An AI character is a private model that has learned a specific person\'s (or pet\'s) face, body
proportions, and identifying features from reference photos you upload. Once trained, the character can
be rendered into any scene — at any angle, in any lighting, wearing any outfit — by the same Flux engine
that generates the scene itself.
Think of a character as a saved likeness, not as a saved photograph. The likeness is what gets used; the
output photo is generated fresh every time.
The Flux engine
Lover Snap uses Flux, an image-generation model by
Black Forest Labs. Flux is notable for:
Photorealism — outputs that read as photographs rather than illustrations or filtered selfies.
Strong handling of human faces, including likeness preservation across generations.
Reasonable speed at consumer-quality settings (about 23 seconds per generation).
Flux is the same family of models used by professional studios for high-fidelity AI imagery in 2026.
Training process
How it works
From a few photos to lifelike new memories.
1
Upload selfies
Pick 3-16 photos of yourself or the person you want to include. Variety helps: close-ups, full-body shots, different angles and outfits. Avoid sunglasses or hats that hide the face.
2
Train your AI character
Our Flux engine builds a private AI character that resembles the person in your reference photos. Training takes about 30 minutes. It only ever runs for you.
3
Generate up to 16 photos at once
Pick a scene — quiet kitchen, sunlit beach, anywhere — and generate photorealistic images in about 23 seconds each. Download the ones that feel right.
What makes a good reference set
3-16 photos from one era of the person\'s life (or one period of an adult animal\'s life).
Mix of close-ups, medium shots, and at least 2-3 full-body shots.
Different angles, expressions, lighting, and settings.
No sunglasses, hats that shadow the face, or group photos where the subject is small in frame.
High resolution (300+ DPI for scans, native iPhone photo quality or better).
JPG, PNG, WebP, or AVIF — HEIC files must be converted first.
See how to prepare reference photos for the full guide
(originally written for memorial use, but applies to all character types).
Privacy and data handling
Reference photos and trained characters are:
Encrypted at rest (AES-256) and in transit (TLS 1.3).
Scoped to a single user account — nobody else can access your character.
Never used to train shared, public, or improved models.
Deletable on request at any time.
Permanently removed within 30 days of account closure.
Refusals
Character training will be refused for reference photos that depict:
Flux is a photorealistic image-generation model by Black Forest Labs. Lover Snap licenses it for AI character creation and scene rendering. It is the same family of models used by professional studios for high-fidelity AI imagery.
How long does character training take?+
About 30 minutes per character on Lover Snap's infrastructure. The app sends a notification when training finishes; you can close the app while it runs.
How many reference photos do I need?+
3-16 photos. Variety matters more than count — include close-ups, medium shots, and at least 2-3 full-body shots from different angles and lighting.
Can the same character be reused across many photos?+
Yes — once trained, a character can be used in unlimited generations. There is no per-generation training cost.
Is the trained character shared with anyone?+
No. AI characters are private to your account, encrypted at rest, and never used to train shared models. Deletion is supported anytime.
