v0.3.0-beta

Change Log

• Adding ControlNet support for SD 1.5
• Transition to TensorRT 10
• Simplified install

Warning
This version is NOT backward compatible. Previously exported engines will not work and need to be re-exported.

Full Changelog: v0.2.1...v0.3.0

v0.2.1

What's Changed

• fix typo by @Rudra-Ji in #79
• Fix for #225 Install.py silently blocked by blocking call of nvidia-cudnn-cu11 uninstall due to interactive python shell opening by @andrewtvuong in #226
• Resolve AttributeError on generation without Unet by @AetherMagee in #224
• Fixes crashes cause by lora checkpoint configs missing sd version by @Zn10plays in #251
• Fix AttributeError when accessing profile_idx of NoneType by @shinshin86 in #259

Full Changelog: v0.2.0...v0.2.1

v0.2.0

Features

• Faster LoRA checkpoint conversion.
• Combining multiple LoRAs at runtime.
• Use prompt embedding for LoRA.
• SDXL LoRA support.
• Automatic engine selection.
• Improved HIGHRES.FIX engine selection
• PyTorch fallback.
• Print the correct profile is loaded.

Minor

• Use script callback to handle inference.
• External data format for ONNX
• Deleted dead code and duplications

v0.1.0

Initial Release