Commit Graph

  • 51435f1718 Merge pull request #2303 from kohya-ss/sd3 main Kohya S. 2026-04-02 12:40:48 +09:00
  • fa53f71ec0 fix: improve numerical stability by conditionally using float32 in Anima (#2302) sd3 Kohya S. 2026-04-02 12:36:29 +09:00
  • 1dae34b0af Merge pull request #2298 from kohya-ss/sd3 v0.10.2 dev Kohya S. 2026-03-30 07:53:22 +09:00
  • dd7a666727 Merge pull request #2301 from woct0rdho/resize-lora-logging Kohya S. 2026-03-30 07:50:15 +09:00
  • b2c330407b Print verbose info while extracting woctordho 2025-09-04 15:38:53 +08:00
  • c018765583 Merge pull request #2300 from kohya-ss/doc/re-update-README Kohya S. 2026-03-29 22:10:53 +09:00
  • 3cb9025b4b doc: update change history in README files to include LECO training support for SD/SDXL Kohya S 2026-03-29 22:07:52 +09:00
  • adf4b7b9c0 Merge pull request #2299 from kohya-ss/docs/fix-missing-docs Kohya S. 2026-03-29 22:02:43 +09:00
  • b637c31365 fix: update table of contents and change history in README files for clarity Kohya S 2026-03-29 21:58:38 +09:00
  • 7cbae516c1 Merge pull request #2297 from kohya-ss/fix-anima-fp16-nan-issue Kohya S. 2026-03-29 21:31:19 +09:00
  • 5fb3172baf fix: AdaLN modulation to use float32 for numerical stability in fp16 Kohya S 2026-03-29 21:25:53 +09:00
  • 5cdad10de5 Fix/leco cleanup (#2294) Kohya S. 2026-03-29 20:41:43 +09:00
  • 89b246f3f6 Merge pull request #2296 from kohya-ss/add-svd-lowrank-niter Kohya S. 2026-03-29 20:40:55 +09:00
  • 4be0e94fad Merge pull request #2194 from woct0rdho/rank1 woctordho 2026-03-29 19:35:00 +08:00
  • 0e168dd1eb add --svd_lowrank_niter option to resize_lora.py Kohya S 2026-03-29 20:33:33 +09:00
  • 2723a75f91 Merge pull request #2240 from woct0rdho/svd-lowrank Kohya S. 2026-03-29 19:54:44 +09:00
  • 5f793fb0f4 Log d*lr for ProdigyPlusScheduleFree (#2289) Kohya S. 2026-03-29 18:47:09 +09:00
  • feb38356ea Merge pull request #2291 from kohya-ss/fix-anima-validation-with-text-encoder-output-cache Kohya S. 2026-03-22 22:23:27 +09:00
  • cdb49f9fe7 fix: Anima validation dataset not working with Text Encoder output caching due to caption dropout Kohya S 2026-03-22 22:19:47 +09:00
  • a437949d47 feat: Add support for Safetensors format in caching strategies (WIP) feat-safetensors-cache Kohya S 2026-03-22 21:15:12 +09:00
  • bd19e4c15d Merge branch 'main' into sd3 Kohya S 2026-03-22 21:10:51 +09:00
  • 343c929e39 Log d*lr for ProdigyPlusScheduleFree woctordho 2025-09-24 00:32:00 +08:00
  • b2abe873a5 Merge pull request #2283 from kozistr/deps/pytorch-optimizer Kohya S. 2026-03-19 09:18:06 +09:00
  • 7c159291e9 docs: add skip_image_resolution to config README (#2288) Kohya S. 2026-03-19 09:17:29 +09:00
  • 1cd95b2d8b Add skip_image_resolution to deduplicate multi-resolution dataset (#2273) woctordho 2026-03-19 07:43:39 +08:00
  • 1bd0b0faf1 build(deps): bump pytorch-optimizer into 3.10.0 kozistr 2026-03-02 14:39:48 +09:00
  • d633b51126 Merge branch 'dev' into sd3 Kohya S 2026-02-26 08:22:30 +09:00
  • 1a3ec9ea74 Merge pull request #2280 from kohya-ss/fix-main-wd14-tagger-unbound-local-error Kohya S. 2026-02-26 08:21:42 +09:00
  • e1aedceffa fix: rename character_tags to img_character_tags to fix unboundlocalerror Kohya S 2026-02-26 08:18:45 +09:00
  • 2217704ce1 feat: Support LoKr/LoHa for SDXL and Anima (#2275) Kohya S. 2026-02-23 22:09:00 +09:00
  • f90fa1a89a feat: backward compatibility for SD/SDXL latent cache (#2276) Kohya S. 2026-02-23 21:44:51 +09:00
  • 98a42e4cd6 Merge pull request #2277 from kohya-ss/feat-stability-with-fp16-for-anima Kohya S. 2026-02-23 21:15:49 +09:00
  • 892f8be78f fix: cast input tensor to float32 for improved numerical stability in residual connections Kohya S 2026-02-23 21:12:57 +09:00
  • 50694df3cf Multi-resolution dataset for SD1/SDXL (#2269) woctordho 2026-02-23 14:30:36 +08:00
  • 609d1292f6 Fix the LoRA dropout issue in the Anima model during LoRA training. (#2272) duongve13112002 2026-02-23 13:13:40 +07:00
  • 48d368fa55 Merge pull request #2268 from kohya-ss/sd3 Kohya S. 2026-02-16 08:07:29 +09:00
  • 3265f2edfb Merge pull request #2267 from kohya-ss/fix-github-actions-error Kohya S. 2026-02-16 08:01:20 +09:00
  • ef051427df fix: str is not "no" to str != "no" Kohya S 2026-02-16 07:58:15 +09:00
  • 573a7fa06c Merge pull request #2262 from duongve13112002/fix_lumina Kohya S. 2026-02-16 07:54:49 +09:00
  • ae72efb92b Merge pull request #2264 from kohya-ss/release/v0.10.1 v0.10.1 Kohya S. 2026-02-13 08:34:42 +09:00
  • 449e70b4cf README: Update change history for version 0.10.1 with Anima model support Kohya S 2026-02-13 08:31:22 +09:00
  • b237b8deb3 Merge pull request #2263 from kohya-ss/sd3 Kohya S. 2026-02-13 08:20:44 +09:00
  • 34e7138b6a Add/modify some implementation for anima (#2261) Kohya S. 2026-02-13 08:15:06 +09:00
  • 9144463f7b Merge branch 'dev' into sd3 Kohya S 2026-02-13 08:14:21 +09:00
  • 1640e53392 Fix bug and optimization Lumina training Duoong 2026-02-12 22:52:28 +07:00
  • e21a7736f8 Support Anima model (#2260) duongve13112002 2026-02-08 08:18:55 +07:00
  • 8b5ce3e641 Merge pull request #2255 from cgcalatrava/fix-diffusers-unet-import Kohya S. 2026-01-20 07:50:04 +09:00
  • da07e4c617 Make UNet2DConditionModel import compatible with old and new diffusers versions cgcalatrava 2026-01-19 20:53:00 +01:00
  • 966e9d7f6b Merge pull request #2254 from kohya-ss/dev v0.10.0 Kohya S. 2026-01-19 22:00:25 +09:00
  • 2a2760e702 Merge pull request #1374 from kohya-ss/sd3 Kohya S. 2026-01-19 21:50:22 +09:00
  • b996440c5f Doc update sd3 branch documentation (#2253) Kohya S. 2026-01-19 21:38:46 +09:00
  • 5585347a52 Merge branch 'sd3' into feat-flux-chroma-fp8-scaled feat-flux-chroma-fp8-scaled Kohya S 2026-01-18 16:58:01 +09:00
  • a9af52692a feat: add pyramid noise and noise offset options to generation script (#2252) Kohya S. 2026-01-18 16:56:48 +09:00
  • c6bc632ec6 fix: metadata dataset degradation and make it work (#2186) Kohya S. 2026-01-18 15:17:07 +09:00
  • f7f971f50d Merge pull request #2251 from kohya-ss/fix-pytest-for-lumina Kohya S. 2026-01-18 15:09:47 +09:00
  • c4be615f69 fix(tests): add ip_noise_gamma args for MockArgs in pytest Kohya S 2026-01-18 15:05:57 +09:00
  • e06e063970 Merge pull request #2225 from urlesistiana/sd3_lumina2_ts_fix Kohya S. 2026-01-18 14:39:04 +09:00
  • b9357c662d Merge branch 'sd3' into feat-flux-chroma-fp8-scaled Kohya S 2026-01-18 14:37:12 +09:00
  • 08dc0858b5 Merge pull request #2224 from rockerBOO/block-swap-fp8-scaled Kohya S. 2026-01-18 14:35:32 +09:00
  • 94e3dbebea Merge pull request #2246 from kozistr/deps/pytorch-optimizer Kohya S. 2025-12-21 22:51:32 +09:00
  • 95a65b89a5 build(deps): bump pytorch-optimizer to v3.9.0 kozistr 2025-12-21 15:53:47 +09:00
  • 872124c5e1 Use svd_lowrank for large matrices in resize_lora.py woctordho 2025-11-17 09:20:08 +08:00
  • a5a162044c Merge pull request #2226 from kohya-ss/fix-hunyuan-image-batch-gen-error Kohya S. 2025-10-15 21:57:45 +09:00
  • a33cad714e fix: error on batch generation closes #2209 Kohya S 2025-10-15 21:57:11 +09:00
  • f7fc7ddda2 fix #2201: lumina 2 timesteps handling urlesistiana 2025-10-13 16:08:28 +08:00
  • f9710863ca Set is_swapping_blocks before loading_device, add warning for ignoring fp8_scaled if already fp8 rockerBOO 2025-10-10 15:58:21 -04:00
  • 5e366acda4 Merge pull request #2003 from laolongboy/sd3-dev Kohya S. 2025-10-01 21:03:12 +09:00
  • 650969002f feat: normalize key names when extracting tensors from tuned model feat-extract-flux-lora-from-weights-with-key-prefix Kohya S 2025-09-29 23:18:01 +09:00
  • 5462a6bb24 Merge branch 'dev' into sd3 Kohya S 2025-09-29 21:02:02 +09:00
  • 63711390a0 Merge branch 'main' into dev Kohya S 2025-09-29 20:56:07 +09:00
  • 206adb6438 Merge pull request #2216 from kohya-ss/fix-sdxl-textual-inversion-training-disable-mmap Kohya S. 2025-09-29 20:55:02 +09:00
  • 60bfa97b19 fix: disable_mmap_safetensors not defined in SDXL TI training Kohya S 2025-09-29 20:52:48 +09:00
  • f0c767e0f2 Merge pull request #2213 from kohya-ss/doc-hunyuan-image-training-text-encoder-cpu-note Kohya S. 2025-09-28 18:32:11 +09:00
  • a0c26a0efa docs: enhance text encoder CPU usage instructions for HunyuanImage-2.1 training kohya-ss 2025-09-28 18:21:25 +09:00
  • 67d0621313 Merge pull request #2212 from kohya-ss/fix-hunyuan-image-sample-generation Kohya S. 2025-09-28 18:12:04 +09:00
  • 6a826d21b1 feat: add new parameters for sample image inference configuration Kohya S 2025-09-28 18:06:17 +09:00
  • f6137a7175 feat: add fp8 optimization for FLUX Kohya S 2025-09-25 22:10:11 +09:00
  • 4c197a538b Merge pull request #2207 from kohya-ss/fix-flux-extract-lora-metadata-failed Kohya S. 2025-09-24 21:19:27 +09:00
  • 4b79d73504 fix: update metadata construction to include model_config for flux Kohya S 2025-09-24 21:15:37 +09:00
  • 121853ca2a Merge pull request #2198 from kohya-ss/feat-hunyuan-image-2.1-inference Kohya S. 2025-09-23 19:11:50 +09:00
  • 58df9dffa4 doc: update README with HunyuanImage-2.1 LoRA training details and requirements Kohya S 2025-09-23 18:59:02 +09:00
  • 31f7df3b3a doc: add --network_train_unet_only option for HunyuanImage-2.1 training Kohya S 2025-09-23 18:53:36 +09:00
  • 753c794549 Update hunyuan_image_train_network.py Kohya S. 2025-09-21 13:30:22 +09:00
  • e7b89826c5 Update library/custom_offloading_utils.py Kohya S. 2025-09-21 13:29:58 +09:00
  • 806d535ef1 fix: block-wise scaling is overwritten by per-tensor scaling Kohya S 2025-09-21 13:10:41 +09:00
  • 3876343fad fix: remove print statement for guidance rescale in AdaptiveProjectedGuidance Kohya S 2025-09-21 13:09:38 +09:00
  • 040d976597 feat: add guidance rescale options for Adaptive Projected Guidance in inference Kohya S 2025-09-21 13:03:14 +09:00
  • 9621d9d637 feat: add Adaptive Projected Guidance parameters and noise rescaling Kohya S 2025-09-21 12:34:40 +09:00
  • e7b8e9a778 doc: add --vae_chunk_size option for training and inference Kohya S 2025-09-21 11:13:26 +09:00
  • f41e9e2b58 feat: add vae_chunk_size argument for memory-efficient VAE decoding and processing Kohya S 2025-09-21 11:09:37 +09:00
  • 8f20c37949 feat: add --text_encoder_cpu option to reduce VRAM usage by running text encoders on CPU for training Kohya S 2025-09-20 20:26:20 +09:00
  • b090d15f7d feat: add multi backend attention and related update for HI2.1 models and scripts Kohya S 2025-09-20 19:45:33 +09:00
  • f834b2e0d4 fix: --fp8_vl to work Kohya S 2025-09-18 23:46:18 +09:00
  • f6b4bdc83f feat: block-wise fp8 quantization Kohya S 2025-09-18 21:20:54 +09:00
  • 2ce506e187 fix: fp8 casting not working Kohya S 2025-09-18 21:20:08 +09:00
  • f5b004009e fix: correct tensor indexing in HunyuanVAE2D class for blending and encoding functions Kohya S 2025-09-17 21:54:25 +09:00
  • cbe2a9da45 feat: add conversion script for LoRA models to ComfyUI format with reverse option Kohya S 2025-09-16 21:48:47 +09:00
  • f318ddaeea docs: update HunyuanImage-2.1 training guide with model download instructions and VRAM optimization settings (by Claude) kohya-ss 2025-09-16 21:18:01 +09:00
  • 39458ec0e3 fix: update default values for guidance_scale, image_size, infer_steps, and flow_shift in argument parser kohya-ss 2025-09-16 21:17:21 +09:00
  • 2732be0b29 Merge branch 'feat-hunyuan-image-2.1-inference' of https://github.com/kohya-ss/sd-scripts into feat-hunyuan-image-2.1-inference Kohya S 2025-09-14 20:49:24 +09:00