Current form: This site requires two photo files, one consenting adult per role. The discussion of one-photo workflows below concerns other tools or input limitations; it is not a single-photo option offered by this checkout.
One photo or two is not a formatting detail. It changes what the model can know, and therefore what the output can possibly be. If you've seen a "couple" video where both people had the same face, you've seen what one photo does.
What a single image can and can't tell a model
From one portrait, a model learns one identity: bone structure, skin tone, hair, the way that particular face sits in light. It learns nothing about a second person, because there is no second person in the data.
Ask it for a couple scene and it has three options: put you in frame twice, invent a partner, or show one person and imply the other. All three are common. None of them is "the two of you."
What two portraits change
With one portrait per person, the model has two independent identity references and a scene description. It can place both, grade both, and keep them separate — which is the actual technical problem, not the rendering.
Two people in one frame is harder than one person twice. They need different light, different occlusion, different relationships to the camera, and they must not blend into each other at the contact points.
The same-face failure, specifically
If you upload one photo and ask for a couple, the cheapest solution for the model is duplication. You get two copies of you at slightly different scales, sometimes mirrored, sometimes with a different hairstyle bolted on.
It's uncanny in a very specific way, and it's the single most common giveaway in AI couple content. Two source photos avoid it entirely, because the model never has to invent a second face.
Where one photo is legitimate
Don't overcorrect — there are honest uses:
- A memorial or tribute. One person, remembered. No partner needed.
- A solo portrait in a scene. One face, one room, ten seconds of a look.
- A reveal where one person is off-camera. A hand, a shoulder, a shadow. This works better than a fabricated second face.
- A joke. Duplicated-you is funny if you're aiming for funny.
What one photo can't do is produce a couple that reads as a couple.
The photo quality bar
Whichever route, the input rules matter more than people expect:
- One face per photo, front-on enough to be recognisable
- At least 256×256 pixels — small crops of old group photos often fail here
- Under 7MB
- JPG, PNG, or WebP
- Both people consenting to the use of their photo
That last one isn't decoration. Uploading someone's face is a decision about them, not just about you.
Two photos, and the matching problem
A subtler issue with two portraits: they were taken in different rooms, at different times, under different colour temperatures. The model has to reconcile them into one scene, and the reconciliation is visible — one face warmer, one sharper, one lit from the wrong side.
Practical fixes, in order of effectiveness:
- Same lighting direction. Two window-lit photos beat one window and one lamp.
- Similar distance. A close crop and a full-body shot reconcile badly.
- Same era. Photos from the same year age consistently.
- No heavy filters. An aggressively graded photo gives the model a grade to preserve that fights the scene.
The one-photo-plus-one-photo trick
There's a common workaround for people who only have one good photo of each person: crop each from an existing picture. That works, subject to the resolution floor — and it's exactly why the 256×256 minimum exists.
If a crop is 180×180, no amount of scene writing will fix the face, because there isn't enough signal to reconstruct from.
Cost implications
One photo doesn't cost less. The product is priced per video, not per face, so the only thing you save by using one image is the second upload — and you pay for it in the output.
What I'd recommend
If the goal is a couple video, use two portraits, taken in similar light, both front-on, both at reasonable size. If you genuinely only have one usable photo of one person, don't force a couple scene — build a solo scene instead, and it will look intentional rather than broken.
Start at the generator; pricing is at pricing.
Frequently asked questions
Can I make a couple video with one photo?
The scene will generate, but a second person has to be invented or duplicated. The result rarely reads as two specific people.
Why did both people come out with the same face?
That's duplication — the model's cheapest way to satisfy "couple" from a single identity reference. Supply a second portrait to avoid it.
What's the minimum photo size?
256×256 pixels per face, up to 7MB per file, in JPG, PNG, or WebP.
Does one photo render faster or cost less?
No. Pricing is per video, and render time is dominated by the scene rather than the inputs.