Is it safe to send production footage to the Gemini API? Google's terms, explained for productions
It depends on which key you use. The Gemini API has two regimes. On a free key, Google may use what you send to improve its products and let human reviewers see it. On a paid key, Google says it doesn't, processes your content as your data processor, and keeps it for 55 days for abuse monitoring. A third route, Vertex AI in your own Google Cloud project, adds data residency and a way to switch that logging off. Here is what each means for a production, in Google's words.
Free key: not for client work
- Google uses content "to provide, improve, and develop" its products.
- "human reviewers may read, annotate, and process your API input and output."
- Google's own advice: "Do not submit sensitive, confidential, or personal information to the Unpaid Services."
- In the EEA, Switzerland and the UK, Google's terms allow an app like Shotlore to use "only Paid Services".
Paid key: no training, a processor contract, a 55-day log
- "Google doesn't use your prompts … or responses to improve our products."
- Prompts and responses are processed under Google's "Data Processing Addendum for Products Where Google is a Data Processor", which lists "Gemini API Paid Services". Your company is the controller; Google is the processor. Google is certified under the EU-US Data Privacy Framework and its UK extension, with standard contractual clauses as fallback.
- For abuse monitoring, prompts, context and output are retained "for fifty-five (55) days", can be "accessed for human review only by authorized Google employees" when flagged, are "not used to train or fine-tune any AI/ML models besides those used specifically for policy enforcement", and "may be stored transiently or cached in any country in which Google or its agents maintain facilities."
- A "paid key" means a key in a Google Cloud project with billing linked, whether or not you ever exceed the free quota.
Vertex AI: your own cloud project, your region, logging you can switch off
- "Google won't use your data to train or fine-tune any AI/ML models without your prior permission or instruction."
- Abuse-monitoring prompts are "stored securely for up to 90 days in the same region or multi-region selected by the customer", and "Customers may request for an exception" by form; customers on a Google Cloud Master Agreement "are exempt from prompt logging … by default".
- Regional endpoints keep ML processing in the region you pick (EU multi-region and London are listed for Gemini models; check the current table for the model you use).
- Google's ISO 27001, 27017, 27018 and 42001 certifications and SOC 2 reports cover this service; they are not listed for the plain developer API.
What a production should do
- Make the key in a Google Cloud project your company owns, with billing on; restrict it to the Gemini API; set a spend cap.
- Accept Google's Data Processing Addendum on that account and keep a copy; that is your Article 28 contract.
- Decide per job what goes: previews only (never originals), fewer frames, no positions for sensitive places, no audio where contributors speak without consent.
- Tell the commissioner, in the words their policy uses (see broadcaster AI rules), and keep a record of what was sent.
- For a title with a "no copies anywhere" clause, use on-device tagging or the Vertex route.
How Shotlore fits
Shotlore sends only small preview frames, the brief, time, position and place name, under your own key; never the clips or raw files. It has a free on-device mode that sends nothing, and a Vertex route in development. The full list of what leaves the computer is on the security and confidentiality page; the cost per clip is on the questions page.
Sources: Gemini API Additional Terms, Gemini API abuse monitoring, Google Data Processing Addendum, Vertex AI abuse monitoring. Not legal advice.