Smart Figure Detailer

AI Workflow / 2026
ComfyUI · SAM · Detailer
Role — Concept, Development & Visualization

In a wide architectural rendering, the scale-figures—the people that give a scene life—often occupy only a few hundred pixels. That is too little information for a diffusion model to resolve a believable face, hands, or the fall of fabric, which is why AI visualizations so often betray themselves through soft, slightly melted figures.

Smart Figure Detailer is a computational workflow built to fix exactly that. A segmentation model detects every person in the frame and isolates them as clean masks. Each figure is cropped, upscaled to a proper working resolution, and passed through a detail pass that rebuilds facial features, clothing folds, and contact shadows—then composited back into the original image. The building, lighting, and composition stay untouched; only the people are quietly brought up to standard.

01
Detect & Isolate A Segment Anything (SAM) pass finds each figure and outputs a binary mask, separating people from the scene.
02
Upscale The masked figures are cropped and upscaled, giving the model enough resolution to work with—sharper, but not yet refined.
03
Enhance A detail pass rebuilds faces, fabric, and shadows at full fidelity, then composites the result back into the render.