StarSample V2.0

StarSample V2.0 is a 2x image restoration model for cartoon footage. It was trained around My Little Pony: Friendship is Magic frames, but the useful angle is broader: it targets flat-color animation, compression damage, focal blur, GIF-like artifacts, and difficult detail areas that generic photo upscalers often handle poorly.

The practical workflow is simple. Start with a degraded frame, screenshot, or crop from a compressed cartoon source, run it through the 2x model, and compare whether edges, flat fills, and small details look cleaner without turning into smeared AI texture. That makes it more useful as a restoration tool than as a general creative generator.

The release is also reproducible enough to evaluate. The OpenModelDB page includes the model details and examples, and the GitHub release provides public safetensors and ONNX downloads. There is no hosted Space, so this is a local tool workflow rather than a browser demo.

Start from the OpenModelDB page and download the 2x StarSample V2.0 release. Use a local upscaling tool that supports ESRGAN-style models, then run one small crop before processing a full frame or clip. Pick a source with obvious compression blocks, soft lines, or blurred background detail.

Compare the result at normal viewing size and zoomed in. The main thing to check is not whether the image becomes sharper in a generic way. Check whether line art stays stable, flat colors stay clean, and small facial or background details avoid the waxy smudge that restoration models can introduce.

If the 2x model pushes too much scale into the image, also look at the StarSample V2.0 NS variants from the same release. Those are meant for 1x restoration, which may be a better fit when you want cleanup without changing resolution.

This is not a broad image model and the license is noncommercial. It is trained around a specific cartoon source, so test it on your own art style before assuming it will generalize. There is also no live demo; local setup is part of the evaluation.

• Restore compressed cartoon screenshots before editing or archiving.
• Upscale animation frames where photo-focused models create bad texture.
• Clean GIF-like artifacts, blur, and blocky source compression in flat-color art.
• Compare 2x upscaling against 1x cleanup when resolution changes are not the goal.
• Build a repeatable local restoration step for a small batch of animation frames.

Try the demo

View model page