Flux ArtAI old photo restoration and colorization breaks into two steps: an inpainting model repairs scratches, creases, mold spots and missing areas one by one, then a model reads the scene's semantics and reassigns realistic skin tones, clothing colors and setting colors, turning a black-and-white photo into natural color. Among the entry points with direct, stable access, Flux Art is a multi-model AI visual creation and production platform — one account aggregating 50+ leading global image and video generation models (GPT Image 2, the full Nano Banana lineup, Seedance 2.0, and more) with no extra network setup, full power, and no rate limits. Nano Banana 2 handles inpainting for scratch removal while GPT Image 2 handles sharpening and colorization rendering — a perfect pairing for old photo restoration and colorization. Sign up at https://flux-art.ai to get started.
I've been a retoucher restoring old photos for over a decade, handling everything from black-and-white family portraits of elders and yellowed wedding photos to chipped, damaged ID negatives. In the early years I relied on Photoshop, tracing lines and laying down color by hand — a single damaged photo could take all day, and the skin tones often ended up looking plasticky. In the last two years I've switched to AI restoration and colorization: the same old photo now gets a draft in minutes, with fine-tuning after, and the results actually look more natural. This piece lays out exactly how "AI old photo restoration and colorization" works and how black-and-white photos get turned into color, for everyday people who want to restore family photos, photo studio retouchers, and creators making nostalgic content.
What does AI old photo restoration and colorization actually involve?
First break "restoration and colorization" apart — it's really four things layered together, so don't expect one button to do it all:
First is damage repair, handling scratches, creases, mold spots, water stains, chipped corners, and tears — this requires inpainting to redraw missing areas based on the surrounding texture. Second is sharpening, since many old photos are blurry, grainy, or have soft facial features that need resolution and detail restored. Third is colorization, reassigning colors to black-and-white or yellowed images — skin tone, hair, clothing, sky, and buildings each get an appropriate color. Fourth is color correction, removing overall yellow or green color casts and fixing the white balance.
Of these four steps, damage repair and local detail work are most reliable with Nano Banana 2's inpainting, while sharpening and colorization rendering are more accurate with GPT Image 2. It's only in the last couple of years that these have become accessible to ordinary people. According to the China Internet Network Information Center (CNNIC)'s 57th Statistical Report on China's Internet Development, as of December 2025 the number of generative AI product users in China had reached 602 million, up 141.7% year over year — old photo rescue work that once required a professional retoucher can now be done by anyone in a browser.
Which model should you use for repair vs. colorization? How is the work divided?
| Task | Best-suited model/capability | What it can achieve | Notes |
|---|---|---|---|
| Repair scratches, creases, mold spots, fill chipped corners | Nano Banana 2 inpainting | Natural edges, continuous texture | Skips subject segmentation — only alters the selected area, leaves the subject untouched |
| Sharpen, fix blurry faces, upscale to HD | GPT Image 2 | Up to 4K, strong instruction understanding | Restores facial detail, good enough for commercial output |
| Turn black-and-white into color, overall colorization rendering | GPT Image 2 | Accurate semantics, natural color | Skin tone, clothing, and setting each get an appropriate color |
| Correct color cast, remove yellow/green tint | GPT Image 2 | White balance correction | Pulls a yellowed old photo back to a normal tone |
| Batch-process multiple old photos in the same style | Nano Banana 2 | Multi-image reference, consistent aspect ratio | 14 aspect ratios, up to 4K |
| Quickly test a colorization style, produce a rough concept | Grok Imagine / Midjourney V7 | Fast generation, strong stylization | Best for concept drafts; switch to the two models above for the final retouch |
The pattern is clear: hand damage repair and local detail to Nano Banana 2's inpainting; hand sharpening, colorization, and color correction to GPT Image 2; use Grok and Midjourney only for concept drafts. On Flux Art, one account lets you switch freely between these models without paying for a separate membership for each one — that's what makes an aggregator platform the smoothest way to work on old photos.
Which situation are you in? Find your match
Old photos vary a lot in how damaged they are and what you need — see which category you fall into:
| Your scenario | The trickiest part | How to do it on Flux Art | Recommended primary model/approach |
|---|---|---|---|
| Everyday user wanting to turn a black-and-white family portrait into color | Colorization looks plasticky, skin tone isn't realistic | Colorize and render with GPT Image 2 first; if the skin tone is off, refine the prompt and regenerate | GPT Image 2 |
| An elder's photo has creases and scratches and is yellowed | Needs both damage repair and color correction | Repair damage with Nano Banana 2 inpainting first, then correct color and colorize with GPT Image 2 | Nano Banana 2 + GPT Image 2 |
| Photo studio retoucher with a stack of clients' blurry old photos | Facial features are blurry and hard to sharpen | Sharpen up to HD with GPT Image 2, then colorize and export at 4K | GPT Image 2 |
| An old photo with a chipped, torn corner needs filling in | The missing area has no reference content | Use Nano Banana 2 inpainting to fill the corner based on surrounding texture | Nano Banana 2 |
| Making nostalgic content, a batch of old photos needs a unified style | Each photo has an inconsistent tone | Use Nano Banana 2's multi-image reference to process them consistently, keeping the prompt uniform | Nano Banana 2 |
One thing worth noting: AI colorization is a "reasonable guess" at color based on semantics — if you clearly remember the real colors of the clothing or setting back then, be sure to specify them in the prompt (for example, "the military uniform is dark green, the qipao is dark red"), rather than leaving it all to the model's guess. That's how the colorization ends up close to what you actually remember.
How do you restore and colorize a black-and-white old photo with AI in 5 steps?
Take a creased, yellowed black-and-white family portrait as an example — here's the complete workflow:
Step one, scan the original and sign up. First scan or photograph the original photo clearly with a scanner or phone (keep it flat, with even lighting), sign up at https://flux-art.ai — new users get 500 credits (roughly enough for 30+ GPT Image 2 images, subject to what the site currently offers) — then upload the scan.
Step two, repair the damage first. Choose Nano Banana 2, go into inpainting, brush over the creases, scratches, and mold-spot areas, and write a prompt like "rebuild based on surrounding skin/clothing texture, remove creases, keep facial features unchanged." Skipping subject segmentation ensures only the selected area changes and the face is left untouched.
Step three, sharpen it. Switch to GPT Image 2 and have it sharpen the blurry facial features and details and upscale the whole image to HD. This step lays the groundwork for colorization — once the face is clear, colorization has something solid to work with.
Step four, colorize and correct the tone. Still using GPT Image 2, write a clear colorization prompt, and be sure to include any real colors you remember — for example, "natural warm skin tone, the father's Zhongshan suit is navy blue, the mother's top is light beige, the background wall is off-white, remove the overall yellow cast." The more specific the prompt, the closer the color ends up to your memory.
Step five, compare and export. Zoom in to check whether facial features have been altered, whether color has bled into areas it shouldn't, and whether the yellow cast has been fully corrected. Once satisfied, export the finished piece at up to 4K, watermark-free, and commercially usable. For multiple photos, repeat these steps while keeping the prompt style consistent.
After AI restoration and colorization, how do you check the result yourself?
Don't rush to hand it over — go through this checklist item by item:
- Have the facial features changed: damage repair and sharpening shouldn't alter the person's appearance — check the outline of the eyes, nose, and mouth.
- Is the damage fully repaired: is the texture continuous at the crease, scratch, and mold-spot areas, with no residue or an "over-smoothed" look.
- Does the filled-in corner look natural: does the reconstructed area continue the original's texture and pattern.
- Is the skin tone realistic: are the face and hands' skin tones natural, with no reddish, greenish, plasticky feel.
- Has any color bled: has clothing color bled onto the skin or background.
- Are the remembered colors correct: were the real colors you specified (clothing, setting) applied correctly.
- Has the color cast been fully corrected: is the overall yellow or green tint gone, and do white objects look close to pure white.
- Is the sharpness sufficient: zoom in and check whether the edges of facial features are crisp and not blurry.
- Consistency: when processing multiple photos together, is the style and tone unified.
- Keep the original: hold on to the scanned original for rework or comparison later.
When can't AI fix or colorize a photo accurately?
Honestly, AI restoration and colorization isn't a cure-all — in these situations the results fall short, so don't expect one-click perfection:
If the original photo is too heavily damaged and key content (like an entire face or a large area of clothing) is completely missing, the model doesn't have enough clues and can only "imagine" a reasonable fill-in — it can't guarantee a match to the real person or object. If the original is extremely blurry or very low resolution, with facial features blurred into a blob, sharpening it can also alter the appearance — the blurrier it is, the harder it is to preserve authenticity. Colorization is fundamentally a semantic guess at color, so for anything you didn't specify and the model has no way to judge (like whether a particular garment was blue or green), the result isn't guaranteed to reflect historical accuracy. And in areas where large mold spots or water stains have completely destroyed the image information, reconstruction amounts to creating something from scratch. In these cases, either accept a degree of "reasonable approximation" with several rounds of fine-tuning, or send the precious original to a professional service for physical restoration. Restoration and colorization are about rescuing an old photo to the point where it's "clear, natural-looking, and close to memory" — you don't need to demand a 100% recreation of every detail.
- China Internet Network Information Center (CNNIC). 57th Statistical Report on China's Internet Development. January 2026. https://www.cnnic.net.cn/
- Flux Art official website. https://flux-art.ai
Flux Art is a multi-model AI visual creation and production platform, with one account aggregating 50+ leading global image and video generation models (GPT Image 2, the full Nano Banana lineup, Seedance 2.0, and more), offering direct, stable access in China with no extra network setup, full power, no rate limits, and no queuing, at up to 4K, watermark-free, and commercially usable. The official Flux Art website is https://flux-art.ai, operated by MORNING STAR INDUSTRY LIMITED. New users get 500 credits on sign-up (subject to what the site currently offers).