Ditto
Ant Group motion-space diffusion model for controllable real-time talking-head synthesis.
Motion-space diffusion with TensorRT models for streaming talking-head output.
Ditto is Ant Group's motion-space diffusion model for controllable, real-time talking-head synthesis. It suits developers building streaming avatar features on their own NVIDIA servers.
What it is
The README ships TensorRT models for Ampere and newer GPUs, an ONNX-to-TensorRT conversion script for other cards, and an online configuration with streaming HuBERT audio features. It names an A100 as the GPU it was run on.
Who it's for
Suits engineers who need low-latency avatar rendering inside their own product. It does not suit non-developers: there is no UI, speech layer or language model.
Key features
Motion-space diffusion
Generates controllable talking-head motion.
Real-time focus
Built for realtime synthesis rather than batch rendering.
TensorRT models
Prebuilt engines for Ampere and newer GPUs.
ONNX conversion
A script converts ONNX to TensorRT for other GPUs.
Online config
A streaming configuration uses streaming audio features.
Apache-2.0
Released under Apache-2.0.
Up and running in 3 steps
Our honest verdict
What's great
- Apache-2.0 licence on the repository.
- Designed for real-time use, which suits live avatars.
- Ships TensorRT engines so you skip some conversion work.
- Backed by a published paper from Ant Group.
Where it falls short
- The README lists CentOS 7.2, an A100 and TensorRT 8.6.1; other setups need conversion work.
- No speech recognition, language model or text-to-speech; you build the conversation layer.
- Checkpoints must be downloaded and arranged by hand.
- Research code with limited documentation.
Pricing & license
Free under Apache-2.0. The README lists no hosted or paid plan; you pay only for your GPU hardware.
Open remote AI-training roles
768 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Is Ditto real time?
It is designed for realtime talking-head synthesis and includes an online configuration for streaming audio, on supported NVIDIA GPUs with TensorRT.
Can I use it commercially?
The repository is Apache-2.0. The README does not state a separate weights licence, so check the model files you download.
What GPU do I need?
The README names an A100 as its reference GPU and provides TensorRT models for Ampere and newer cards.
