How it works.

• Prompt-to-image generation with compatible browser WebGPU models.
• Simple model picker with only currently loadable browser models such as SD-Turbo, Z-Image Turbo and Janus-Pro-1B.
• Technical model details and seed controls stay in Advanced settings.
• Research-only experimental models are not shown in the regular public flow.
• Image size options for Square, 16:9, 9:16, 4:3 and 3:4 when the selected model supports them.
• Visible model download and generation progress.
• Export and import model caches for supported local ONNX models.
• Prompt queue for stacking multiple local image generations.
• Automatic local gallery with click-to-expand image viewing.
• No account, no UnfilteredPixel prompt API, and no prompt upload.

Does the prompt leave my device?

No prompt is sent to UnfilteredPixel. The selected model runs locally in your browser after model and runtime files are downloaded from public hosts.

Why does it download so much?

AI image models are large. UnfilteredPixel shows model names directly, keeps technical details in Advanced settings, and only enables sizes that the selected browser runtime can handle. In this build, SD-Turbo is the stable local runtime.

Can I keep model downloads locally?

Yes. Supported ONNX model caches can be exported to a local folder and imported later into the browser cache.

Does it work in every browser?

No. Local image generation needs WebGPU and a compatible GPU. If WebGPU is unavailable, the app keeps the rest of the interface usable and explains the limitation.

Can I choose any model?

UnfilteredPixel shows clear model names but keeps provider IDs, runtime details, size support, seed settings and negative prompt details inside Advanced settings.

How does it choose a model?

SD-Turbo is the default because it is the stable local browser runtime in this build. Z-Image Turbo and Janus-Pro-1B are shown as experimental options because this app includes runtime paths for them.