AI knows what you look like.
Helix knows who you are.
A persistent, user-owned identity layer that holds your face, your family, your pets — and the relationships between them. Train once. Be recognized everywhere.
The problem · two layers
AI image models are stateless with respect to who you are.
Every AI app you use makes you start from zero — upload your photos, wait for training, hope the output looks like you. As more apps emerge that need to generate content featuring the people you love, the cost of repeated re-onboarding compounds.
And the deeper problem is worse: even when the face is structurally correct, it usually isn't you.
Onboarding friction.
Every AI app re-trains identity from scratch. Users bounce at the upload step. Builders pay full compute every time.
Expressional identity drift.
Current identity-preservation tech — LoRA, IP-Adapter, InstantID, even frontier models — optimizes for structuralsimilarity: bone structure, eye spacing, hair, skin tone. The people who love you don't recognize you by your bone structure.
The recognition test
“That's not her smile in photo three.”
Recognition isn't a benchmark. It's a wife looking at an AI-generated photo of her daughter and knowing — instantly, pre-verbally — that something is off.
The face was structurally correct. The identity was wrong. Helix exists to close that gap: the smile, the micro-expressions, the way one eyebrow goes up first, the specific asymmetry of a laugh.
The missing layer
One identity. Every app. Owned by you.
Helix sits between you and the AI apps you use. Once your identity — and your family's — is trained, any compliant app can request it. No re-uploads. No drift. No starting over.
Build your identity once.
Upload photos of you, your family, your pets. Helix learns not just faces, but expressions, asymmetries, and the relationships between everyone.
Own it. Permission it.
Your identity vault lives with you. Grant any AI app scoped, revocable access — like an OAuth token, but for who you are.
Be recognized, everywhere.
When an app generates with your Helix ID, the output is expressionally you — the smile your kid would point to. Across every app you ever use.
Stop uploading the same photos to every new AI app.
Your face, your family, your pets — held in one identity vault you control.
- Train your Helix ID once. Use it forever.
- Grant per-app access. Revoke anytime.
- Outputs that your family actually recognizes.
Drop in identity. Skip the training step entirely.
A request away from a fully-trained identity — instead of another LoRA pipeline you have to maintain.
- One SDK call replaces your upload & training flow.
- No more bouncing at the upload step.
- Better outputs without your own ML investment.
A decentralized protocol for multi-subject asset consistency across every generative engine.
Helix isn't a single app — it's the open identity layer that lets any model (diffusion, video, 3D, agentic) and any application request a verified, permissioned identity for one person, a family, or an entire cast of subjects. Secure, cross-platform, engine-agnostic.
Decentralized
User-owned identity vaults. No single point of capture, no platform lock-in.
Cross-platform
OAuth-style scopes. Any compliant app or engine can request access.
Multi-subject
People, pets, relationships — bound together as one identity graph.
Engine-agnostic
Works with diffusion, transformer, video, 3D, and agentic models.
Permissioned
Per-app, per-purpose, per-duration grants. Revocable at any moment.
Verifiable
Cryptographically signed identity assertions across the ecosystem.
“The metric that matters isn't cosine similarity in face embedding space. It's whether the person who loves you recognizes you in the picture.”
— The Helix thesis
Build on the Helix protocol.
We're onboarding a small group of generative engines and application-layer teams to the protocol first. Tell us about your platform.