Anima Base Tiled High-Resolution Upscale Workflow

This workflow is designed for Anima Base tiled image upscaling and high-resolution refinement. Its main purpose is to take an existing anime-style image, enlarge it with an anime-focused upscaler, divide the enlarged result into manageable tiles, refine each tile through Anima Base, and then assemble the tiles back into a higher-resolution final image with better detail and stronger visual quality.

Unlike a simple one-click upscale workflow, this setup does not only enlarge the image. It combines RealESRGAN anime upscaling, total-pixel scaling, tile splitting, WD14 tag extraction, prompt rebuilding, Anima Base latent refinement, tiled VAE decoding, and final tile reassembly. This makes the workflow more useful when creators want to improve large anime illustrations without forcing the entire image through one huge generation pass.

The workflow uses anima_baseV10.safetensors as the main UNet model, qwen_3_06b_base.safetensors as the Qwen image CLIP route, and qwen_image_vae.safetensors as the VAE. It also includes anima-turbo-lora-v0.1.safetensors, which helps speed up the Anima Base refinement process. For the first upscale stage, the workflow uses RealESRGAN_x4plus_anime_6B.pth, a practical anime-oriented upscaling model for increasing the source image size before tile-based enhancement.

The first part of the workflow loads the input image and sends it through RealESRGAN upscaling. After that, ImageScaleToTotalPixels pushes the image toward a larger total-pixel target, with the example setting around 9.4 megapixels. This creates a much larger working canvas, but processing the entire image at once would be heavy and unstable. That is why the workflow then moves into tiled processing.

The tile section is one of the key parts of this graph. TTP_Tile_image_size calculates tile dimensions based on the image, using width and height factors with an overlap rate. TTP_Image_Tile_Batch then cuts the image into tile batches, while TTP_Image_Assy later reconstructs the final image using tile positions, original size, grid size, and padding. This tiled method helps preserve large-image structure while allowing each section to be refined more safely.

The workflow also includes WD14Tagger. This node analyzes the image and generates descriptive anime tags. Those tags are then combined with an optional user prompt through JWStringConcat before entering CLIPTextEncode. This means the workflow can automatically recover the image’s visual content and use it as guidance during the refinement stage, instead of relying only on a manual prompt.

The Anima Base refinement stage uses VAEEncode, ClownsharKSampler_Beta, and VAEDecodeTiled. The sampler settings are conservative, with moderate denoise, low CFG, and a fixed / incrementing seed route. This is important because tiled upscale workflows should enhance detail without completely changing the character, composition, pose, outfit, or background. The negative prompt also suppresses common failures such as low quality, blurry details, bad anatomy, bad hands, extra fingers, deformed faces, text, watermark, logo, and JPEG artifacts.

The final result is assembled back into one complete image, and the workflow includes an image comparer so users can check the original input against the reconstructed high-resolution output. This makes it useful for testing whether the upscale actually improves the image without damaging identity or structure.

This workflow is ideal for anime illustration upscaling, Anima Base high-resolution refinement, large image enhancement, tiled redraw, Civitai preview images, RunningHub demos, character illustration repair, and final publish-ready image polishing. If you want to see how RealESRGAN anime upscale, Anima Base, WD14 tag extraction, tiled refinement, and final image reassembly work together, watch the full tutorial from the YouTube link above.

⚙️ Try the Workflow Online

👉 Workflow: https://www.runninghub.cn/post/2055118475793838082?inviteCode=rh-v1111

Open the link above to run the workflow directly online and view the generation results in real time.

If the results meet your expectations, you can also deploy it locally for further customization.

🎁 Fan Benefits: Register now to get 1000 points, plus 100 daily login points — enjoy 4090-level performance and 48 GB of powerful compute!

📺 Bilibili Updates (Mainland China & Asia-Pacific)

If you are in Mainland China or the Asia-Pacific region, you can watch the video below for workflow demos and a detailed creative breakdown.

📺 Bilibili Video: https://www.bilibili.com/video/BV1ci5q65EGQ/

I will continue updating model resources on Quark Drive:

👉 https://pan.quark.cn/s/20c6f6f8d87b

These resources are mainly prepared for local users, making creation and learning more convenient.

⚙️ 在线体验工作流

👉 工作流： https://www.runninghub.cn/post/2055118475793838082?inviteCode=rh-v1111

打开上方链接即可直接运行该工作流，实时查看生成效果。

如果觉得效果理想，你也可以在本地进行自定义部署。

🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！

📺 Bilibili 更新（中国大陆及南亚太地区）

如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。

📺 B站视频： https://www.bilibili.com/video/BV1ci5q65EGQ/

我会在 夸克网盘 持续更新模型资源：

👉 https://pan.quark.cn/s/20c6f6f8d87b

这些资源主要面向本地用户，方便进行创作与学习。