Bring softer portraits back into focus with local, adjustable image restoration. No account. No photo upload.
✦Private by designPhotos stay in this browser
Photo workspaceNo photo selected
▧
Your portrait goes here
Drop an image here or browse your device. You can also paste an image.
JPG, PNG, WebP, AVIF · Up to 30 MB
FACE AREA
Standard face enhancement, with optional local AI
01
Choose a portrait
Import a JPG, PNG, WebP or supported AVIF photo. Draw the face frame around one person.
02
Enhance locally
Press Enhance face in Standard mode to apply local tone, denoise and detail filters without downloading a model.
03
Compare and export
Adjust the restoration blend, inspect the before/after comparison and save the completed image.
This tool is intended for front-facing portraits. It cannot guarantee accurate reconstruction of eyes, identity or features that were not present in the original image. Low-memory devices may be unable to run the large AI model.
✳
Processing
Enhancing your portrait on this device. Large images may take longer.
Face Enhancer did not start
An essential script may be missing or blocked. Existing files and photos were not cleared. Ensure ai-face-enhancer/scripts/ was installed completely.
Using AI Face Enhancer
Choose a supported portrait and position the source-square frame around one face. Adjust the eye-line rotation manually if needed, apply enhancement, then review the face and blending carefully before export. Model licensing information is available in the linked provenance sources.
The large GFPGAN model and ONNX Runtime are downloaded on demand; choosing a local model does not supply the runtime. Manual rotation is not five-landmark alignment. AI can alter identity or invent features, especially for off-angle faces or missing detail. Native face detection is browser-dependent, and no batch mode or quality guarantee is supplied. Model licensing and component restrictions require release review.
Position a square crop around one supported face, adjust its rotation manually if needed, then request enhancement and carefully review the result and blending.
The selected crop is processed and exported as an image. A local model option does not include the ONNX runtime; model licensing and component restrictions require review before release.
Example workflow
Open a supported portrait and position the source-square frame around one face.
Adjust the eye line rotation manually and apply enhancement after any required model setup.
Compare the result with the source, especially facial identity and blending, before exporting.
Authored example: Use an authored close-up portrait scenario with the face slightly off-angle. Practice positioning the square crop and adjusting rotation; this does not demonstrate model quality or identity preservation.
Limits and model handling: GFPGAN and ONNX Runtime are downloaded on demand; model size is large. Manual rotation is not five-landmark alignment. AI may alter identity or invent features, particularly for off-angle faces or missing detail. There is no batch mode or quality guarantee, and licensing review remains a release gate.
AI Face Enhancer questions
Does enhancement preserve a person’s identity exactly?
No. The model may alter identity or invent facial features, particularly when the face is off-angle or detail is missing. Review the output against the original.
Does choosing a local model avoid downloading the runtime?
No. A local model file does not supply the ONNX Runtime. Model and runtime setup remain separate.
