Three ways to make your AI twin (2026)
Every twin starts from your photos. What changes is what happens next: nothing (Instant), a few minutes of training on a face profile (Studio), or a longer training that gives you weights to download (Trained). We rendered the same designed person in the same six scenes with all three and measured the likeness with a face recogniser.
Instant is ready the moment your selfies are in and already scores 0.62. Studio is the sharpest likeness we measured, 0.70, after a few minutes of training. Trained gives you your own weights, but in our tests its likeness equals Instant and a small mark can land on the other side of the face.
The photos every twin starts from
A designed person (nobody real): 20 photos taken from one to three metres, no mirror, one beauty mark at the lip.
At a glance
| Way | Wait | Price | Likeness | Best for |
|---|---|---|---|---|
| Instant | None | 2 credits a photo | 0.62 | Trying it now, a first pack in minutes |
| Studio | A few minutes | 100 credits once, then 2 to 3 a photo | 0.70 | The best likeness, the same face on every scene |
| Trained | About 25 to 35 minutes | 100 credits once, then 1 a photo | 0.62 - 0.66 | Owning your weights and using them elsewhere |
Prices and features checked in September 2026. Tell us what changed: [email protected].
The same six scenes, three ways
Instant · Likeness 0.62
Your photos guide the engine on every picture. Natural proportions, one mark, no training to wait for.
Studio · Likeness 0.70
A profile trained on your face. The quality tier: you pay more, you get the best.
Trained · Likeness 0.62 - 0.66
Your own weights, yours to download. Same likeness as Instant in our tests; a small mark can land on the other side.
The trainers behind the Trained tier
Three engines, same photos, same scenes. Ideogram scored best, so the Trained tier runs on it.
Likeness 0.66. Ideogram: editorial framing, one mark on most scenes, sometimes mirrored.
Likeness 0.62. FLUX portrait: tight portraits, the face very close.
Likeness 0.63. Qwen Image: the best skin of the open models, the mark can double.
Likeness is the cosine similarity of a face recogniser between each photo and the portrait, averaged over the photo scenes. It measures identity only, not framing or proportions.
Our honest take on training
We tested it. Weights trained on a face for about 35 minutes gave the same likeness as selfies used directly, with no training at all. Photo models now learn a face from a few photos, and each new one does it better. We think face training is on its way out.
Training keeps one real advantage: the weights are yours to download and use anywhere. So the option stays. If you just want great photos of yourself, pick Instant or Studio.
Questions
Do I need to train anything?
No. The Instant twin works from your photos alone. Training is a choice for a sharper likeness (Studio) or for owning weights (Trained).
Why is Studio the best?
It learns your face once, in a few minutes, and keeps it the same on every scene. In our tests it scored 0.70 against 0.62 for the others.
Can I switch later?
Yes. A twin keeps its photos: you can train it as Studio or Trained at any time, and use it Instant meanwhile.
Do I still need to train an AI model on my face?
For most photos, no. In our tests your selfies used directly (Instant) matched weights trained for 35 minutes, and a profile learned in a few minutes (Studio) beat both. Train only if you want to own the weights.
Your first photo is free.
One selfie, the first photo free, no card. An Instant twin takes ten selfies and a plan.