The Studio

This app needs your graphics card to run. If you're seeing this, your browser or computer is blocking access. How to fix it: 1. Update your browser: Ensure you're using the latest version of Chrome, Edge, or Firefox. 2. Turn on "Hardware Acceleration": 1. Go to your browser Settings. 2. Search for "System" or "Graphics." 3. Toggle "Use graphics acceleration when available" to ON. 3. Restart: Close all browser windows and try again. Still stuck? Search Google for: "How to enable WebGPU in [Your Browser Name]" Note for Linux users: You may need to enable the #enable-unsafe-webgpu flag in your browser settings (chrome://flags).

This tool always upscales by 2×, doubling the width and height of your image or video. In the side-by-side preview both versions are scaled to the same display size, so the sharpness gain may look subtle — open the downloaded file at full resolution or in fullscreen to see the real improvement.

Need 4× or 8×? Download your result, re-upload it, and run another pass. Each extra pass doubles the resolution again — 2×+2×=4×, then one more gives you 8×.

The AI model processes images using the GPU, which works on RGB color channels only and does not carry transparency through the super-resolution pipeline. As a result, any transparent areas in your PNG are filled with black in the output.

This is a current limitation of the tool. If your image has a transparent background, the workaround is to place it on a solid colored background before upscaling, then remove that color afterward in an image editor.

upscale8 uses WebGPU to run a neural super-resolution model directly on your device. Your media stays local while the model predicts higher-detail pixels from low-resolution input patches.

For images, the app decodes the file, runs the model on the selected profile, then exports the upscaled output using a matching format. For video, frames are processed sequentially and encoded again in-browser, which is why longer clips can take significantly more time than still images.

Use Real-life for photos, camera video, and natural textures. Use Anime for line art, flat shading, and cel animation. Use 3D for rendered scenes, game captures, and CGI material where edges are synthetic but not hand-drawn.

Small network is best for weak GPUs or quick previews. Medium is the balanced default. Large usually provides better fine detail recovery but has the highest memory and compute requirements.

Start with Medium plus the correct content type, then compare against Small to estimate whether Large is worth it for your source. Highly compressed inputs can show sharper edges but also stronger artifact visibility, so evaluate a short sample first.

For video, test the first 5 to 10 seconds before committing to a full clip. If your browser gets unstable, lower the network size, close other GPU-heavy tabs, and keep resolution and clip length moderate.

If WebGPU is unavailable, processing will not start. Update your browser, ensure hardware acceleration is enabled, update your GPU drivers, and restart the browser. Some devices or enterprise policies may still block WebGPU.

Video upscaling can be slow and memory-intensive on longer clips. Browser codecs also vary by platform, so exported format support can differ between Chrome, Edge, Firefox, and Safari.

Processing is designed to run locally in your browser session. Uploaded files are converted to temporary object URLs and cleaned up after use or when the page is reset.

No account is required for core upscaling. Still, as with any website, analytics, ads, and browser extensions may collect telemetry independently from the local model pipeline.