SCENES & CONTINUITY · SOUL ID · 6 MIN READ · UPDATED AUGUST 2026

Soul ID Explained

Soul ID is ManifoldGen's identity layer: one locked character spec — anchors, descriptor, derivation rules — reused everywhere so every output inherits the same face.

Soul ID is the name we give to the practice the consistency guides teach: identity lives in assets you own, not in prompts you retype. A character has a Soul ID when anyone on the team can produce a new image of them without guessing — because the face of record and the rules for deriving from it are written down.

The three parts

  • Anchor set — the canonical sheet: frontal, three-quarter, full-body portraits at high fidelity. The face of record.
  • Descriptor block — the frozen identity paragraph, pasted verbatim into every brief. Never paraphrased, versioned like code.
  • Derivation rules — how new images may legally be made: reference edit from anchors; masked inpaint for repairs; never a text-only regeneration of the face.

How it flows through the tools

  • Soul Moodboard fuses anchor references with style references into one visual direction for the project.
  • Nano Banana 2 / GPT Image 2 reference editing moves the anchored face into new scenes — identity rides with the pixels.
  • H3 Image Editor (REF2VA) regenerates a scene while preserving identity and geometry, for heavier recontextualization.
  • Inpaint repairs drifted regions while the mask protects approved pixels.
  • Relight adapts anchors to scene lighting without touching facial structure.
  • Character Animator carries identity through motion driven by a performance clip.
  • Image Upscale finishes selects for delivery.

What Soul ID is not

  • Not a magic one-click clone — the anchor set still needs to be built deliberately.
  • Not seed locking — seeds die at every route, model, and resolution change.
  • Not style transfer — style is how the world renders; identity is who is standing in it. They are separate layers.

Setting one up takes ten minutes

  1. 01Pick or produce the anchor portraits (photo-to-persona flow works for real faces).
  2. 02Write the descriptor block; save it next to the images.
  3. 03Test derivation: three scenes via reference edit, audited against the anchors.
  4. 04Freeze v1. Assets plus descriptor live in one folder; nothing generates outside the rules.
The rule of thumb

If a new image cannot be traced back to an anchor, it is not your character. Traceability is the product — every output should be able to answer "which anchor did you come from?"

The same pattern scales beyond people: give a location an anchor set and derivation rules and you have Soul ID for sets — which is exactly what sequence work demands (see Characters and Locations Across Shots).

Put the workflow to work.

The Studio has the reference editing, relighting, and batch tools these guides assume.

Open Studio