Baikal Anime Upscaler

Baikal Anime Upscaler

x2 upscaling for anime and illustrations, with Baikal LoopSR v2, LoopSR v1 and SwinFIR v31.

Download models · Example workflow

Examples

Bicubic x2 → Baikal LoopSR v2 x2. Same display size, no additional sharpening. V2: EMA step 30000, loops 4.

Full-resolution inputs/results: Hugging Face examples.

Installation

Clone into ComfyUI/custom_nodes, install requirements with ComfyUI's Python, then restart:

git clone --branch dev https://github.com/SnJake/ComfyUI-Baikal-Anime.git
python -m pip install -r ComfyUI-Baikal-Anime/requirements.txt

For Windows portable, from its root:

python_embeded\python.exe -m pip install -r ComfyUI\custom_nodes\ComfyUI-Baikal-Anime\requirements.txt

Use one checkout; update the existing installation instead of installing a duplicate. Place weights in ComfyUI/models/anime_upscale/ or let the loader download them.

Usage

Category Baikal/Upscale: Baikal Model Loader → Baikal Anime Upscale → Save Image.

| Model | File | |---|---| | Baikal LoopSR v2 x2 | Baikal_LoopSR_x2_v2.safetensors | | Baikal LoopSR v1 x2 | Baikal_LoopSR_x2.safetensors | | Baikal SwinFIR v31 | Baikal_SwinFIR_Anime_x2_v31.safetensors |

LoopSR supports loops 1–4. Default tiling: 256 / overlap 32 / halo 96 in LR pixels; tile: 0 attempts a whole-image forward. Precision: auto/BF16/FP16/none, CPU or CUDA. The loader's UPSCALE_MODEL output also works with the standard image-upscaling node.

Standalone LoopSR:

python tools/infer.py --weights Baikal_LoopSR_x2_v2.safetensors --input input.png --output output.png

LoopSR adapts shared loops, XSA and deep supervision from Looped Diffusion Transformer. Technical notes.

MIT · License