One-node ComfyUI workflow generates $500 color analysis boards from a single portrait
A GitHub workflow using gpt-image-2-image-to-image generates Korean-style color analysis editorial boards from a single portrait—undertone, makeup palette, capsule wardrobe, and legible typography—without the multi-step fumbles of standard pipelines.
A ComfyUI workflow posted this week turns a single portrait into a 4K Dior-style color analysis board—the kind TikTok users fly to Seoul or pay NYC stylists $300–500 an hour to produce. The workflow ships as a one-node JSON file on GitHub and uses a single edit model to hold the reference face while rendering dozens of small labeled panels (undertone strip, makeup grid, capsule wardrobe swatches) with legible magazine typography.
The creator spent two days trying standard ComfyUI pipelines and hit three blockers: flux-dev "color analysis board for this person" returned Pinterest moodboards of unrelated stock photos, nano-banana-edit kept the face but rendered palette swatches as blurred rectangles with hallucinated hex codes, and anything below 1K resolution made the small magazine-style labels unreadable. The winning stack is MuAPIImageToImage with model gpt-image-2-image-to-image, a very specific aesthetic anchor prompt, and 3840×2160 output.
Model choice is load-bearing. The workflow uses gpt-image-2-image-to-image because it's the only edit model tested that holds the reference identity and renders dozens of small legible labels in the same image without text drift. Flux Kontext gets the face but garbles text; nano-banana gets text but loses the face.
Two prompt phrases control the entire vibe. The aesthetic anchor is "high-end editorial Color Analysis Board in a luxury fashion magazine style (Dior / Ralph Lauren aesthetic), clean beige/ivory background, minimal elegant typography, grid-based layout." Without "Dior / Ralph Lauren" the model defaults to scrapbook Pinterest energy with mismatched fonts. Without "grid-based layout" you get a single hero panel instead of the 8-panel magazine spread.
4K is non-negotiable. The workflow outputs at 3840×2160 (wired into extra_params_json). The board has 8+ small labeled panels—swatches, undertone strip, makeup grid, capsule wardrobe—and at 1024 resolution the labels under each swatch turn to mush.
One node, one shot. The entire workflow is a single MuAPIImageToImage node. Drop in a portrait, get back a full editorial board with best colors, undertone, makeup guide, hair, jewelry, and capsule wardrobe in one generation. The workflow JSON is public at .
