The blinked frame · One instruction · The same face, awake
The shutter caught a blink and there is no second take — the toast is over, the group has scattered, the light has gone. Rebuild the eyes on the frame you have and keep everything that was already right about it.
Our own run on this site's engines. Drag the divider and check the brows first — if they jumped, the model rebuilt the face instead of the eyes. Then the smile, which should be identical in both. Every output is AI-generated; the result follows the face in your photo.
Drag to compare before and after
The model reconstructs the eye region: irises and pupils placed correctly, a catchlight in each, lashes and the eyelid crease rebuilt, and the surrounding skin reshaped the way a genuinely open eye sits.
There is a reason this problem is so common and so annoying. A blink lasts about a third of a second, and in a group of ten people the odds that at least one is mid-blink in any given frame are close to even. Photographers deal with it by shooting bursts and compositing the best eyes from a neighbouring frame — which works beautifully and requires that a neighbouring frame exists.
Most people asking to open eyes have exactly one picture, taken by a relative, of a moment that will not repeat.
The pair above is our own run and the check is quick. Look at the eyebrows: they should sit where they sat in the closed version, because opening your eyes barely moves them. Look at the spacing and the tilt of each eye against the nose. Look for a catchlight in both — the same window, on the same side.
Then look at the mouth, the hair and the collar and confirm nothing there has been quietly redrawn.
Where it gets harder: profiles and steep angles, subjects wearing glasses (the frame and the reflection both have to survive), and pictures small enough that the eye region is a couple of hundred pixels across. If the whole photo is soft rather than just the eyes closed, fix the resolution first — restoring an old or low-quality print is the better opening move, and the blink fix goes second.
Creative engine
Upload the frame, name the eye colour, generate and compare at full size.
In a group shot, crop to the person who blinked, fix that crop, and composite it back — the model has more to work with and the rest of the frame is never at risk.
Say brown, green, hazel. Left unsaid the model guesses from the face, and it guesses wrong often enough that it is worth one extra word.
The bigger the face in the picture, the better this goes. Everything else is detail.
In a group shot, cut down to that one head and shoulders before uploading. It hands the model far more pixels to work with and puts the other nine faces completely out of reach of the edit.
"Open her eyes looking at the camera, brown irises, natural lashes, keep the smile and everything else unchanged." Colour and direction are the two things the model cannot infer from a closed lid, so supply both.
Once for the eyes, once for everything you did not ask to change. Most rejected runs fail on the second pass — a subtly different jawline, a smile that shifted a millimetre — and that is easier to catch side by side than from memory.
Three unrepeatable moments and one badly timed blink each.
Fourteen people, one usable composition, one uncle mid-blink. Crop to him, fix it, put him back — and the frame everyone actually liked becomes the one that gets printed.
You get one frame where they are laughing at the right thing, and their eyes are shut. There is no reshoot of a two-year-old's specific moment of delight.
Good light, good angle, half-closed eyes that make you look exhausted. Fix the eyes, then treat the rest of the frame like any other portrait pass — a straighter smile if the mouth is the other thing bothering you.
Checked before this goes in an album.
It reconstructs eyes consistent with your face rather than copying your real ones, because a closed lid contains no information about the iris underneath. Shape, spacing and set come out right; the exact pattern of the iris is invented. State the colour and the result is close enough that nobody who knows you will pause on it.
Ask for one at a time. A prompt naming two subjects tends to get one good result and one redrawn face, whereas cropping to each person, fixing them separately and compositing gives you full control and puts everyone else out of harm's way.
Usually, and it is the case worth checking most carefully. The lens edge, the frame and any reflection all have to survive the rebuild, and a run that quietly removes the reflection or thins the frame reads as wrong even though the eyes look fine. Compare the glasses specifically before accepting.
That is the easier case — a partly visible iris gives the model real colour and position information to work from, so results are more faithful than on a fully closed lid. Say "open the eyes fully" so it does not treat the squint as the intended expression.
Not if the run is a good one, and print is a fair test because it removes the option of squinting at a screen. The tells to look for are a missing or mismatched catchlight, lashes that stop abruptly, and skin around the socket that is smoother than the rest of the face.
Uploaded over HTTPS for the edit, stored in your history, never used to advertise, and deleted when you delete it. Every output is AI-generated.
One credit pool covers Nano Banana images and Seedance and Veo video. The cost shows before every run, and failed jobs are not charged. Use a subscription for ongoing work, or a one-time pack when you just need to top up.
What's included
Secure checkout by Stripe. Card details never touch our servers.
What's included
Secure checkout by Stripe. Card details never touch our servers.
What's included
Secure checkout by Stripe. Card details never touch our servers.
What's included
Secure checkout by Stripe. Card details never touch our servers.
One-time top-ups — buy extra credits any time you run low.
What's included
Secure checkout by Stripe. Card details never touch our servers.
What's included
Secure checkout by Stripe. Card details never touch our servers.
What's included
Secure checkout by Stripe. Card details never touch our servers.
Charges appear as “KAVEL AI” on your card statement. You can cancel any time from Settings → Billing; cancellation takes effect at the end of the period you already paid for.
Powered by
Use one balance across every supported image and video model — eleven of them today, listed below — and check the credit cost before each request.
Video models
Image models
Model credit guide
You're only charged for successful generations. The exact estimate in the generator varies by model, length, resolution, audio, and number of images.
| Type | Model | Credit cost |
|---|---|---|
| Video | Seedance 2.5 | The long-take tier, priced per second: 5s 480p ≈ 211 credits, 5s 720p ≈ 473. A full 30s take runs ≈ 1,260 at 480p and ≈ 2,835 at 720p. Supplying a reference clip lowers the per-second rate. |
| Video | Seedance 2.0 | 5s 720p image-to-video ≈ 188 credits; text-to-video ≈ 308 credits. Scales with resolution and length. |
| Video | Seedance 2 Fast | Faster and lower cost. 5s 720p text-to-video ≈ 248 credits. |
| Video | Seedance 2 Mini | The cheapest Seedance tier. 5s 720p ≈ 154 credits, 5s 480p ≈ 72 credits. |
| Video | Seedance 1.5 Pro | Audio doubles the rate. 5s 720p ≈ 27 credits silent, ≈ 53 with audio; 1080p ≈ 57 and ≈ 113. |
| Video | Veo 3.1 | Billed per video, not per second (Lite tier). About 45 credits at 720p and 53 at 1080p. |
| Video | Kling 3.0 | Audio raises the rate. 5s 720p ≈ 105 credits silent, ≈ 150 with audio; 1080p ≈ 135 and ≈ 203. |
| Video | MiniMax H3 | Fixed 2K, no resolution ladder. Priced per second — a 5s clip ≈ 158 credits. |
| Image | Nano Banana 2 | Generate or edit from text and images. About 8 credits per 1K image, 12 at 2K, 18 at 4K. |
| Image | Nano Banana Pro | Consistent run times across generations. About 12 credits per 1K or 2K image, 21 at 4K. |
| Image | Nano Banana 2 Lite | Faster, simpler variant. A flat 8 credits per image at every resolution. Start here to test. |
| Image | GPT Image 2 | The lowest-cost image model here. About 3 credits per 1K image, 6 at 2K, 12 at 4K. |
| Image | Seedream 5.0 Lite | Flat pricing — about 8 credits per image whether you generate or edit. |
Other single-region edits that leave the photograph alone.
Upload the blinked shot, name the eye colour, and get the picture everyone remembers taking — then [fix the smile](/image/ai-smile-generator) if that is the other thing wrong with it.
A real edit made with this tool — drag your own photo in on the left, the prompt is already written, then press Generate.
