v25.0.3

• Upgrade Gradio, diffusers and huggingface-hub to latest release to fix issue with ASGI.
• Add a new method to setup and run the GUI. You will find two new script for both Windows (gui-uv.bat) and Linux (gui-uv.sh). With those scripts there is no need to run setup.bat or setup.sh anymore.

v25.0.2

• Force gradio to 5.14.0 or greater so it is updated.

v25.0.1

Fix issue with huggingface download issue

v25.0.0

v25.0.0 Release Notes

Major Features

• Flux.1 and SD3 Model Support: Added full support for Flux.1 and SD3 models, including GUI elements, parameters, LoRA training, and DreamBooth/finetuning capabilities.
• Advanced Learning Rate Options: Introduced custom learning rate scheduler support in the GUI, reworked TE1/TE2 learning rate handling for SDXL DreamBooth, and added lr_warmup_steps override.
• New Optimizers and Schedulers: Added support for AdEMAMix8bit, ScheduleFree, RAdamScheduleFree, ProdigyPlusScheduleFree, and Differential Output Preservation.
• LoRA Enhancements: Added LyCORIS LoRA support for Flux.1, increased maximum Flux-LoRA merge strength to 2, and introduced options for training specific blocks and swapping (blocks_to_swap).

Key Improvements

• GUI Enhancements: Implemented dark mode styles, improved visual sectioning of LoRA parameters, and added auto-detection of model types for safetensors files.
• Performance and Compatibility: Updated to the latest sd-scripts versions multiple times, improved requirements handling across platforms (Windows, Linux, Docker), and ensured compatibility with Python 3.10–3.12.
• Command and Output Improvements: Enhanced "print command" output format and used output model name as the default Weights & Biases (wandb) run name when not specified.

Bug Fixes

• Fixed pip upgrade issues and requirement validation problems.
• Corrected toml configuration issues for noise offset and adaptive_noise_scale.
• Resolved learning rate parameter syntax errors (text_encoder_lr, lr_warmup_steps, max_train_steps).
• Fixed various GUI issues, including Gradio on new installations, prompt.txt location, and SD3/Flux.1 LoRA training parameters (blocks_to_swap, fused_backward_pass).
• Addressed typos in README, GUI files, and logic errors in training scripts.

Detailed Changes

Feature Additions

• Added .webp image extension support for BLIP2 captioning.
• Introduced GUI support for Flux.1 and SD3, including parameters like guidance_scale, fp8_base, split_qkv, and cpu_offload_checkpointing.
• Added training support for T5-XXL with learning rate (t5xxl LR) and max token length fixes.
• Enhanced LoRA training with options like Rank for layers, uniform weighting scheme, and blocks_to_swap.
• Added startup options: --noverify to skip requirements validation and --skip_cache_check for the GUI.

Enhancements

• Updated sd-scripts submodule to the latest version multiple times, incorporating new features like fused_backward_pass for SD3 and pytorch-optimizer support.
• Improved Docker support by mounting the models folder as a volume and updating docker-compose.
• Added quiet output support for Linux setup and tensorboard host configuration via TENSORBOARD_HOST environment variable.
• Updated dependencies (e.g., torch, onnxruntime-gpu for CUDA 12, prodigyopt to 1.1.2) and requirements files for various platforms.

Dependency Updates

• Bumped crate-ci/typos from 1.21.0 to 1.30.0 across multiple updates.
• Updated docker/build-push-action from 5 to 6.
• Updated transformers, wandb, accelerate, IPEX, and ROCm for improved functionality.

Miscellaneous

• Updated README with a fork section and v25.0.0 release details.
• Fixed minor typos in GUI files (e.g., "Regularization") and README.
• Removed unsupported parameters from Flux LoRA network and reverted problematic max_train_steps changes.

For a full list of changes, refer to the commit history.

v24.1.7

• Fix issue with fastapi latest release causing gradio to fail. Updated to latest Gradio module release.

Full Changelog: v24.1.6...v24.1.7

v24.1.6

What's Changed

• Bump crate-ci/typos from 1.21.0 to 1.23.6 by @dependabot in #2677
• Fix issue with missing key. Upgrade Gradio release for security issue. by @bmaltais in #2682

Full Changelog: v24.1.5...v24.1.6

v24.1.5

What's Changed

• Security fix by @bmaltais in #2668

Full Changelog: v24.1.4...v24.1.5

v24.1.4

What's Changed

• Minor fixes by @DevArqSangoi in #2480
• Fix finetuning meta file creation by @bmaltais in #2488

Full Changelog: v24.1.3...v24.1.4

v24.1.3

What's Changed

• Fix issue with tensorboard by @bmaltais in #2472
• Fix creation of output folders by @bmaltais in #2473
• Catch when accelerate command is not found by @bmaltais in #2477

Full Changelog: v24.1.2...v24.1.3

v24.1.2

What's Changed

• fix: vae path validation by @notjedi in #2462
• Fix issue with svd merge int parameters handling by @bmaltais in #2464

New Contributors

• @notjedi made their first contribution in #2462

Full Changelog: v24.1.1...v24.1.2