BROWSER EXPERIMENTS · MODEL DOWNLOADS ARE OPTIONAL

Images made and edited in your browser

The lab gathers several independent image tools. Models download only when you start a tool, and inference runs in the browser. The StyleGAN2 model is a recovered export from my final-year project; the other tools are separate experiments.

Anime character generated by the recovered StyleGAN2 model from seed 106. — One output from the recovered model. The project presentation's reported FID is not rerun by this browser demo.

Four separate tools

StyleGAN2: generate a 512 × 512 character from the recovered model; GPU and CPU paths are available.
AnimeGANv3: refine a generated image or a photo. The photo stays in the browser.
Real-ESRGAN: upscale an image locally in tiles to limit memory use.
SD-Turbo: a separate text-to-image browser experiment based on Stable Diffusion 2.1, not the final-year StyleGAN project.

Models are substantial downloads: the homepage lists about 145 MB for StyleGAN2 and up to about 2.4 GB for SD-Turbo. The site updater leaves model files alone.

Open StyleGAN2 · Open AnimeGAN · Open the upscaler · Open SD-Turbo

Credits and boundaries

The page identifies model sources, licences, and credits in its terms and credits. The image lab is for experimentation; its browser output does not add new benchmark evidence to the final-year project.