Phota AI: Persistent Identity Profiles for Personalized Image Generation (2026)
Last updated August 7, 2026
Phota (Phota Labs) is not a generator — it is a persistent-identity layer that trains a private profile from 5–50 photos of a person or pet, referenced via [[profile_id]] in prompts, and keeps that likeness consistent across Nano Banana 2, GPT Image 2, Qwen Image, FLUX 2 Dev, and Reve. Training costs $1.49–2.90; images $0.02–0.20. Available inside invideo.
Updated August 2026
Phota is not an image generator — it is a persistent-identity layer that sits on top of them. Built by Phota Labs (photalabs.com), it trains a private, reusable profile of a person (or pet) from as few as 5 photos, then keeps that likeness consistent across generations and edits on whichever foundation model you route the request to — Nano Banana 2, GPT Image 2, Qwen Image, FLUX 2 Dev, or Reve. Prompts reference a profile with [[profile_id]] syntax, and the identity carries across its four operations — Edit, Generate, Enhance, and Remix. Profile training costs $2.90 (Full) or $1.49 (Quick), and images run $0.02–0.20 depending on the base model and whether a trained identity appears in the output (official pricing docs, August 2026). In invideo's model picker it appears as Phota — personalized photo generation and editing with profile-based identity control.
What is Phota AI and who makes it?
Phota Labs is a startup whose product is "persistent digital likeness": train a profile once, and Phota applies it across interchangeable third-party generators instead of losing that work each time a better model ships. The company states it is backed by a16z and Figma Ventures — a claim from its own site, not independent coverage. It ships a browser tool (Phota Studio at studio.photalabs.com) and a production API documented at docs.photalabs.com.
Since searches for "phota ai" surface both: photalabs.com is the actual company; phota.ai is an unrelated parked "launching soon" page as of August 2026.
How do Phota identity profiles work?
The workflow has three steps:
- Train a profile from 5–50 photos of the subject: Quick Train ($1.49) uses 5–10 images, Full Train ($2.90) uses 10–50 for highest fidelity — two tiers that likely explain why the company's pages quote both "1–3 minute" and "~15 minute" training times.
- Reference it in prompts using
[[profile_id]]— e.g., "[[maya_profile]] presenting on a conference stage." Supported subjects are people and pets (cats and dogs), per the official docs. - Pick a base model per request. The identity layer is model-agnostic: the same profile renders through any supported base model — Nano Banana 2 is the default — with no per-model retraining.
Profiles are private, and Phota Studio states they are "never used for public model training."
What are the four Phota operations?
| Operation | What it does | When to use it |
|---|---|---|
| Edit | Identity-preserving edits to an existing photo | Change outfit, background, or lighting without the face drifting |
| Generate | New scenes featuring the trained likeness | Content the camera never captured — new settings, poses, scenarios |
| Enhance | Quality fixes on existing images | Restoring or sharpening real photos |
| Remix | Style transfer from a reference image | Recasting the subject in a specific aesthetic |
All four accept [[profile_id]] references: one trained identity, four ways to deploy it.
How much does Phota cost?
Official rates from docs.photalabs.com/pricing.md (August 2026); per-image pricing has a base rate and a higher with-identity rate when a trained profile appears in the output:
| Item | Base price | With trained identity |
|---|---|---|
| Full Train (10–50 images) | $2.90 one-time | — |
| Quick Train (5–10 images) | $1.49 one-time | — |
| Nano Banana 2, 1K (default) | $0.08 | $0.10 |
| Nano Banana 2, 4K | $0.16 | $0.20 |
| Qwen Image | $0.04 | $0.05 |
| FLUX 2 Dev, 1K | $0.02 | $0.03 |
| Reve, 1K | $0.04 | $0.05 |
| Enhance / Remix, per output | $0.12 | $0.15 |
GPT Image 2 requests are token-metered rather than flat-priced. The spread matters: identity-locked output on FLUX 2 Dev costs $0.03 an image, so a 100-image consistency test costs about $3 plus one $1.49 Quick Train.
What is Phota good for?
The common thread is any workflow where the same face must survive dozens of generations:
- Headshots and profile imagery. Train once, then generate polished variations in new settings and outfits — the core AI headshot job, without per-shoot retraining.
- Creator avatars and thumbnails. A persistent likeness across weekly thumbnails and channel art is exactly the consistency problem AI avatar workflows hit; Phota solves it at the identity layer rather than per image.
- Recurring characters in brand content. Phota Labs' own listed commercial use cases include ad creative, storyboards with character consistency, and identity-locked keyframes for AI video — the core need of serialized character generation.
What are Phota's limitations?
An honest read, as of August 2026:
- It is a young company with no published benchmarks. There are no third-party arena placements or identity-fidelity metrics to cite; quality claims rest on the company's own demos.
- Output quality is bounded by the base models. Phota controls who is in the image; how well it is rendered is the underlying generator's job.
- Commercial licensing terms are not published — enterprise use runs through direct contact, so review terms before client work.
- Likeness rights are on you. Training a profile of a real person should only happen with their consent; the tooling does not police whose photos you upload.
Questions people ask about Phota
Is Phota an image generation model?
No — it is an identity layer that routes to third-party foundation models (Nano Banana 2, GPT Image 2, Qwen Image, FLUX 2 Dev, Reve). Its job is keeping a trained likeness consistent across whichever model renders the image.
How many photos does Phota need to train a profile?
A minimum of 5 and up to 50 for highest fidelity: Quick Train uses 5–10 images ($1.49), Full Train uses 10–50 ($2.90), per the official docs, August 2026.
Can Phota create images of pets?
Yes — profiles support people and pets (cats and dogs), using the same [[profile_id]] prompt syntax.
Are my photos used to train Phota's models?
Phota Studio states profiles are private and "never used for public model training" — training photos build your own profile only (studio.photalabs.com, August 2026).
How much does a Phota image cost?
Between $0.02 (FLUX 2 Dev 1K, base) and $0.20 (Nano Banana 2 4K with identity) per image, depending on base model, resolution, and identity use — plus the one-time training fee (pricing docs).
Is Phota the same as phota.ai?
No. The company behind the identity technology is Phota Labs at photalabs.com; phota.ai is an unrelated parked page as of August 2026.
Where can you use Phota?
Phota is available inside the invideo agent, where the picker lists it as "Phota — personalized photo generation and editing with profile-based identity control." That placement fits how the platform works: invideo already runs the foundation models Phota routes to, so a profile-locked likeness can anchor a project's stills while the agent's 200+ models handle everything around them — and identity-consistent keyframes are exactly what character-driven video sequences need to start from.
Version history: Phota Labs launched Studio and its production API in 2026; the company publishes no dated changelog, and the details above reflect the official docs as of August 2026.