Getting great results

Do you still need to train an AI on your face?

Mostly no. In our tests, selfies used directly as references gave the same likeness as weights trained for 35 minutes. We think face training is on its way out, and we keep it for the cases where it still earns its place.

What we measured

No training (Instant): your selfies guide the photo engine on every picture. Likeness 0.62.
A profile learned in a few minutes (Studio): 0.70, the sharpest we measured.
Weights trained for about 35 minutes (Trained): 0.66 on the best trainer, 0.62 to 0.63 on the others.

Same designed person, same scenes, same face recogniser as judge, on a scale from 0 to 1. The photos of every way are on the comparison page (Compare, then Instant, Studio or Trained).

Why training is fading for faces

Photo models already know faces very well: they only need to be told which one. The newest engines take several reference photos in one request. FLUX.3, released in October 2026, takes up to ten; Nano Banana Pro and GPT Image work the same way. A handful of selfies now does the job a LoRA used to do.

Training also has a cost: a wait, a price, and defects of its own. In our runs, trained weights sometimes mirrored the face or doubled a beauty mark.

Where training still earns its place

You want to own the weights and use them in other tools.
A product, a garment or a drawing style the model has never seen: references alone drift more there than on a face.
A character that has to stay identical across hundreds of pictures or in long videos: training can still help there.

That is why Doppelshot keeps a Trained option, next to the two ways that need no long training.

What to pick in Doppelshot

Instant: ready the moment your selfies are in. Start here.
Studio: a few minutes, the same face on every scene.
Trained: about 35 minutes, your own weights to download.

Each new model keeps a face better from fewer photos. Your twin is made of your selfies, not of one model, so it follows the best engine as they come out.