FaceRef.org

ControlNet Face Reference

Generate clean face angle references, Loomis or Asaro structure sketches, and depth-style control maps for Stable Diffusion, FLUX, and ComfyUI portraits.
Apr 10, 2026

Why this page exists

Most portrait inconsistency problems do not start with the prompt. They start because the model is guessing the head angle, facial structure, and lighting at the same time.

A clean ControlNet face reference reduces that randomness.

What FaceRef helps you lock in first

Use FaceRef to decide four things before you generate:

  1. the yaw of the head
  2. the pitch of the head
  3. the overall facial structure
  4. the main light direction

Once those are stable, the prompt can focus on style, material, and mood instead of rebuilding the face from scratch.

Loomis or Asaro: choose the kind of guidance you need

When you want clearer construction, start with a Loomis-style breakdown.

When you need stronger lighting logic and clearer planes, use an Asaro-style structure. This is especially useful when you want the conditioning image to preserve the cheek, brow, nose wedge, and jaw relationships under dramatic light.

Depth-style maps and line-driven maps

Different ControlNet workflows respond better to different kinds of simplification.

  • depth-style guidance is useful when volume matters most
  • line-driven guidance is useful when sharper structure matters most

The important point is that the exported image should preserve the geometry you do not want the model to reinvent.

A practical workflow

  1. set the head angle in FaceRef
  2. adjust the light until the shadow family is intentional
  3. export the structure image that best matches your workflow
  4. use that image as the conditioning layer for your portrait generation
  5. iterate the prompt while keeping the reference stable

Start with the workspace

Open the FaceRef workspace and build the head angle before you touch the prompt stack.