New performances from your own footage.
Custom avatars are a project we run with you. The agent finds your talent across your library and prepares a dataset, a model is trained on that one person, and you judge new takes for the scenes you need.
You bring
- Footage you're authorized to use
- The participating talent
- The scenes and formats you have in mind
You get back
- A dataset built from your own library
- A model trained on one person
- New takes for your scenes, ready to judge
Demo-led. Availability, turnaround and costs are scoped with the team for each project.
The library does the first half of the work.
A useful avatar starts with the right source material. The same indexing that makes your footage searchable is what finds it.
-
Find
StarZero's index of your footage, transcripts and speakers locates the talent's appearances and speech across the library.
-
Prepare
Clips with useful framing and coverage are selected and captioned as training material.
-
Train
A model is trained on that one person, with custom LoRA work where the source and the outputs call for it.
-
Review
New takes are generated for your scenes, and you judge identity, movement and speech across every one.
Judge it on the takes for your own scenes.
A demo should answer one question: can this person appear convincingly in the new scene, with the performance your production needs?
What to inspect
- Likeness and recognizability
- Speech, expression and movement
- Consistency from take to take
- How much selection and finishing the result needs
What should we bring?
Footage you're authorized to use, the participating talent, and a few specific scenes or performances you want. The team assesses the source and tells you if more material is needed.
Is this the same as speaker identification?
No. Speaker identification names who is speaking in recordings you already have. The avatar pipeline uses that material to train a model for new, generated performances.
Can we make an avatar in the app today?
Start with a conversation and a demo. Availability, dataset requirements, turnaround, training costs and production use are scoped with the team.
Bring your talent and one real scene.
We'll look at the source with you and show what a trained model can do with it.