Baikal Swin Anime

--- license: mit pipeline_tag: image-to-image library_name: pytorch tags: [super-resolution, image-restoration, anime, illustration, comfyui, transformer, loopsr, swinfir] --- # Baikal Anime Upscaler **x2 upscaling for anime and illustrations**, using Baikal LoopSR v2, LoopSR v1 or SwinFIR v31. [ComfyUI nodes](https://github.com/SnJake/ComfyUI-Baikal-Anime/tree/dev) · [Example workflow](https://github.com/SnJake/ComfyUI-Baikal-Anime/blob/dev/workflows/Baikal_LoopSR_x2_v2.json) | Model | Download | |---|---| | Baikal LoopSR v2 x2 | [Baikal_LoopSR_x2_v2.safetensors](https://huggingface.co/SnJake/Baikal-Anime-Upscaler/resolve/dev/Baikal_LoopSR_x2_v2.safetensors?download=true) | | Baikal LoopSR v1 x2 | [Baikal_LoopSR_x2.safetensors](Baikal_LoopSR_x2.safetensors) | | Baikal SwinFIR v31 | [Baikal_SwinFIR_Anime_x2_v31.safetensors](Baikal_SwinFIR_Anime_x2_v31.safetensors) | ## Examples **Bicubic x2 → Baikal LoopSR v2 x2.** Same display size, no extra sharpening. V2: EMA step 30000, loops 4.   [Full input](examples/illustration-input.png) · [Full x2 result](examples/illustration-loopsr-x2.png)   [Full input](examples/anime-input.png) · [Full x2 result](examples/anime-loopsr-x2.png) ## ComfyUI Install [ComfyUI-Baikal-Anime](https://github.com/SnJake/ComfyUI-Baikal-Anime/tree/dev), restart, and place weights in **`ComfyUI/models/anime_upscale/`**. The loader can also download them. Category **`Baikal/Upscale`**: **Baikal Model Loader → Baikal Anime Upscale → Save Image**. LoopSR supports **loops 1–4**. Default LR tile/overlap/halo: **256 / 32 / 96**. The custom loader also provides `UPSCALE_MODEL` for the standard upscaling node. LoopSR adapts shared loops, XSA and deep supervision from [Looped Diffusion Transformer](https://arxiv.org/abs/2609.40305). [Technical notes](https://github.com/SnJake/ComfyUI-Baikal-Anime/blob/dev/docs/TRAINING.md). MIT · [License](LICENSE.md)

--- license: mit pipeline_tag: image-to-image library_name: pytorch tags: - computer-vision - image-to-image - image-restoration - image-enhancement - super-resolution - comfyui - pytorch - swinir - swinfir - transformer - anime-upscale --- # SnJake Anime Upscale A **experimental** lightweight upscaler (x2) for anime/illustration images. Designed for clean, sharp results with minimal artifacts. V2 is slightly sharper and removes edge noise artifacts.  ## Examples    # How to use in ComfyUI The model is designed to work with the **SnJake Anime Upscale** ComfyUI node. 1. Install the node from [GITHUB REPO](https://github.com/SnJake/SnJake_Baikal_Swin_Anime). 2. Download the weights from this repository. 3. Place the file(s) into `ComfyUI/models/anime_upscale/`. 4. Select the weights in the node dropdown and run the workflow. # Training Details V1: - Dataset: 40,000 images from Danbooru2024: https://huggingface.co/datasets/deepghs/danbooru2024 - Validation: 600 images - Epochs: 70 V2: - Slightly sharper output, no edge noise artifacts. - Epochs: 20 - Dataset: 49,606 images from Danbooru2024: https://huggingface.co/datasets/deepghs/danbooru2024 - Perceptual backbone: Custom SimSiam pre-trained convnextv2_tiny (Experimental) - Loss schedule: gradual ramp‑in of perceptual/auxiliary losses for stable training. V2.1: - Removed Nearest from resample_methods - Epochs: 30 V2.2: - Epochs: 40 (For now) V3 (SwinFIR): - Epochs: Stage 1 - 20; Stage 2 - 18 - Perceptual backbone: Custom SimSiam pre-trained convnextv2_base (Experimental). - Dataset: ~50,000 images from Danbooru2024: https://huggingface.co/datasets/deepghs/danbooru2024 V3.1 (Current Best - SwinFIR): 🏆 - Epochs: Stage 1 (Charbonnier) - 20; Stage 2 (GAN Fine-tuning) - 16. - Perceptual backbone: Reverted to the robust ImageNet-pretrained convnextv2_base.fcmae_ft_in22k_in1k for superior high-frequency fea...

Source context: 45 downloads · 4 likes · Pipeline image-to-image · Library pytorch · Repo SnJake/Baikal-Anime-Upscaler