Blog/From Synthetic Portrait to Live Browser Identity: The Latentface Workflow
Product guidePillar6 min read

From Synthetic Portrait to Live Browser Identity: The Latentface Workflow

Latentface Team·

A practical guide to choosing a model-generated portrait, starting the camera only when you decide, and carrying that identity into calls or broadcasts.

What Latentface does today

Latentface is a browser-based realtime face-swap workspace. Its first screen is a field of six pre-generated synthetic portraits. None depicts a real person, and the site does not generate a new person while you browse the field.

Choose a point, confirm the portrait, then decide separately whether to start the camera. That second action is intentional: inspecting an identity should never silently request camera permission.

The complete workflow

  1. Find an identity. Click or tap a point in the field. Keyboard users can move between anchors with the arrow and bracket keys.
  2. Confirm the portrait. “Choose this face” fixes the model-generated portrait but keeps the realtime engine and camera unmounted.
  3. Start live explicitly. “Wear this face live” opens the realtime workspace. The shared camera consent screen remains the authority for permission.
  4. Change without reconnecting. The field stays available from the instrument rail. Moving to the previous or next anchor updates the target portrait while leaving the session intact.
  5. Choose an output. Use the result in the browser, or open Send to create the browser-source URL used by OBS Window Capture. The URL is minted only when you ask for it and expires after 12 hours.

When a handoff fails

Latentface keeps the camera session separate from the portrait handoff. If a portrait asset cannot load, the browser refuses the file transfer, or the shared picker has not mounted yet, the instrument names that exact boundary and tells you the next safe action. It does not claim that a face was applied when it was not.

The shared realtime surface separately reports camera denial, a missing or busy camera, connection failure, queue state, session expiry, and payment boundaries. A failed portrait change does not tear down an otherwise healthy session.

Choosing a useful source portrait

The built-in field is the quickest route because its portraits have already been selected as clean synthetic swap targets. You can also skip the field and use the shared picker for your own image. For a stable result, use a front-facing, well-lit portrait with one unobstructed adult face. Avoid group photos, heavy sunglasses, and extreme side angles.

Only use images you own or have permission to use. Latentface's built-in portraits are synthetic specifically so the default journey does not depend on a celebrity or another real person's likeness.

What is not part of the product

Latentface does not currently sell a public REST API, API keys, Python or TypeScript SDKs, per-call metering, or a face-generation endpoint. Older developer-console plans were retired when the brand became a consumer realtime product. Current product claims live on the browser workflow above and are tested against the public site.

Try the instrument

Open the Latentface field and choose an identity before your camera is requested.

Try Latentface today

Every face here was generated by a model — none is real. Find one in the latent field and wear it live. Real-time face swap in your browser, no download.