v3.7.9 - HiDream, Lens, Boogu, LongCat-Image

New Models

• HiDream — O1
• Lens — Base, Turbo
• Boogu Image — Base, Edit, Turbo
• LongCat Image — Base, Edit, Turbo

Backend Updates

• Diffusers v0.40.0
• Transformers v5.17.0
• ONNX Runtime v1.29.0
• Stable Diffusion.cpp master-859-7f410a3

Backend Folder Migration

The backend folder layouts have changed in this release.
On startup, Amuse will automatically migrate and rename existing runtime folders to match the new backend folder layout.
No manual migration should be required.

Full Changelog: v3.7.5...v3.7.9

v3.7.5 - Python Model Router Bugfix
This release fixed the python model router bug introduced in v3.7.4

Root Cause:
Debug code left in place during v3.7.4 build

Full Changelog: v3.7.4...v3.7.5

v3.7.3 - Technical Release

This is a mostly technical release with few changes that are directly visible to users.

The main focus was improving the foundations of Amuse, particularly performance and integration. Large tensor transfers now make better use of shared memory, and I’ve added a basic .NET wrapper API for StableDiffusion.cpp.

I’ve also expanded StableDiffusion.cpp support, fixed several video workflows, added new model support, and cleaned up some of the internal architecture.

Not the most exciting release on the surface, but these changes make Amuse more capable and give me a much better foundation for what comes next.

Full Changelog: v3.7.0...v3.7.3

v3.7.0 - StableDiffusion.cpp Backend

Amuse v3.7.0 introduces a new StableDiffusion.cpp backend and adds support for Qwen 3.8 27B.

Unlike Python-based backends, StableDiffusion.cpp does not require Python or a large Python ML environment, making it easier to install, distribute, and run. It brings support for a growing range of image and video generation models through a lightweight, native backend.

Initial Model Support

The initial implementation supports:

• FLUX.1
• FLUX.2 Klein
• Ideogram 4
• Qwen Image
• Qwen Image Edit
• Anima
• Ernie
• Krea 2
• Z-Image
• LTX-2.3

More models will be implemented in upcoming Amuse builds.

Device Support

The StableDiffusion.cpp backend supports:

• Vulkan
• CUDA
• ROCm

This allows supported models to run across a wider range of hardware, including AMD, Intel, and NVIDIA GPUs, without requiring a Python-based runtime.

More Information: StableDiffusion.cpp https://github.com/leejet/stable-diffusion.cpp

Qwen 3.8 27B

Amuse v3.7.0 also adds support for Qwen 3.8 27B, expanding the range of language models available in Amuse.

Full Changelog: v3.6.3...v3.7.0