ComfyUI lab 10: Restyle without losing structure

Style transfer fails when “preserve structure” has no technical meaning. This graph defines it as Canny edges: a compact map of intensity boundaries. Proteus then interprets the prompt under that constraint.

Understand what Canny preserves

Edges describe contours, not identity, depth, material, or color. A strong Canny map can preserve architecture and pose while still allowing every surface to change. It can also preserve unwanted texture, hair strands, or background clutter. Blur, crop, or simplify the input when those edges are not part of the desired composition.

Run an input sweep at fixed prompt and seed. At strength 0, the reference should stop controlling structure. Near 0.6, large boundaries should remain while the prompt changes appearance. At very high strength, results may become traced or brittle. The useful range depends on edge density.

Compact node versus explicit graph

The custom Proteus node is convenient and open, but it hides model loading and edge extraction inside one box. The existing SDXL ControlNet gallery graph exposes those operations as core nodes. Use the compact graph for a known pipeline; use the expanded graph for teaching, swapping models, or debugging. Abstraction is useful only when you can still identify what it contains.

Run the lab

• Open this graph in Comfy Studio
• Download the exact API-format workflow
• CLI: app comfy download proteus-style-node

Record the graph JSON, input assets, seed where relevant, model or service version, custom-node commit, and final artifact together. That bundle—not a screenshot—is the reproducible creative unit.

Open-source map

• Open Proteus style node — MIT compact SDXL ControlNet integration
• Proteus V0.4 — Model card, weights, and model license

Creative Systems Labs: 10 of 12. Previous: Control Z-Image with line structure Next: Design a text-to-3D handoff