A GPU when you need one.
Your own computers stay the default. When a job needs more than your server has, Frameleaf Cloud can run it on a rented GPU, only when you ask and only after you've seen the cost.
Descriptions
Richer captions and tags from larger vision models, in your language. Describe a whole library in one go.
Enhance
Upscale photos 2× or 4×, with an optional pass that restores soft or old faces.
Video restoration
Faithful or Creative, up to 4K. Always preview five seconds first.
Transcription
Speech to captions in eight languages, up to four hours per video.
Voice-over
Natural text-to-speech narration for Studio projects, in English.
Smooth motion
Frame interpolation at 2× or 4×, up to 120 fps and 4K.
You choose where every job runs.
A model slider for each kind of job goes from lighter to heavier models. Nothing ever falls back to the cloud silently, and if the cloud is unavailable, nothing is sent anywhere else.
- White runs on your server's processor.
- Green fits the GPU you already have.
- Blue runs only on Frameleaf Cloud, and always asks first.

Private by design.
Never the original
Jobs get a copy with location and camera metadata removed. Video audio is stripped before upload.
Zero retention
Files are deleted when the job finishes, within 24 hours at most, and never used to train models.
Never in the cloud
Face recognition, smart search, text recognition and Studio exports always stay on your network.
Consent you control
An administrator accepts the processing terms, decides who may spend, and can withdraw consent at any time.
Pay for the seconds you use.
Cloud jobs are paid from a prepaid AI credit balance. There's no subscription quota and no expiring bundle.
See it first
Every job shows its model and a likely-to-most cost range before you confirm.
Hold, then settle
A hold covers the estimate; you pay only what the job used and the rest is released immediately.
Your limits
A daily spending cap, per-person monthly limits, and optional auto top-up with a monthly ceiling.
Fair on failure
If Frameleaf or the GPU provider fails, you're not charged.