AI Upscaler Studio 1.0.3 Portable

Languages: English | File Size: 4.46 GB

Studio-grade upscaling for printable art. Batch-upscale a whole folder at once, up to 10x - to exact print sizes at 300 DPI, entirely on your own NVIDIA GPU. No uploads, no accounts, no slow cloud queue.

Built for people who actually print.
No frills, no fluff - just a fast, local batch image upscaler for Windows that turns low-res art into print-ready files on your own NVIDIA GPU. Six things that matter, when each one matters.

Press-grade speed
FP16, TF32, torch.compile, 768-pixel tiles. A 4K upscale of a 1080p mockup finishes in about eight seconds on an RTX 3060 Ti.

Whole-folder batch
Drop a folder, set the scale, get an archive back. No limits, no babysitting individual files.

1x to 10x scale
A smooth slider, not three preset buttons. Hit exact print dimensions every time.

Stickers stay crisp
The alpha channel runs through the model on its own. No serrated edges on transparent PNGs.

Stays on your machine
Nothing uploads. Or, if you'd rather, use the optional cloud press with your own Replicate key.

Three models, three jobs
Pick by intent, not by filename. Low Resolution for compressed source, CGI / Anime for digital art, High Fidelity for grain-preserving photo work - each backed by a different ESRGAN weight.

Print-size presets, baked DPI
Pick a poster size in inches (5x7 up to 24x36, plus A2 / A1). The output lands at the exact pixel count for the chosen DPI, with the print-DPI tag written into the file so Etsy and Redbubble accept it on first upload.
Release Notes

System Requirements
AI Upscaler Studio needs an NVIDIA GPU to press at the speed it's built for. If you don't have one, please read this before buying.

- Windows 10 or 11
- NVIDIA GPU with 4 GB+ VRAM (GTX 1660 / RTX 2060 or newer) - required for local mode, 64-bit
- 5 GB free disk space for installer and models

Optional
A free Replicate API key for cloud upscaling - no GPU needed, but per-image cost applies

Not supported
AMD, Intel Arc, or Mac - these fall back to CPU and run 20-60x slower

Screen :