How to document Google Veo work for client delivery

Google Veo is the video model behind Google Flow, the AI filmmaking tool Google describes as built for cinematic clips, scenes, and stories. Flow gives Veo three documented ways in: a text prompt, a pair of start and end frames, or ingredient images that keep the same characters and objects across clips. That structure is good news for documentation, because it means every Veo generation has identifiable inputs you can archive. The bad news is that none of those inputs travel with the finished clip.

Google's marking system is the most consistent part of the story. Every output generated in Flow with Veo or Imagen carries an invisible SynthID watermark, and whether a clip also carries a visible watermark depends on your subscription tier. Google states that videos generated on the free, Plus, and Pro tiers carry a visible mark indicating the content was made with Veo, while Ultra tier videos carry one only where local regulations require it. Two clips from the same project can therefore look differently marked, and nothing in the files explains why. Your production record has to carry the tier.

Verification is also documented, with limits worth knowing before you rely on it. Gemini can check whether a file carries a SynthID watermark, and Google says the watermark usually survives re-scaling, re-coloring, compression, and other alterations. But the check only recognizes content made by Google AI tools, a negative result does not rule out other systems, and the documented quota is roughly ten video checks in any rolling 24-hour window. SynthID can tell a client that Google AI made the clip. It cannot tell anyone which prompt, project, or account made it, which is exactly the part a delivery record exists to answer.

What Google Veo stores on its side

• Google states that all outputs generated in Google Flow using Veo and Imagen include invisible SynthID watermarks. Source
• Videos generated on the free, Plus, and Pro tiers carry a visible watermark indicating the content was made with Veo, while Ultra tier videos carry one only where local regulations require it. Source
• Flow documents three generation methods: text-to-video from a described scene, frames-to-video from specified start and end images, and ingredients-to-video that reuses the same images for characters and key objects across clips. Source
• Gemini can verify whether an image, video, or audio file carries a SynthID watermark, with a quota of approximately ten checks per media type within any 24-hour rolling window. Source
• Google states the SynthID watermark will usually still exist even if the content is re-scaled, re-colored, compressed, or altered in other ways. Source
• Gemini can currently only recognize content created by Google AI tools, so a file without a detected watermark could still have been created by other AI systems. Source

What gets lost without your own record

• SynthID tells a checker that Google AI made or edited the content; it carries no prompt, no project name, and no account, so it cannot replace a production record.
• The visible watermark depends on the subscription tier at generation time, so two clips from one project can be marked differently and nothing in the files explains why.
• Flow keeps prompts, ingredient images, and scene edits inside the project, and the downloaded clip carries none of that context.
• Gemini's verification quota of roughly ten video checks per rolling 24-hour window makes it impractical to verify a large delivery in one sitting.
• A negative SynthID check does not establish that footage is free of AI generation, because the check only recognizes Google AI content.

The documentation checklist for Google Veo

1. Keep one Flow project per client job and record the project name in your delivery record, so generations, ingredients, and edits stay traceable to the job.
2. Save the full prompt for every text-to-video generation exactly as entered.
3. For frames-to-video, archive the exact start and end frames you supplied alongside the output clip.
4. For ingredients-to-video, store the ingredient images with the delivery, because consistency across clips depends on reusing exactly those files.
5. Write down the subscription tier active on the generation date, since it determines whether the download carries a visible Veo watermark.
6. Run a SynthID check in Gemini on each final delivered file, and record the result and the check date, planning around the quota of roughly ten video checks per day.

If your delivery falls under the EU AI Act's Article 50 transparency obligations, SynthID gives you a documented, machine-readable marking on the Google side, and the Gemini check gives you a way to demonstrate it on a specific file. What it does not give you is the deployment-side record: what was generated, what was disclosed to whom, and when. Penalties for transparency violations can reach up to €15 million or 3% of worldwide turnover, which is reason enough to keep that record with the project rather than in someone's chat history.

Verify strategically rather than exhaustively. With roughly ten video checks available per rolling day, run the SynthID check on final delivered files, not on every iteration, and store the result with a date next to the clip's hash. That combination of a marking check plus your own prompt and project records covers inclusion, integrity, and time. It still proves nothing about causation, and a good delivery note says so plainly.

Frequently asked questions