AI b-roll of a real place on YouTube: the realistic-scene label test
YouTube asks for a label when AI footage alters a real place or shows a realistic scene that never happened. Fantasy and animation are exempt. Here is the test.

Yes, if the AI clip looks real. YouTube's help page says to disclose realistic content that alters footage of a real event or place, or generates a realistic scene that did not occur. AI b-roll of a recognizable street or skyline that never looked that way qualifies. A fantasy scene, an obviously animated one, or a minor edit such as a color adjustment does not. The label sits in the player for photorealistic content and in the expanded description for non-photorealistic content, and YouTube says disclosure alone does not limit reach or monetization.
The rule as YouTube words it
YouTube Help: Disclose use of altered or synthetic content, read 2026-10-04, lists four triggers: making a real person appear to say or do something they did not; altering footage of a real event or place; generating a realistic scene that did not occur; and music that is the main focus of the video. It lists as not needing a label: non-realistic content such as fantasy scenes and fully animated AI content, and minor edits such as beauty filters, color or lighting adjustments, special effects filters, caption creation, voice or audio repair, and cloning your own voice. The page also warns that consistent non-disclosure risks manual labeling, content removal, or Partner Program suspension.
| B-roll you generate | Which listed line it touches | Likely label? |
|---|---|---|
| Photoreal skyline of a named city, with buildings that do not exist | Alters footage of a real place | Yes |
| Photoreal crowd at a real event that never took place | Generates a realistic scene that did not occur | Yes |
| Stylized, clearly animated street scene | Non-realistic or animated content | No, but it may show in the description |
| Dragon over a mountain | Fantasy scene | No |
| Real footage you dimmed or cropped | Color or lighting adjustment | No |
Where it goes in the API
The Data API exposes the choice as status.containsSyntheticMedia, a boolean you can set in videos.insert or videos.update to disclose realistic altered or synthetic content (Videos resource, read 2026-10-04). Studio has the same question in its upload flow. Sume does not call either; you set it where you upload.
Because the generated clip's provenance is lost after a re-encode, record at generation time which clips came from a model. The Video Router returns a job for each clip; keep your own table of job to final cut so the answer at upload time is a lookup, not a guess.
A practical gate
Once clips are in a Timeline 1.0 render, the finished MP4 mixes generated and real shots with no per-shot flag. Run a short check: for each slot, list whether the source was generated, and answer the YouTube question for the full file. If any generated slot is photoreal and shows a real place or event, or a person doing something they did not do, disclose. Video inspect can sample up to 24 stills per call so a reviewer sees the cut quickly.
This is a reading of YouTube's own page, not legal advice, and YouTube may still add a label itself.
Sources
Related posts
More in Use cases
- YouTube AI Shorts tools missing in EEA, UK, Switzerland: use an API
YouTube lists several Shorts AI tools as not available in the EEA, UK or Switzerland. If you are there, generate the clip with an API and upload the MP4.
- YouTube Audio-to-Short: 8 s to 3 min, English transcript required
YouTube's Audio-to-Short turns audio-first videos into Shorts of 8 seconds to 3 minutes. It needs an English transcript and desktop. Prepare both with Sume.
- YouTube Dream Track for Shorts: a Lyria soundtrack through an API
Dream Track makes 30-second Lyria soundtracks inside YouTube Shorts. If you need that outside the app, Sume's Music Router returns a file for any video.
- YouTube end screens need a 25-second video: plan the clip length
YouTube end screens need videos 25 seconds or longer. Most Sume video models stop at 15 seconds, so here is how to reach 25 with long models or Timeline.
Written by Sume