AI Act Article 50 and standard editing: which Sume tools use no model

The Commission's Article 50 FAQ says the obligation to mark AI-generated or manipulated synthetic content does not apply when the AI system performs an assistive function for standard editing. Sume's docs for video trim, video filter, audio detach and Timeline render say they run on worker ffmpeg with no provider inference. Whether a given edit counts as standard editing is the Commission's call in its guidelines, which this post did not read in full; this is not legal advice.

The carve-out is quoted from the FAQ (last update shown 24 July 2026); tool behavior from the Sume docs. Both read 2026-09-30.

What exactly does the FAQ say?

Article 50(2) obliges providers of generative AI systems to mark outputs in a machine-readable way. The FAQ then says that obligation does not apply when the AI system performs an assistive function for standard editing, and that the guidelines give a diverse set of practical examples of what is standard editing and what goes beyond it. The list of examples is in the guidelines, not in the FAQ.

Which Sume tools involve no model?

Sume's own docs describe these as worker ffmpeg only, with no provider inference. That describes how Sume processes the file; it does not decide the legal category.

Which Sume tools are not in that list?

Captions are not: speech-to-captions uses speech-to-text, so do not treat them as ffmpeg-only. Anything that calls a generation model, such as video, image or avatar generation, is outside this post. A trim of an AI-generated clip is still a trim of an AI-generated clip; the mark question for the original generation stays with whoever is the provider of that system.

What should I do with this?

Keep per-file records of the source and the edit, so you can show what was changed; what to log per generated file lists the fields Sume returns. For the platform side of the same question, see YouTube and upscaled AI video. Ask a lawyer where your edit sits against the guidelines.