AI Photo Restore — bring 1900s prints back to life
Drop a damaged photo. Get a restored version in 15–45 seconds. GFPGAN + CodeFormer + DeOldify + SwinIR + Photo-Restore XL multi-pass routing repairs scratches, sharpens faces, colorizes black-and-white, and fixes faded color in one click. Free, no watermark, full commercial license.
Face-aware restoration · Family portraits, ID photos, group shots — sharpens eyes, lips, skin without plastic look.
From a torn 1940s print to a museum-grade portrait in four clicks
No darkroom, no Photoshop healing brush, no $200/photo restoration studio. Drop the scan, pick the engine, toggle the passes you need, hit restore — most photos finish in 15–45 seconds.
Scan or photograph the print
Use a flatbed scanner at 600 DPI for best results, or a phone camera in even daylight without flash. PNG, JPG or WEBP up to 16 MB. Even smartphone snaps of glossy prints work — the AI handles glare and angle.
Pick the right engine
GFPGAN handles 80% of family portraits. CodeFormer for severely damaged faces. DeOldify for black-and-white colorization. SwinIR for fabric and background detail. Photo-Restore XL when you need every fix in one shot.
Toggle the passes you need
Scratch removal for torn or creased prints. Colorize for B&W or sepia. Denoise for grain and JPEG artifacts. Face Enhance for any portrait. Each pass adds 1–3 credits and 3–10 seconds — only enable what your photo actually needs.
Download in 15–45 seconds
PNG keeps full archival quality, JPG saves bandwidth, WEBP balances both. Side-by-side preview lets you compare before downloading. Commercial license included on every render — print, share, frame, gift.
Six teams that bring damaged photos back to life
From genealogists to wedding archivists, the bottleneck is the same: a torn, faded or scratched print that holds the only record of a person, place or moment. Below are the six teams that replace $200/photo restoration studios most often.
Families & genealogists — grandma's wedding, restored
A creased 1952 wedding portrait with water stains and a tear across the bride's veil. Photo-Restore XL repairs the tear, GFPGAN sharpens both faces, DeOldify adds era-appropriate ivory and blush tones. Print framed by Sunday for the family reunion.
Museums & historical societies — the 1900s archive, indexed
A 12,000-print county historical archive scanned years ago at 200 DPI. Batch CodeFormer + SwinIR runs overnight, recovers detail in faces and signage. Curators tag and publish 10× faster than the previous "send to specialist" workflow.
Wedding photographers — a parent's heirloom on the welcome table
Couples want their grandparents' wedding photos at the reception. GFPGAN + Scratch Removal restores both sets of grandparents in under 5 minutes per photo. Same-day delivery, no studio handoff, included as a premium add-on.
Funeral homes — obituary portraits at zero notice
Family arrives with a faded driver's license photo or wallet print, needs a 16×20 portrait by tomorrow morning. CodeFormer rebuilds severely damaged faces; Photo-Restore XL handles color and grain in one pass. Dignified, fast, no per-job retoucher fee.
Veterans & military families — service portraits, modernized
WWII and Vietnam-era service photos faded to sepia, edges crumbling. DeOldify recreates uniform colors with historical accuracy; Scratch Removal handles album damage. Perfect for memorial walls, anniversary gifts and VFW tributes.
Heritage brands — founder portraits for the "Our Story" page
A 1920s photo of the company founder in a cluttered warehouse. SwinIR sharpens background detail, GFPGAN brings the face into modern focus, DeOldify adds plausible color. The hero image of every "About Us" page rewrite, no studio shoot needed.
Five engines, picked for the damage
GFPGAN handles 80% of family portraits — face-aware sharpening without plastic look. CodeFormer for severely damaged or very low-res faces. DeOldify for B&W colorization. SwinIR for fabric, paper grain and backgrounds. Photo-Restore XL when you need every fix in one shot — switch between them per photo, no separate accounts.
Six rules for restorations that look authentic, not Instagram
Practical, tested advice for getting restorations that an archivist or family member will trust — not the over-saturated, plastic-skinned "AI miracle" look that screams fake.
Scan, don't photograph, when possible
A flatbed scanner at 600 DPI gives the AI 3× more detail to work with than a phone camera. If you must use a phone, shoot in even daylight without flash, hold parallel to the print, and crop tightly. Better source = better restoration. Always.
Match the engine to the damage
Light fading + small scratches → GFPGAN + Scratch Removal. Severely blurred or low-res face → CodeFormer (use w0.5–0.7 to preserve identity). B&W with no scratches → DeOldify alone. Heavy multi-damage → Photo-Restore XL. The wrong engine doubles your re-render rate and can falsify identity.
Don't over-colorize family portraits
DeOldify guesses skin tone, hair color and clothing dye. For known relatives, lower the colorize strength to 50–60% and treat the output as a starting point. Cross-check with surviving color photos or family memory before printing — over-saturated guesses on a great-grandmother feel disrespectful.
Always run Face Enhance for portraits
General restoration smooths faces into a "wax doll" look — pores disappear, eyes lose catch-lights, lipstick edges go fuzzy. Face Enhance runs GFPGAN as a second pass focused only on detected faces. Costs 2 extra credits but is the difference between a publishable portrait and an obvious AI render.
Audit at 100% zoom on faces, not full screen
Identity drift, hallucinated features and mismatched skin texture all hide at full-screen view. Always zoom 100% on every face and compare to the original. If a feature looks invented (different eye color, missing freckle, altered jawline), lower restoration strength or switch to GFPGAN at lower w-value — never publish without checking.
PNG for archive, JPG for sharing
PNG is lossless — required for museum archives, family heirloom storage, and any future re-restoration. JPG is fine for sharing on Facebook, family group chats and email. WEBP for web galleries. Wrong format = lost archival value (JPG generations stack up and degrade) or wasted bandwidth (PNG too heavy for messaging).
AI Photo Restore — Frequently Asked Questions
Everything you need to know about restoring damaged photos on ZNIX.ai.
What is AI photo restoration?+
AI photo restoration uses deep-learning models to repair scratches, sharpen blurred faces, colorize black-and-white prints, fix faded color, and remove stains, dust and tape marks — all in a single pass. ZNIX.ai routes your scan to one of five engines — GFPGAN, CodeFormer, DeOldify, SwinIR and a Photo-Restore XL multi-pass pipeline — so you can pick the strongest model per photo. Family portraits, museum archives, wedding heirlooms, military service photos — bring them back to life.
Is the AI photo restorer free?+
Yes. New accounts get free credits on signup — enough to restore several photos at no cost. Paid plans start under $20/month and unlock the Photo-Restore XL all-in-one engine, batch queues, the colorization model, and priority queue. No watermark on free or paid output. Compared to traditional restoration studios charging $50–$300 per photo, the cost-per-photo is typically 99%+ lower.
Which engine should I pick?+
Light damage + family portraits → GFPGAN (the default) with Scratch Removal. Severely damaged or very low-res faces → CodeFormer at w0.5–0.7 (preserves identity). B&W or sepia with no scratches → DeOldify alone. Fabric, backgrounds, paper grain → SwinIR. Multi-damage (scratches + fade + low-res + B&W) → Photo-Restore XL all-in-one. The right engine halves your re-render rate.
How long does a restoration take?+
Most single-pass restorations finish in 15–25 seconds. Photo-Restore XL with all four passes (scratch + colorize + denoise + face) takes 30–60 seconds. Batch queues run in parallel, so 50 family-album scans usually finish inside 10 minutes. Compare to studio turnaround of 3–14 days per photo.
Will the AI invent fake details on a relative's face?+
AI restoration models do generate detail in damaged regions — that's how they exceed pixel-by-pixel healing. ZNIX.ai's engines are tuned for fidelity over creative invention; CodeFormer specifically lets you control identity preservation via the w-value (use 0.5–0.7 for faces). Always audit at 100% zoom and compare to surviving color photos. For legal, forensic or genealogical use where invented detail is a liability, lower restoration strength to 30–50% or use GFPGAN alone.
Can DeOldify give realistic colors?+
DeOldify produces era-aware, plausible color — skin tones, hair color, clothing dyes that match historical norms for the photo's decade. It does not "know" the actual color of your great-grandmother's dress. For known relatives, treat colorization as a starting point; lower strength to 50–60% and cross-check against surviving color photos or family memory. For museum-grade work, DeOldify is the strongest open-source option but always disclose colorization in captions.
Can I use restored photos commercially?+
Yes. Every restored output includes a full commercial license — use them in books, films, museum exhibits, brand "Our Story" pages, anywhere. Output is yours. No attribution required and no per-use fees, regardless of volume. The original photo's copyright (if any) is your responsibility — restoration does not create new rights, just a better-quality version of what you already own.
Can I batch restore an entire family album?+
Yes, on paid plans. Drop up to 100 photos per batch, pick one engine + pass combo, and the queue runs in parallel. Common workflows: families digitizing a 200-photo grandparent album, museums batching a county historical archive, wedding photographers restoring a couple's combined heritage prints. Mix engines per batch by uploading separately.
Are my family photos kept private?+
Yes. Photos uploaded to ZNIX.ai are processed in-session and auto-deleted after restoration completes. They are never added to any training dataset, shown to other users, or shared with third parties. Your photo, your data, your output. Premium plans include private workspaces with no provider-side logging — important for sensitive estate, legal or memorial work.
How is this different from sending to a Photoshop retoucher?+
A skilled Photoshop retoucher charges $50–$300 per photo and takes 3–14 days. The result is excellent but unsustainable for a 200-photo family album. ZNIX.ai handles 80% of family-restoration cases at near-studio quality in under a minute, at 1% of the cost. For museum-grade or commercial publication where invented detail is unacceptable, a human retoucher is still the gold standard — use ZNIX.ai for the 80%, send the irreplaceable 20% to a specialist.
Bring your photos back to life — drop your scan now
Scan an old print, pick the engine, toggle scratch / colorize / denoise / face passes. Five engines in one workspace, archival-grade output, batch queue. Free credits on signup, full commercial license, no watermark.
No credit card · Free credits on signup · Commercial license · No watermark