How to Fix a Blurry Face in a Portrait Across Topaz and Magnific
Recover a soft or motion-blurred face by combining the strengths of Topaz Photo AI and Magnific without making it look fake.
A blurry face is the hardest thing to fix, because the tools have to invent features that were lost. Push too hard and you get a stranger with plastic skin. The reliable path is to use each tool for what it does best: Topaz to recover and clean, Magnific for a gentle detail pass. This guide walks the full combined workflow.
What you need
- Topaz Photo AI installed and licensed
- A Magnific account with credits
- The blurry portrait at the highest resolution you have
- If possible, a sharp reference photo of the same person
Step 1: Start in Topaz with Recover Faces
Open the portrait in Topaz Photo AI and enable Face Recovery on the blurred face. Keep the strength moderate, around 60, and add light denoise. Topaz is conservative, so it tends to keep the person looking like themselves, which is exactly what you want as a base.
Step 2: Export a clean intermediate file
Export the Topaz result as a high-quality PNG or TIFF. This becomes the input for Magnific. Saving lossless here matters because the next pass will amplify any compression artifacts you leave behind.
Step 3: Run a low-creativity Magnific pass
Upload the Topaz output to the Magnific Upscaler. Set Creativity low, around 2 to 3, and Resemblance high, around 14, with the Portrait preset. The goal is to add micro-detail to skin and eyes, not to redraw the face, so resist the urge to crank Creativity.
Step 4: Compare against the reference and decide
Open the final image next to the sharp reference photo if you have one. Check that the eye spacing, nose, and mouth still match the real person. If the tools drifted the likeness, lower Magnific Creativity and Topaz strength and rebuild from the original.
Result: a motion-blurred face becomes clearly recognizable, with defined eyes and natural skin texture, and crucially it still looks like the same person rather than an AI approximation of one.
Watch related tutorials
12:00
08:30
06:50
18:42
21:09
23:55