Training a material LoRA for Qwen Image Edit 2511: 30 pairs still miss the fabric
Qwesbrz · reddit · 2026-09-17
A user documents training a material-swap LoRA for Qwen Image Edit 2511: 30 input/target pairs from 10 Velvet 506 furniture photos, full config (1024px, 300 steps, dim 16, AdamW8bit, lr 5e-5). Results get the color right but miss fabric texture, folds and stitching, and leak into areas that shouldn't change. Full settings and failure cases included.
More from Multimodal
- Midjourney wallpaper recipe: sref + profile mix for eerie film-blur portraits — michaelrabone · 2026-11-02
- Cavallaero Teaser: A Painterly Rider Races Through a Golden Forest — CommunicationTop4244 · 2026-11-02
- Anima's snapref Offers a New Reference-Based Fix for Character Consistency — Intrepid_Swim_8669 · 2026-11-02
- Pencil-textured folding fan: surreal product concept render goes viral — Delahuntagram · 2026-11-02
- Free Midjourney SREF gallery adds 1,200+ handpicked style codes overnight — tisch_eins · 2026-11-02
- Adobe's EditAgent beats prior image-editing agents but needs ~10x the compute — emmanuelvivier · 2026-11-02