A good result starts with the highest-quality original you have. Keep that file and work on a copy. Enlarging a thumbnail before you remove its background does not reveal the original hair, lettering or fine edges: it only gives an algorithm more pixels to estimate.
1. Check the source dimensions
Open Image Dimensions to check width, height and aspect ratio. A 600 × 400 photograph has 240,000 pixels. Enlarging it to 1200 × 800 creates four times as many output pixels, but it does not create four times as much reliable information. Use the source photo rather than a screenshot of its preview.
2. Fix tone before increasing sharpness
Load the picture in Image Enhancer. Begin with brightness and contrast close to 100. Increase brightness gently if the subject is dark, then compare with Show original. Auto levels can stretch a narrow tonal range, but it may not suit a deliberately soft or low-contrast picture.
Try a small sharpness amount and inspect high-contrast edges. A pale outline beside dark hair or lettering is a sharpening halo; reduce the setting. Saturation changes color intensity, not exposure. If skin or product colors look unnatural, return saturation toward 100 rather than compensating with more contrast.
3. Decide whether AI is appropriate
The optional AI photo enhancer processes a copy up to 768 pixels on the long edge and returns twice those dimensions. That bounded size keeps the work more practical in a browser. It is useful for experimenting with small photos, but its predicted texture is not evidence of what was originally present. Do not use enhanced unreadable numbers, faces or labels as verification.
First use downloads the model and library files. The picture is processed on your device rather than uploaded to an image API. Model loading and processing speed depend on the connection, browser and available device memory. If the AI mode cannot finish, the manual controls remain available.
4. Choose a background-removal method
In Background Remover, use AI remove portrait background for a person. The model is designed for portraits, so a chair, pet or transparent bottle may not be separated correctly. Keep a contrasting source background where possible and check around hair, hands and clothing.
For a product or logo on a flat background, sample the background color instead. Start with Connected image edges so a similar color inside the subject is less likely to disappear. Increase tolerance only as much as needed. All matching colors also removes matching pixels that are isolated inside the picture.
5. Inspect and refine
The checkerboard represents transparency and is not part of the PNG. Use the Erase brush to remove leftover pixels, and Restore to bring back parts of the original subject. Work with a smaller brush around fine edges. Undo returns a recent change; Restore original starts again from the selected image.
Check the cutout on the background where it will actually be used. An edge that looks clean on white can have a visible fringe on black. Keep important text, logos and fine lines in the inspection rather than judging only the center of the subject.
6. Export for the destination
Download a transparent PNG for a cutout. JPG cannot retain transparency. AI photo enhancement uses a white-backed RGB image, so perform a portrait cutout after enhancement if that is your intended workflow. For a website photograph that does not need transparency, JPG or WebP can be more suitable; use Compress Image after choosing final pixel dimensions.
A concrete example
For a small profile image, start with a 600 × 800 portrait. Make a modest tone adjustment, compare it with the original and export the result. If you try AI mode, its input copy becomes 576 × 768 and its 2× result is 1152 × 1536. Remove the portrait background next, refine any missing edges and download PNG. Finally crop and resize a copy for the profile destination, keeping the larger cutout for reuse.
Keep the right expectations
No automatic model succeeds on every subject. Transparent objects, motion blur, severe compression and tiny faces are difficult cases. A cleaner original photograph or a careful manual edit can be more useful than stronger processing. Review the output before publishing it.
Model information: MODNet portrait model and UpscalerJS ESRGAN Slim.