Does color grading or blur need a YouTube AI label? Sume video filter

YouTube lists beauty filters, color and lighting adjustment, background blur and upscaling as minor edits needing no AI label. One $0.02 filter job does most.

4 min readSume
All posts

No. YouTube's disclosure page says minor edits do not need an AI disclosure, and it defines minor edits as ones that are primarily aesthetic and do not alter the content in a way that could mislead the viewer about what actually happened. Its list includes beauty filters, color adjustment or lighting filters, special effects filters such as background blur or vintage effects, caption creation, and video sharpening, upscaling or repair.

Sume's video filter does this kind of pixel pass on one clip for $0.02 per encode job, and it returns a new MP4 without touching the source.

What the page exempts

Keep the intent test in view: aesthetic, not misleading. The table maps YouTube's minor-edit list to what a filter job can do.

Minor edits YouTube lists vs Sume video filter (read 2026-10-04)
YouTube minor editSume video filter route
Color adjustment or lighting filterseq, colorbalance, hue, or the dim op
Background blur or vintage effectsgblur, vignette, noise in filtergraph
Sharpeningunsharp or cas in filtergraph
Beauty filtersNo named beauty feature; use tone and blur filters in filtergraph
Caption creationVideo captions, $0.20 per job

Check, then run

The filter API has a free check endpoint, POST /v1/video-filter/check, that validates the program and the source without creating a job. Then the encode call needs an Idempotency-Key. This program warms the picture slightly and adds a light vignette:

curl -X POST https://api.sume.com/v1/video-filter \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: grade-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/talk.mp4",
    "filtergraph": "eq=contrast=1.06:saturation=1.1,vignette=PI/10"
  }'

Limits that matter

  • The source must be 300 seconds or shorter; longer sources fail with output_duration_exceeded.
  • The filtergraph is capped at 2048 characters and 32 filters, and only allowlisted pixel filters are accepted.
  • The default mode is async; poll GET /v1/jobs/:id/result or send mode: sync to wait up to 30 seconds.

Why a separate pass is worth it

Because the filter job never rewrites the source, you can keep the untouched clip as your reference and publish the graded copy. If a viewer, a brand partner or a platform ever asks what was changed, the answer is a short, boring list: a tone curve, a vignette, a crop. That is the same shape of answer YouTube's page wants, where the edit is aesthetic and the facts on screen are unchanged.

Run the free check first when you are experimenting. An unknown filter name fails up front with the allowlist in the error, before any compute is reserved, so a typo costs nothing.

Caveat

A grade is only minor while it stays aesthetic. Relighting a scene so it shows a different time of day, or blurring a sign that changes what happened, can move you into 'alters footage of a real event' territory. If YouTube's systems still label the upload, you can usually change the answer under Attributes, except for content made with YouTube's tools, with C2PA metadata, or labeled after manual review.

Sources

Related posts

More in Media tools

All Media tools posts

Written by Sume