kohya-ss/sd-scripts
Source: https://github.com/kohya-ss/sd-scripts
Structured fields
{
"archived": false,
"revision": {
"ref": null,
"pinned": false,
"source_url": "https://github.com/kohya-ss/sd-scripts"
},
"checked_at": "2026-10-17T00:00:00.000Z",
"usefulness": "De facto reference trainer for LoRA/fine-tune/Textual Inversion on SD1.x/2.x, SDXL, SD3/3.5 and FLUX.1 (and newer image models per README); many GUIs and ComfyUI LoRA loaders follow its LoRA conventions. Use for character/style LoRA training recipes.",
"source_type": "repository",
"availability": "available",
"dependencies": [
"PyTorch >= 2.6.0",
"CUDA toolkit matching the PyTorch build",
"diffusers 0.40 and transformers 5.x as of v0.12.0 (per README)"
],
"next_check_at": "2026-10-18T00:00:00.000Z",
"hardware_notes": "No VRAM figures on the README page; NVIDIA CUDA path documented; Windows ARM64 support noted; xformers may be unavailable on some GPU architectures.",
"known_limitations": [
"v0.11.0 was a major internal refactor; v0.10.6 is described as the stable version before it, so older tutorials/configs may not match current main."
],
"upstream_updated_at": "2026-10-17T04:13:20.000Z"
}Evidence claims
- [upstream_claim] README lists training/generation/utility scripts for Stable Diffusion and other image models: LoRA, fine-tuning, Textual Inversion, inpainting-model training, conversion, tagging and LoRA merging. source
- [upstream_claim] Supported models listed include SD 1.x/2.x, SDXL, SD3/SD3.5, FLUX.1, LUMINA, HunyuanImage-2.1 and Anima; newest changelog entry seen is Version 0.12.0 dated 2026-09-24 (dependency update to transformers 5.x and diffusers 0.40, removal of gen_img_diffusers.py). source
- [upstream_claim] Requires PyTorch 2.6.0 or later; README advises PyTorch 2.8.0 with CUDA 12.8/12.9 for RTX 50-series. source
- [upstream_claim] Research review verdict at export: reviewed (as of 2026-10-17). Recorded by the research pipeline, not a hub review outcome; evidence.review stays unreviewed until an independent reviewer records one.
Rights
| component | license | hosting | redistribution | modification | training | Evidence and notes |
|---|---|---|---|---|---|---|
| code | Apache-2.0 | permitted | permitted | permitted | unknown | Repository page states Apache-2.0 overall with some portions under separate terms (Memory Efficient Attention Pytorch MIT, bitsandbytes MIT, BLIP BSD-3-Clause). LICENSE file itself not opened. |
| model_weights | unknown | unknown | unknown | unknown | unknown | Weights of base models loaded by this tool carry their own licenses (for example FLUX.1 [dev] is non-commercial per its model card); LoRAs/checkpoints it produces inherit constraints from the base model and training data, not from the tool's code license. Not assessed here. |