Locate the person in the material.
Use indexed footage, transcripts, and speaker information to locate suitable appearances and speech from the source library.
CUSTOM AVATARS / WORK WITH OUR TEAMNew scenes.
New scenes.
New performances.
A familiar face.
Turn footage of your talent into a custom avatar production pipeline. Agents find suitable material, prepare a dataset, train a person-specific model, and generate new takes to evaluate.
Demo-led access. Scope the use case and current availability with the team.
FROM THE LIBRARY TO A NEW PERFORMANCEThe person.
The presence.
The possibilities.
Illustrative portrait · not an avatar resultThe presence.
The possibilities.
THE STARTING POINT
Footage of the participating talent and a clear production use case.
THE CUSTOM WORK
Dataset preparation, model training, generation, and evaluation of the takes.
THE DECISION
Judge likeness, performance, consistency, and usefulness for the intended production.
THE PIPELINE IS PART OF THE PRODUCTThe library does
The library does
more of the work.
A useful avatar starts with the material behind it. Starzero’s video understanding helps agents prepare the source dataset.
Build the training material.
Curate examples with useful framing and coverage. Prepare the selected material and captions for the custom training process.
Create a model around the talent.
Train a person-specific model, including custom LoRA work where appropriate to the pipeline. Match the approach to the source and intended outputs.
Evaluate new performances.
Generate takes for the proposed use case. Inspect identity, movement, speech, and how the result holds up across the scenes you actually need.
A DEMO SHOULD ANSWER THE CREATIVE QUESTION
Can this person appear convincingly in the new scene, delivering the performance our production needs?
What to inspect
- Likeness and recognizability
- Speech, expression, and movement
- Consistency across different takes
- How much selection and finishing the result needs
Evaluate the actual output for the use case, rather than inferring quality from a model name.
CONNECTED TO THE PRODUCTIONA new performance.
A new performance.
A place in the edit.
An avatar take is material for a larger creative idea. Shape the surrounding sequence, create graphics and sound, and direct the final piece in the Starzero workspace.
Bring the scenes and output formats you have in mind. They help define the source requirements and what the team should demonstrate.
Explore the creative workspace ↗Before the demo.
What should we bring?
Footage you’re authorized to use, the participating talent, and a few specific scenes or performances you want to create. The team can assess source suitability and discuss what additional material may be needed.
Is this the same as speaker annotation?
No. Speaker annotation names people speaking in existing recordings. The avatar pipeline uses source material to train a custom model for new generated performances. Character identity in existing footage is another distinct capability.
Can we create an avatar immediately in the app?
Start with a conversation and demo for this offering. Current availability, dataset requirements, turnaround, training costs, and production usage are scoped with the team.