v22.1.0

• 2023/10/10 (v22.1.0)
  • Remove support for torch 1 to align with kohya_ss sd-scripts code base.
  • Add Intel ARC GPU support with IPEX support on Linux / WSL
    • Users needs to set these manually:
      • Mixed precision to BF16,
      • Attention to SDPA,
      • Optimizer to: AdamW (or any other non 8 bit one).
    • Run setup with: ./setup.sh --use-ipex
    • Run the GUI with: ./gui.sh --use-ipex
  • Merging main branch of sd-scripts:
    • tag_images_by_wd_14_tagger.py now supports Onnx. If you use Onnx, TensorFlow is not required anymore. #864 Thanks to Isotr0py!
      • --onnx option is added. If you use Onnx, specify --onnx option.
      • Please install Onnx and other required packages.
        1. Uninstall TensorFlow.
        2. pip install tensorboard==2.14.1 This is required for the specified version of protobuf.
        3. pip install protobuf==3.20.3 This is required for Onnx.
        4. pip install onnx==1.14.1
        5. pip install onnxruntime-gpu==1.16.0 or pip install onnxruntime==1.16.0
    • --append_tags option is added to tag_images_by_wd_14_tagger.py. This option appends the tags to the existing tags, instead of replacing them. #858 Thanks to a-l-e-x-d-s-9!
    • OFT is now supported.
      • You can use networks.oft for the network module in sdxl_train_network.py. The usage is the same as networks.lora. Some options are not supported.
      • sdxl_gen_img.py also supports OFT as --network_module.
      • OFT only supports SDXL currently. Because current OFT tweaks Q/K/V and O in the transformer, and SD1/2 have extremely fewer transformers than SDXL.
      • The implementation is heavily based on laksjdjf's OFT implementation. Thanks to laksjdjf!
    • Other bug fixes and improvements.

What's Changed

• "Chinese User.bat and useful readme" by @boombbo in #1564
• Update lora_gui.py by @DevArqSangoi in #1572
• Add IPEX by @Disty0 in #1499
• fix: linting, spelling add international English localisation by @sammcj in #1589
• v22.1.0 by @bmaltais in #1594

New Contributors

• @DevArqSangoi made their first contribution in #1572
• @Disty0 made their first contribution in #1499

Full Changelog: v22.0.1...v22.1.0

v22.0.1

Fix iossue with duplicate diffusers module version in requirements.txt

v22.0.0

• 2023/10/01 (v22.0.0)
  • Merging main branch of sd-scripts:

    • SAI Model Spec metadata is now supported partially. hash_sha256 is not supported yet.

      • The main items are set automatically.
      • You can set title, author, description, license and tags with --metadata_xxx options in each training script.
      • Merging scripts also support minimum SAI Model Spec metadata. See the help message for the usage.
      • Metadata editor will be available soon.

    • --v_pred_like_loss ratio option is added. This option adds the loss like v-prediction loss in SDXL training. 0.1 means that the loss is added 10% of the v-prediction loss. The default value is None (disabled).

      • In v-prediction, the loss is higher in the early timesteps (near the noise). This option can be used to increase the loss in the early timesteps.

    • Arbitrary options can be used for Diffusers' schedulers. For example --lr_scheduler_args "lr_end=1e-8".

    • LoRA-FA is added experimentally. Specify --network_module networks.lora_fa option instead of --network_module networks.lora. The trained model can be used as a normal LoRA model.

    • JPEG XL is supported. #786

    • Input perturbation noise is added. See #798 for details.

    • Dataset subset now has caption_prefix and caption_suffix options. The strings are added to the beginning and the end of the captions before shuffling. You can specify the options in .toml.

    • Intel ARC support with IPEX is added. #825

    • Other bug fixes and improvements.

  • New SDXL presets
  • Update wandb module version
  • Add support for Chinese zh-CN localisation. You can use it with .\gui.bat --language=zh-CN
  • Add presets support to Finetuning. You can add your own finetuning user presets under the /presets/finetune/user_presets folder.

What's Changed

• Add Chinese zh-CN localizations by @boombbo in #1542
• v22.0.0 by @bmaltais in #1559

New Contributors

• @boombbo made their first contribution in #1542

Full Changelog: v21.8.10...v22.0.0

v21.8.10

• Minor point upgrade. Mostly adding a new preset.

What's Changed

• v21.8.10 by @bmaltais in #1540

Full Changelog: v21.8.9...v21.8.10

v21.8.9

• Update sd-script to caode as of Sept 3 2023
  • ControlNet-LLLite is added. See documentation for details.
  • JPEG XL is supported. #786
  • Peak memory usage is reduced. #791
  • Input perturbation noise is added. See #798 for details.
  • Dataset subset now has caption_prefix and caption_suffix options. The strings are added to the beginning and the end of the captions before shuffling. You can specify the options in .toml.
  • Other minor changes.
  • Added support fir Chinese locallisation

What's Changed

• localization implement #649 by @yxJoy in #1432
• FIX: lr_scheduler_args set incorrectly in run_cmd by @dioxic in #1454
• Add SmilingWolf/wd-v1-4-moat-tagger-v2 by @zlstringham in #1465
• Update README.md - adding free SDXL training tutorial by @FurkanGozukara in #1473
• v21.8.9 by @bmaltais in #1476

New Contributors

• @yxJoy made their first contribution in #1432
• @dioxic made their first contribution in #1454
• @zlstringham made their first contribution in #1465

Full Changelog: v21.8.8...v21.8.9

v21.8.8

• 2023/08/05 (v21.8.8)
  • Fix issue with aiofiles: #1359
  • Merge sd-scripts updates as of Aug 23 2023
  • Add new blip2 caption processor tool
  • Add dataset preparation tab to appropriate trainers
  • Add GUI support for new block_lr lora network parameter
  • Add support for experimental LoRA-FA network
  • Fix LyCORIS extraction issue with code

What's Changed

• Manual Caption/Tag Tool by @channelcat in #1352
• Fix bitsandbytes 0.40.1 download link for Windows by @Trojaner in #1366
• Added today released 85 minutes 73 chapters tutorial by @FurkanGozukara in #1368
• update of readme file by @FurkanGozukara in #1378
• v21.8.8 by @bmaltais in #1428
• v21.8.8 by @bmaltais in #1429

New Contributors

• @channelcat made their first contribution in #1352

Full Changelog: v21.8.7...v21.8.8

v21.8.7

• 2023/08/05 (v21.8.7)
  • Merge latest sd-scripts updates.
  • Updated layout? This is up for debate... but I think it make things easier to find. Tab instead of endless scrolling...
  • Fix issue with LoRA merge GUI
  • fix: macos arm64 'tensorflow' not found

What's Changed

• I have reworked the GUI to put the basic, advanced and samples config training sections as seperate tabs instead of the long list. Let you know if you like it opr hate it. I think it make for less scrolling...
• fix: macos arm64 'tensorflow' not found by @TitorPs360 in #1329
• Update setup_windows.py by @bmaltais in #1336
• Merge dev2 by @bmaltais in #1357

New Contributors

• @TitorPs360 made their first contribution in #1329

Full Changelog: v21.8.6...v21.8.7

v21.8.6

Notes:

Run setup again to properly install this new release before starting the GUI.

• 2023/08/05 (v21.8.6)
  • Merge latest sd-scripts updates.
  • Allow DB training on SDXL models. Unsupported but appear to work.
  • Fix finetuning latent caching issue when doing SDXL models in fp16
  • Add SDXL merge lora support. You can now merge LoRAs into an SDXL checkpoint.
  • Add SDPA CrossAttention option to trainers.
  • Merge latest kohya_ss sd-scripts code
  • Fix Dreambooth support for SDXL training
  • Update to latest bitsandbytes release. New optional install option for bitsandbytes versions.

What's Changed

• Max norm adjust by @M4X1K02 in #1272
• v21.8.6 by @bmaltais in #1332

New Contributors

• @M4X1K02 made their first contribution in #1272

Full Changelog: V21.8.5...v21.8.6

v21.8.4

• 2023/07/27 (v21.8.4)
  • Relocate LR number of cycles and LR power options
  • Add missing LR number of cycles and LR power to Dreambooth and TI scripts
  • Fix issue with conv_block_dims and conv_block_alphas
  • Fix 0 noise offset issue
  • Implement Stop training button on LoRA and other training tabs
  • Update LyCORIS network release to fix an issue with the LoCon extraction.

What's Changed

• adding first SDXL tutorial link to readme by @FurkanGozukara in #1229
• v21.8.4 by @bmaltais in #1271

Full Changelog: v21.8.3...v21.8.4

V21.8.5

Backrev the LyCORIS module version due to bug reports.