AI Act closed-loop film previs: where preview ends, output begins

The EU FAQ puts closed-loop film production outside marking scope unless the output is final. Avatar preview stills are never captioned; the final MP4 can be.

4 min readSume
All posts

The Commission's Article 50 FAQ lists, among outputs outside the marking obligation, outputs used only in closed-loop industrial and product development environments, for example film production, unless they are the final output. So internal previs that never leaves the production loop sits differently from a file you publish. This is a plain reading of the FAQ, not legal advice.

The FAQ text is from the Commission's Article 50 FAQ, read 2026-10-01. Below is how Sume's docs separate previews from final output.

What does the FAQ actually say about film production?

It says the Guidelines on Transparency of AI-Generated Content clarify that some outputs fall outside scope, and one entry is outputs that are only used in closed loop industrial and product development environments, for example for film production, unless they are the final output.

The word only and the clause unless they are the final output carry the weight. A frame reviewed internally and discarded is one thing; the same frame in the delivered cut is another. The FAQ does not define a pipeline stage for you.

Where does Sume separate preview from final?

In Avatar videos, optional captions burns styles into the clean final MP4 after generation, and preview stills are never captioned. So burned-in text only lands on the final file, not on the stills you review.

That is a statement about captions, not about legal status. If your label is a burned-in caption, you add it at the final render and your preview stills stay clean.

Preview versus final in the Sume docs, read 2026-10-01
ArtifactWhat the docs say
Avatar preview stillNever captioned
Final avatar MP4captions burns styles in after generation
Frames from a clipvideo_frames extracts stills at times you name, per Video frames; unbilled
Clip inspectionvideo_inspect returns probe facts, stills and optional STT

How do I review a clip without making a new deliverable?

Video frames takes one workspace clip and returns durable image artifacts at the times you name, leaving the source untouched. Video inspect reads one media.sume.com clip and returns probe facts, stills and optional transcription; it never produces an MP4.

Both are review tools. Neither changes the final file you would publish.

What should I decide before shipping?

Decide, per asset, whether it is only in the closed production loop or will be seen by the public. Anything that becomes the final output follows the normal labelling path; see what creators must do under Article 50. Keep the label step on the final render and ask your own adviser about edge cases.

Sources

Related posts

More in Use cases

All Use cases posts

Written by Sume