok Shop's high-quality video requirements page. The clip-making facts come from the Sume video docs.

The page also asks for clear audio and video and for real-time speech or sign language in LIVEs, and it prohibits AI-generated voices in livestreams. This post is about the post-video rule only.

What does TikTok Shop's page say?

The page says product promotional content must be dynamic, face-visible and interactive, and that post-video content needs "at least 3 seconds of dynamic content." It lists prohibited still-frame content as still images, screenshots, scrolling images, screen recordings, slideshows and animated content. It also says showing your face on camera and showing the product from various angles are expected.

Note that "animated content" is on the still-frame list. The page does not define the term, so I would not assume an AI-generated product clip is exempt or covered, and I would not guess how the platform measures it.

Which AI clips are at risk?

The ones where the first three seconds are nearly a still: a slow zoom on a product photo, an image-to-video clip that barely moves, or an intro card. A clip that opens with a product still, then moves after second three, leaves you with a thin first window. Slideshow-style clips are the clearest case; there is a separate post on slideshows.

How do I sample the first frames?

Import the clip to media.sume.com if it is not there, since video-frames takes only this workspace's media. Give explicit times in at[], one to 24 values; every value must be at least 0 and below the clip's duration, or the worker fails with frame_time_out_of_range. A clip under three seconds fails for 2.9, which is itself a useful signal.

The route always answers 202 and the job is polled at GET /v1/video-frames/{id}, as the docs describe.

Run the check on every clip in a batch, not a sample, since generation varies from run to run and one weak opening is enough to fail the rule. Four stills per clip is cheap compared with a removed video and the account penalties the page says violations can bring.

curl -X POST https://api.sume.com/v1/video-frames \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: tts-first3-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/clip.mp4",
    "at": [0, 1, 2, 2.9],
    "format": "jpeg"
  }'

How do I make a clip that clears the bar?

Ask for movement in the first beat. A prompt like "hands pick up the bottle and turn the label toward the camera" gives the generator something to do from frame one. Pick a duration of at least 4 seconds on models with a 4-second floor, since the Auto route accepts 3 to 10 seconds and seedance-2.5 takes 4 to 30.

Then run the frame check. If the four stills look the same, change the prompt rather than the check.

What can Sume not tell me?

It cannot tell you how TikTok Shop scores a clip. The frames show you what changed; whether that change counts as dynamic is TikTok's call, and the page does not give a threshold. If a clip is borderline, regenerate with more action instead of arguing the edge.