AI photoshoot with your own face: which photos to upload

How it works

There are two routes. The first is training a character: the model learns your face from 15 to 30 photos, and you can then pick it in any generation. The second is reference based generation without training. Nano Banana Pro, for example, accepts up to five character images. It is faster, but the likeness is usually weaker.

For regular content, avatars and a consistent series, a trained character works better. For a one off picture, references are enough.

Which photos to upload

The model learns from variety. If every shot has the same angle and light, it learns those conditions rather than your face. Video model developers give the same advice for characters: front, left and right views, a neutral face and a smile.

• 15 to 30 photos of one person
• Different angles: straight on, three quarter, profile
• Different expressions: relaxed, smiling, serious
• Varied light and backgrounds, but the face always well lit
• Sharp shots where the face fills a good part of the frame
• A few waist up and full length photos if you want full outfits

Which photos to avoid

One bad shot in the set can show up in every generation. Fifteen good photos beat thirty random ones.

• Filters, masks and heavy skin retouching
• Sunglasses, caps and hair across the face
• Group photos and shots with other faces in frame
• Blurry, overexposed or very dark shots
• Near duplicates from the same burst
• Old photos where you look noticeably different today

Prompt ideas for the shoot

Describe the outfit, place, light and type of shot, and leave the appearance to the character. Do not ask to change facial features, or the likeness will fade.

• "Business portrait in a bright office, navy blazer, soft window light, waist up"
• "Walking along an autumn embankment, beige coat, film look, warm tones"
• "Studio portrait on a black background, rim light, black turtleneck, close up"
• "Summer cafe, white shirt, laughing and looking away, natural daylight"

Consent to use a likeness

Russian civil law allows a person’s image to be published and used only with their consent, with narrow exceptions. So you can train a character on your own photos or on photos of someone who gave written consent.

Photos of children cannot be used for training. Creating pictures with someone else’s face that pass as real is not allowed by law or by the service rules.

Privacy

Solistra has no public feed, and only you see your character and pictures. Files and the database are stored in Russia. Deleting a character immediately erases the training photos and the model weights.

How to do it in Solistra

A finished character can be picked in Image, in Video through the first frame and in effects.

• Open Characters and upload 15 to 30 photos of yourself
• Confirm that the photos show you or that you have the person’s written consent
• Wait for the character to finish training
• Pick the character in Image, Video or an effect and describe the look