A Video API That Doesn't Train on Your Data
Most AI video APIs are deliberately vague about whether your uploads become training data. BetterVideo's answer is plain: we never train on your video. Our models are pre-trained and fixed — your footage passes through and comes back enhanced, and nothing about it is kept to improve a model.
Zero training on uploads — by architecture, not just policy.
What "training on your data" actually means
AI video models learn from examples. When a provider trains or fine-tunes on user uploads, your footage becomes part of that example set — its visual patterns get absorbed into the model's parameters. That is fundamentally different from storing a file: once content is baked into a model, it is extremely difficult to isolate or remove. If the video you send is sensitive, "it helped improve our AI" is not a comforting outcome — it means your content now lives, in some diffuse form, inside a system other people use.
Why most APIs stay vague about it
User uploads are the cheapest, most abundant source of fresh training data, so many providers keep the door open with soft language: "to improve our services," "aggregated and anonymized data," "for machine learning purposes," or simply no statement at all. The absence of a clear "we do not train on your uploads" is itself the tell. If a provider wanted you to know they don't, they would say so plainly.
How BetterVideo guarantees it
Our enhancement models ship with fixed, pre-trained weights baked into the processing image. When your video runs, those models are applied as read-only functions: frames in, enhanced frames out. There is no path for your footage to update the models — the processing containers have no write access to the weights, and there is no feedback loop from uploads to training.
This matters because a policy can change overnight; an architecture is much harder to quietly reverse. We built the private behavior to be the default path, not an optional setting that depends on a promise being kept.
How to verify any video API (a quick checklist)
- Search the provider's terms for "train," "machine learning," "improve," and "model."
- Look for a plain, explicit statement that uploads are not used to train or fine-tune models.
- Treat soft phrases like "to improve our services" as a yellow flag.
- Ask directly: "Are uploaded videos used to train, fine-tune, or update your models in any way?"
- Check whether the provider describes its processing architecture — separation of processing and training is the strongest signal.
Frequently Asked Questions
No. Our models are pre-trained with fixed weights and applied to your video as a read-only function. Your footage is never added to a training set, and the models are never updated based on your uploads.
It is architectural, not just a promise. The processing containers that handle your video have no write access to the model weights, and there is no feedback loop from uploads to training. The processing and training systems are separated.
Look for an explicit, unambiguous statement that uploads are not used to train, fine-tune, or update models — phrased plainly, not buried under 'to improve our services.' If a provider won't state it clearly, assume the opposite.
Enhance video without feeding a model
Free sandbox credits. Pay per minute. Your uploads never become training data.
Never trained on. Never sold. Auto-deleted.