Training

Finetune a model for your use-case

Fine-tune and deploy multimodal models on data that matches your use-case. Work with the team behind Dia and Narvatar.

Fine-tuning shaped around your use-case

Start with your data and target behavior. Together, we select the model and training approach that fit.

Tune behavior, language, identity, and output quality around your data and use case.

Explore custom training when fine-tuning an existing model cannot meet your requirements.

Transfer capability into a smaller model built for tighter latency and serving budgets.

Train components that combine speech, visual identity, motion, and interactive generation.

Our own models span dialogue generation and human-centric video.

Text to SpeechOpen Source

Our dialogue TTS model series, with 2M+ downloads and 20K+ GitHub stars. Built to generate speech that carries a conversation.

AvatarIn-house

Our human-centric avatar model under 5B, with streaming and non-streaming variants. Optimized for realtime performance and low serving costs.

Checkpoints and associated assets, with the handoff scope agreed at the start.

Results on agreed test data, compared against the starting model and your target behavior.

A deployment configuration for our optimized stack, with latency, capacity, and cost targets.

Share your use-case, model, and performance targets. Work with our engineers on the next step.