# realesrganX4plusAnime_v1.pt

> Real-ESRGAN x4plus Anime 6B · 4× anime upscaler for ComfyUI

This is RealESRGAN_x4plus_anime_6B, Real-ESRGAN’s official 4× upscaler for anime and illustration, under the filename Civitai gives it. Put it in ComfyUI/models/upscale_models and load it with the built-in Load Upscale Model node. No custom nodes needed.

GitHub saves the file as RealESRGAN_x4plus_anime_6B.pth. It is byte-for-byte the same file, so either name works in ComfyUI.

## Facts

- **Same file as:** RealESRGAN_x4plus_anime_6B.pth
- **Type:** Upscale model (ESRGAN family)
- **Scale:** 4×
- **Architecture:** RRDBNet · 6 blocks · 64 features · 4.47M params
- **Format:** PyTorch .pt (tensors only)
- **Size:** 17.9 MB (17,938,799 bytes)
- **SHA256:** f872d837d3c90ed2e05227bed711af5671a6fd1c9f7d7e91c911a61f155e99da
- **ComfyUI folder:** ComfyUI/models/upscale_models/
- **Loader node:** Load Upscale Model
- **License:** BSD-3-Clause
- **Released:** Aug 31, 2021 · Real-ESRGAN v0.2.2.4

## What is realesrganX4plusAnime_v1.pt?

realesrganX4plusAnime_v1.pt is not a separate model. It is the RealESRGAN_x4plus_anime_6B checkpoint that Xintao Wang and the Real-ESRGAN team released on GitHub in August 2021, re-uploaded to Civitai, which renamed it on download. We hashed both files: same SHA256, same 17,938,799 bytes.

The model takes an image and returns one four times wider and taller. It is tuned for anime, manga and flat-color illustration: it keeps line art crisp and cleans up JPEG blocks and banding instead of inventing photo-style texture. That makes it a poor pick for real photos and a strong default for anime frames, character sheets and stylized renders.

It uses 6 RRDB blocks instead of the 23 in the general RealESRGAN_x4plus model, so it is about a quarter of the size (17.9 MB vs 67.0 MB) and needs far less compute per frame, which adds up on long video sequences.

## Download and install

- [GitHub · xinntao/Real-ESRGAN](https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.2.4/RealESRGAN_x4plus_anime_6B.pth) · saves as `RealESRGAN_x4plus_anime_6B.pth` · Official release. Recommended.
- [Civitai mirror](https://civitai.com/models/147821) · saves as `realesrganX4plusAnime_v1.pt` · Re-upload of the same weights. May ask you to sign in.

```sh
cd ComfyUI/models/upscale_models
wget -O realesrganX4plusAnime_v1.pt \
  https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.2.4/RealESRGAN_x4plus_anime_6B.pth
sha256sum realesrganX4plusAnime_v1.pt
# f872d837d3c90ed2e05227bed711af5671a6fd1c9f7d7e91c911a61f155e99da
```

pt files are Python pickles and can run code when loaded, so check the SHA256 before you use a copy from anywhere other than the official source.

## How to load it in ComfyUI

Two built-in nodes. Load Upscale Model reads the file, and Upscale Image (using Model) applies it to any image or video frames you feed in.

- **Want 2× instead of 4×?:** Add Upscale Image By with scale_by 0.5 after the model upscale.
- **Upscaling video:** Connect the frames from your video loader. Every frame in the batch is upscaled.
- **Large images:** ComfyUI tiles automatically and retries with smaller tiles if VRAM runs out.

## Workflows: Built for anime

These workflows load another anime model in the same folder, or they are anime workflows that already have a Load Upscale Model node. Select realesrganX4plusAnime_v1.pt there.

- [APISR | Anime Image/Video Upscaler](https://www.runcomfy.com/comfyui-workflows/apisr-in-comfyui-anime-image-super-resolution)
- [ComfyUI VNCCS Clone | Consistent Character Generator](https://www.runcomfy.com/comfyui-workflows/comfyui-vnccs-character-workflow-qwen-image-seedvr2)
- [Vid2Vid Part 2 | SDXL Style Transfer](https://www.runcomfy.com/comfyui-workflows/comfyui-vid2vid-part2-sdxl-style-transfer)
- [AnimateDiff + QR Code ControlNet | Visual Effects (VFX)](https://www.runcomfy.com/comfyui-workflows/ai-vfx-using-comfyui-animatediff-workflow-for-visual-effects)
- [SDXL Turbo | Rapid Text to Image](https://www.runcomfy.com/comfyui-workflows/text-to-image-with-sdxl-turbo)
- [ControlNet Tile + 4x UltraSharp | Image/Video Upscaler](https://www.runcomfy.com/comfyui-workflows/comfyui-hi-res-fix-upscaling-workflow-controlnet-tile-4x-ultrasharp)
- [MiniMax H3 Anime Video Generator | Multi-Shot Anime + Synced Sound](https://www.runcomfy.com/comfyui-workflows/minimax-h3-in-comfyui-multi-shot-anime-synced-sound)
- [AnimateDiff + ControlNet + IPAdapter V1 | Flat Anime Style](https://www.runcomfy.com/comfyui-workflows/transform-video-into-flat-anime-style-using-animatediff-controlnet-in-comfyui)

## Same file, other names, and alternatives

- `realesrganX4plusAnime_v1.pt` · This page · Civitai filename · 17.9 MB · Anime and illustration, 4×
- `RealESRGAN_x4plus_anime_6B.pth` · Identical bytes · official GitHub filename · 17.9 MB · Same file. Keep whichever name your workflow expects
- `RealESRGAN_x4plus.pth` · Same authors · general-purpose model · 67.0 MB · Real photos and mixed content
- `4x-AnimeSharp.pth` · Alternative · community anime model · 67.0 MB · Sharper edges on anime, heavier to run
- `4x-UltraSharp.pth` · Alternative · community general model · 67.0 MB · Detailed renders and photos, the common SDXL default

## Common errors and fixes

### Value not in list: model_name: 'realesrganX4plusAnime_v1.pt' not in [...]

Why: The workflow asks for this exact filename, but ComfyUI cannot find it in models/upscale_models. Usually the file is missing, sits in the wrong folder, or was saved under its GitHub name.

Fix: Move the file into ComfyUI/models/upscale_models/, rename it to match the workflow (or pick it again in the dropdown), then press R in ComfyUI to refresh the model lists.

### The file does not show up in Load Checkpoint

Why: It is not a checkpoint. Upscale models are only listed by Load Upscale Model.

Fix: Add a Load Upscale Model node and connect it to Upscale Image (using Model).

### Load Upscale Model fails with UnsupportedModelError or an unpickling error

Why: The download is incomplete, or you saved an HTML login page instead of the file (common with mirror sites).

Fix: Check that the file is exactly 17,938,799 bytes and that its SHA256 matches the value above. Re-download from GitHub if not.

### CUDA out of memory during Upscale Image (using Model)

Why: ComfyUI already tiles the image and retries with smaller tiles, but very large inputs or long video batches can still run out of memory.

Fix: Downscale the input before upscaling, process video in shorter batches, or run the workflow on a larger cloud GPU.

## FAQ

### Is realesrganX4plusAnime_v1.pt the same as RealESRGAN_x4plus_anime_6B.pth?

Yes. Both files have the SHA256 `f872d837d3c90ed2e05227bed711af5671a6fd1c9f7d7e91c911a61f155e99da` and are 17,938,799 bytes. Civitai renamed the official GitHub release on download. Keep one copy and use whichever name your workflow references.

### Where do I put realesrganX4plusAnime_v1.pt in ComfyUI?

In `ComfyUI/models/upscale_models/`. Then press R in ComfyUI (or restart it) so the Load Upscale Model dropdown lists the file.

### Which ComfyUI node loads it? Do I need custom nodes?

No custom nodes. Use the built-in [Load Upscale Model](/comfyui-nodes/ComfyUI/upscale-model-loader) node and connect its output to [Upscale Image (using Model)](/comfyui-nodes/ComfyUI/image-upscale-with-model).

### Do I need to rename .pt to .pth?

Not for ComfyUI, which loads both extensions from upscale_models. The rename tip on Civitai is for the AUTOMATIC1111 web UI, which looks for the official filename in its RealESRGAN folder.

### Is the .pt file safe to load?

.pt files are Python pickles, so only load copies you can verify. The GitHub release and the Civitai copy match the hash above, and the file loads with PyTorch `weights_only=True`, which means it contains tensors only.

### Can I upscale 2× instead of 4×?

The model always outputs 4×. For 2×, add an **Upscale Image By** node with `scale_by` 0.5 after Upscale Image (using Model). This usually looks cleaner than a plain 2× resize.

### Does it work for anime video?

Yes. Upscale Image (using Model) processes every frame in an IMAGE batch, so it slots in after a video loader or a video model. With 6 blocks instead of 23, it needs much less compute per frame than RealESRGAN_x4plus or 4x-UltraSharp.

### Can I use it commercially?

Real-ESRGAN is released under BSD-3-Clause, which allows commercial use as long as you keep the copyright and license notice.

## Sources

- [Real-ESRGAN on GitHub](https://github.com/xinntao/Real-ESRGAN)
- [Real-ESRGAN paper (arXiv 2107.10833)](https://arxiv.org/abs/2107.10833)
- [OpenModelDB entry](https://openmodeldb.info/models/4x-realesrgan-x4plus-anime-6b)
- [Civitai model 147821](https://civitai.com/models/147821)

Canonical page: https://www.runcomfy.com/comfyui-workflows/model-files/realesrganX4plusAnime_v1.pt
