v0.6.0 - Better Model Storage

What’s New

Model De-duplication
Many models share the same TextEncoders and AutoEncoders (VAE). Previously, these were duplicated for every model you downloaded. Now, Diffuse identifies these common files and only downloads them once, sharing them across your entire library.

• New Directory Structure: Common files will now be stored in a new shared location.
• Re-downloading Shared Files: Upon first run, Diffuse will download these shared encoders into the new location.
• Existing Files: Files already downloaded in their original model folders will remain unchanged (you are free to delete if necessary).
• Future Savings: Once these shared components are in place, future model downloads will be much smaller and faster.

Mixed Device & Multi-GPU Handling

• PCI Bus Detection: Diffuse now uses PCI Bus IDs instead of relying on the Windows display order.
• Mixed Environments: Improved stability for systems with multiple GPUs or mixed device types (e.g., ROCM + NVIDIA GPU).

New Pipelines

• SkyReelsV2

Bug Fixes & Stability Improvements

• Various fixes and internal cleanups to improve reliability during model loading and inference.

Full Changelog: v0.5.5...v0.6.0

---

Installation & Upgrading

1. Installer Version

Best for a standard Windows setup with automatic shortcuts.

• New Install & Upgrade:
  1. Download and run Diffuse_vX.X.X.exe.
  2. Follow the on-screen instructions.

---

2. Standalone Version

Best for custom drive locations.

New Install:

1. Download and extract Diffuse_vX.X.X.zip to your preferred folder.
2. Run Diffuse.exe.

Note: A fast SSD with plenty of free space is highly recommended, as model downloads can be large.

To Upgrade:

1. Download the new version.
2. Extract the contents.
3. Copy and overwrite all files into your existing Diffuse folder.

Note: Overwriting ensures your existing configurations and downloaded models remain intact while updating the core app.

---

First-run Notice

On the first launch or when loading a model for the first time, setup may take several minutes while components are initialized. This is expected behavior.

Diffuse will automatically handle the following:

• Portable Python: Installs an isolated, portable Python runtime.
• Environment: Creates the required virtual environment.
• Models: Downloads the selected model files from Hugging Face.

No manual Python setup or coding knowledge is required.

---

Device Support

Support for CUDA based devices.
Experimental support for ROCM based devices.

⚠️ ROCM Compatibility Note:

Experimental support is active, but a known issue with VAE tiling on AMD hardware causes unrecoverable system freezes for LTX-2, WAN, and Qwen. We recommend avoiding these models on ROCm until a driver-level fix is available.