AMD Lemonade 11.0 Adds Text-to-Speech and 3D Generation, Strengthens Local AI Server
Key Takeaways
- ▸Text-to-speech arrives with voice cloning and voice design capabilities via OpenMOSS backend
- ▸New 3D generation modality ports Trellis.2 image-to-3D pipeline with GUI viewer
- ▸ModelScope support added as alternative model registry alongside Hugging Face
Summary
AMD has released Lemonade 11.0, a major update to its local AI server platform that adds significant new generative capabilities. The release introduces text-to-speech powered by the OpenMOSS backend with voice cloning and design features, and a new 3D generation modality that ports the Trellis.2 image-to-3D pipeline via a dedicated /v1/3d/generations API endpoint.
Beyond new modalities, the update enhances model flexibility by introducing ModelScope as a second remote model registry alongside Hugging Face, adding collection.router support for intelligent model routing, and improving deployment reliability. The FastFlowLM NPU backend now auto-installs on Linux, reducing friction for local deployment, while Docker runtime security is enhanced with unprivileged execution under UID 10001.
The release reflects AMD's focus on making local, private AI inference more accessible and feature-rich, with dozens of bug fixes and improvements across ROCm support, HIP runtime detection, and model management workflows.
- FastFlowLM NPU backend auto-installs on Linux; improved Docker security with unprivileged execution
- Enhanced collection.router model support enables intelligent routing with transparent route reporting
