Unblur, Restore, Colorize or Ultra HD: which UnBlur mode to use
Updated 13 October 2026 · UnBlur AI guides · 5 sources
AI results are estimates. AI restoration predicts what a sharper or repaired photo might look like. It can invent or change details, especially faces, text and fine patterns. Always keep your original, and do not use an AI-restored image as evidence or to identify someone.
- Unblur, Restore Old Photo and Colorize use 1 credit each and run the FLUX.1 Kontext Pro image model from Black Forest Labs. 1
- Ultra HD uses 3 credits, runs FLUX.2 Pro and returns the largest result: up to 2048 pixels on a side, about 4 megapixels. 2
- Every mode is a generative model: it predicts what the photo might look like, so faces, text and fine patterns can change. 3,4
- Colorize adds plausible colors, not the real ones. Treat the result as an illustration and keep the black-and-white original. 5
UnBlur AI has four modes, and the right one depends on what is wrong with the photo, not on how important it is. In short: use Unblur for a recent photo that came out soft or shaky, Restore Old Photo for a scanned print with scratches, dust or fading, Colorize for a black-and-white photo you would like to see in color, and Ultra HD when you want the largest, most detailed result the app can produce. This guide explains what each mode is for, what it costs in credits and where it stops working.
The four modes at a glance
| Mode (in the app) | Use it for | Credits | Result size | Watch out for |
|---|---|---|---|---|
| Unblur: "Sharpen blurry shots" | Recent photos with mild camera shake, slight motion or missed focus | 1 | About one megapixel, same shape as your photo | Severe blur, small faces, text |
| Restore Old Photo: "Fix scratches & fading" | Scans or photos of old prints with dust, scratches, creases or faded color | 1 | About one megapixel | Faces that look "too new", lost texture |
| Colorize: "Add color to B&W" | Black-and-white or sepia photos | 1 | About one megapixel | Colors are guesses, not history |
| Ultra HD: "Newest AI, max detail" | When you want more pixels and fine detail, for example for a small print | 3 | Up to 2048 px on a side, about 4 MP (about 2048×1536 for a 4:3 photo) | More detail also means more invented detail |
Unblur, Restore Old Photo and Colorize run FLUX.1 Kontext Pro, an image-editing model made by Black Forest Labs 1,3. Ultra HD runs the newer FLUX.2 Pro from the same company 2. Both are hosted by Replicate, which runs the model and returns the result to our server.
How a job works
- Pick a photo from your photo library. The app works on photos already in your library; to work on an old print, scan it or photograph it with your iPhone first.
- Choose a mode. Before the very first photo is sent, the app asks for your consent and explains what is sent and to whom (see where your photos go).
- The app scales the photo to at most 2048 pixels on the long side and compresses it to about 1 MB before uploading, then waits for the result, which usually takes about 30 to 60 seconds.
- Compare the ORIGINAL and ENHANCED views. Zoom in on eyes, teeth, hands and any lettering.
- If you like it, tap Save to Photos or Share. Your past results are listed in the history on the Profile tab.
If a job fails, or the model's safety filter refuses a photo with a "may contain sensitive content" message, no credits are used. Credits come from one-time credit packs or from the UnBlur Pro subscription, which includes credits; some eligible users are offered a free trial of the subscription. The app does not hand out starter credits at install, and you can use it in guest mode without creating an account.
Unblur: for recent blurry photos
Unblur is meant for the everyday case: a photo of people or a scene that is slightly soft because the phone moved, the subject moved or focus landed in the wrong place. Before spending a credit, it is worth trying the free fixes in how to unblur a photo on iPhone, such as picking a sharper frame from a Live Photo. If you are not sure what kind of blur you have, motion blur vs focus blur explains how to tell.
The model draws a crisp version of the photo. With light blur that usually looks natural. With heavy blur the model has very little to go on, and it may produce a clean picture that is not quite your picture: a slightly different face, a different pattern on a shirt or invented lettering on a sign. That is a property of all generative restoration, explained in what AI photo restoration can and can't do 4.
Restore Old Photo: for scans of damaged prints
Restore Old Photo is tuned for the problems of old prints: dust specks, scratches, small tears, creases, fading and color casts. Our server instructs the model to keep each person's face, identity, expression and pose, and the composition, the same. That is an instruction to a generative model, not a promise, so check the result against the original. The better your scan, the less the model has to guess: see how to restore old photos for the full workflow and restoring faded photos for color problems you can fix by hand.
Colorize: for black-and-white photos
Colorize adds color to a black-and-white or sepia image. A gray value can belong to many colors (a mid-gray could be a red dress or a green one), so research on automatic colorization describes the task as producing a plausible color version, not the true one 5. Skin, sky and foliage usually look convincing; clothing, cars, uniforms and painted walls are guesses. If the real colors matter, read how to colorize black and white photos, and label the result as colorized when you share it.
Ultra HD: when you need more pixels
Ultra HD costs 3 credits and returns the largest image the model accepts in your photo's shape: up to 2048 pixels on a side and about 4 megapixels in total. At around 300 pixels per inch that is enough for small prints such as 4×6 or 5×7 inches, not for posters; the numbers are worked through in upscaling a photo for print. More pixels give the model more room to draw fine texture, which is helpful for hair and fabric but also means more detail that was never in the original.
Which mode should I try first?
- A recent, slightly blurry photo: Unblur. If you want it larger or more detailed, try Ultra HD afterward.
- An old print with damage: Restore Old Photo on a good scan.
- An old black-and-white print: Restore Old Photo first if it is damaged, save the result, then run Colorize on the saved copy. Each pass is a new AI estimate, so compare with the original after each step.
- A screenshot or photo of a document: none of them. AI tends to garble letters; see fixing blurry screenshots and text.
Common questions
Which UnBlur mode gives the biggest image?
Ultra HD. It returns up to 2048 pixels on a side, about 4 megapixels in total. The other three modes return about one megapixel in the shape of your photo. 2
Do I lose a credit if the result fails?
No. If a job fails, or the model refuses the photo as possibly sensitive content, no credits are used.
Can I run two modes on the same photo?
Yes. Save the first result to Photos, then pick it as the input for the next mode. Each pass uses its own credits and adds its own AI changes, so compare with the original after each step.
Are colorized colors accurate?
Not necessarily. Colorization models choose plausible colors from patterns they have learned; they cannot know the real color of a dress or a car. 5
Sources and further reading
- Replicate. FLUX.1 Kontext [pro] by Black Forest Labs (model page). replicate.com
- Replicate. FLUX.2 [pro] by Black Forest Labs (model page). replicate.com
- Black Forest Labs, et al. FLUX.1 Kontext: Flow matching for in-context image generation and editing in latent space. arXiv:2506.15742. 2025. arxiv.org
- Blau Y, Michaeli T. The perception-distortion tradeoff. IEEE CVPR. 2018. arxiv.org
- Zhang R, Isola P, Efros AA. Colorful image colorization. ECCV 2016 (Lecture Notes in Computer Science). arxiv.org