SyntheticPic runs on forge:identity-*, the skill that trains a per-person Flux LoRA — a small model fine-tuned on one person's face from a handful of their own uploaded selfies, rather than a single shared model everyone draws from. Training happens once per account; after that, the same model can be reused to generate new photorealistic portraits on demand, in different framing, lighting, or settings, all recognizably the same person because the model was built specifically for that face. Nothing crosses accounts — each user's model, and everything generated from it, is private to them, so no one else's uploads or outputs can surface in, or influence, another user's results.
The prompt
No intake brief on record: SyntheticPic was built engine-direct, bypassing the standard brief-and-intake flow, and tracked internally as engine run 271.
What it does
- Trains a private per-user face model from a handful of uploaded selfies (a Flux LoRA, via the forge:identity-* skill)
- Generates unlimited identity-locked, photorealistic portraits from that model
- No shared model — every user's face model is private to their account
- Model trains once per account and is reusable for unlimited future generations
- Regenerating a new batch never requires re-uploading selfies once the model is trained
Built in 23 minutes, prompt to live deploy.