← all repositories
pkuliyi2015/sd-webui-stablesr

StableSR for WebUI: Research-Grade Upscaler, Consumer VRAM

A port of the StableSR diffusion upscaler to Automatic1111’s WebUI that trades some fidelity to keep VRAM usage and color drift under control.

sd-webui-stablesr
Not currently ranked — collecting fresh signals.
star history

What it does This extension migrates the StableSR super-resolution model—originally a standalone research project built on Stable Diffusion 2.1—into the Automatic1111 WebUI. It upscales images while trying to preserve facial identity and fine detail, and it layers on a wavelet-based color-fix stage to counter the hue shifts that plague the official implementation.

The interesting bit The author performed aggressive surgery to make it fit: the official VAE decoder’s CFW component was cut because its encoder features balloon past 10 GB for 4K images even in float16, and the custom DDPM sampler from the paper is incompatible with WebUI, so the whole thing runs on Euler a with negative prompts hacked in to recover detail. It is essentially a research demo duct-taped to run on a 6 GB card if you tile aggressively enough.

Key highlights

  • Produces high-detail upscales for realistic, anime, or AIGC images using SD 2.1 checkpoints (512 or 768 variants).
  • Smaller VRAM footprint than ControlNet Tile; paired with Tiled Diffusion & VAE it can push 4K on under 12 GB.
  • Wavelet Color Fix borrows a technique from GIMP and Krita to graft the original image’s low-frequency color information back onto the output, fixing tile-boundary drift.
  • Negative prompts (for example, 3d, cartoon, worst quality) significantly boost detail, an accidental discovery not present in the original paper.
  • Both the code and the resulting images are locked under a strict non-commercial S-Lab License.

Caveats

  • Output fidelity is intentionally lower than the official repo: the missing CFW decoder component and WebUI’s sampler limitations mean you will not match the paper’s examples.
  • The S-Lab License prohibits commercial use of both the extension and the images it generates.

Verdict Worth a look for researchers and hobbyists already living in Automatic1111 who need aggressive upscaling and do not mind trading absolute fidelity for convenience. Everyone else—especially anyone hoping to sell prints or ship a product—should steer clear.

Frequently asked

What is pkuliyi2015/sd-webui-stablesr?
A port of the StableSR diffusion upscaler to Automatic1111’s WebUI that trades some fidelity to keep VRAM usage and color drift under control.
Is sd-webui-stablesr open source?
Yes — pkuliyi2015/sd-webui-stablesr is an open-source project tracked on heatdrop.
What language is sd-webui-stablesr written in?
pkuliyi2015/sd-webui-stablesr is primarily written in Python.
How popular is sd-webui-stablesr?
pkuliyi2015/sd-webui-stablesr has 1.1k stars on GitHub.
Where can I find sd-webui-stablesr?
pkuliyi2015/sd-webui-stablesr is on GitHub at https://github.com/pkuliyi2015/sd-webui-stablesr.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.