Persistent digital likeness for generative media.

A reusable likeness model for every real person and synthetic character — placed into any scene, edited without drift, working with any foundation model.

For people & creators Try Phota Studio

For developers & platforms Read the API docs

Backed by

a16z
Figma Ventures
Essence
& more

The technology

One likeness model. Any foundation model.

Phota adds a persistent likeness layer that carries identity across the image models you already use.

Supported base models OpenAI Nano Banana Qwen Grok Ideogram FLUX 1.1

Any scene, any edit

Placed into any scene and edited without losing identity.

Reusable everywhere

One likeness carries across campaigns, avatars, and content.

Built for production

Large volumes, multi-subject scenes, the same person across contexts.

Solutions

Built for both sides of the problem.

Personal

For everyday consumers — your own likeness, everywhere

AI Headshots The same person, consistent across every style

Avatar Platforms The persistent-likeness layer avatar platforms build on

Photo Editing Edit anything without changing how people look

Commercial

For studios, brands & platforms — reusable characters at scale

Storyboards Consistent characters across every frame

Video Keyframing Likeness-locked keyframes for AI video

Explore all solutions

What people say

When likeness finally gets it right, people feel it.

This is the first time I've had a good image of myself.

Photographer

Good enough only works until someone builds exactly right.

Technology enthusiast

Your subjects are much better than Nano Banana. How did you do it?

Researcher · Google DeepMind

Busy mom here. I never have good photos of myself . This was a real boost to my self-esteem.

Mom

Haven't seen a model capture my dark skin tone this accurately.

Creator

No more excuses not to become an IG influencer.

Creator · Instagram

The comparison

Why not a foundation model? Why not LoRA?

Dimension	Phota	Foundation model alone e.g. Nano Banana, GPT-Image	LoRA fine-tuning
01 Quality	Frontier-model quality + identity control	Great quality, but identity drifts	Open models only; quality and base capabilities can degrade
02 Multi-subject support	Compose any number of subjects with reliable identity	Identity quality degrades further with multiple subjects	No robust way to combine multiple LoRAs
03 Total cost	Train once per identity in ~1–3 minutes No per-base-model retraining No adapter serving	No additional training cost Identity misses paid for in repeated generation	Increases with subjects × base models Every adapter must be stored, deployed, and served
04 Persistence	One model per subject Stable across scenes, edits, and base models	No persistent memory of the subject Identity is re-referenced on every request	Persistent only within one model Model upgrades makes the identity obsolete
05 Flexibility	Trains from 5+ casual photos	Reference images required for every request	20–30+ curated, captioned images per subject

For enterprise, teams & platforms

Persistent likeness, priced for your scale.

Studios, marketing teams, avatar platforms, consumer photo apps — if likeness is critical , we offer custom enterprise plans: volume pricing, dedicated support and deployments. Tell us what you're building.