Envisage: FLUX.1-fill inpainting for nose job visualization with SurgicalScore metric
Outperforms existing editing methods on N=211 noses with a dedicated evaluation protocol.
Mudit Agarwal and Amit D. Bhrany introduce Envisage, a rhinoplasty goal visualization pipeline built on FLUX.1-Fill inpainting. It uses 8 clinical presets and hard-mask compositing to edit just the nose from a single frontal photo. Because full-face identity metrics fail for localized edits, they propose SurgicalScore, a mask-decomposed 0-1 scoring system. On 211 cases, Envisage achieved the highest SurgicalScore (0.599) and the smallest ArcFace identity loss (-0.048 vs -0.139 for ICEdit). The work includes blepharoplasty and rhytidectomy presets.
- Uses FLUX.1-Fill inpainting with 8 rhinoplasty presets and hard-mask compositing to edit only the nose region.
- Introduces SurgicalScore, a mask-decomposed 0-1 metric that scores edit direction, magnitude, masked LPIPS, realism, and outside-mask preservation.
- Achieves highest SurgicalScore (0.599 on N=211) and smallest ArcFace identity loss (-0.048) vs. ICEdit, Kontext, and InstructPix2Pix.
Why It Matters
Enables accurate, metric-driven AI visualization for cosmetic surgery planning, replacing misleading full-face identity scores.