Qwen Image 2.1 masked inpainting: working crop-and-stitch graph, wiring and prompting differences from Flux
Reasonable_Arm7239 · reddit · 2026-09-24
The author shares a full migration of their Flux inpainting workflow to Qwen Image 2.1. Key wiring: the cropped image goes into image1 on TextEncodeQwenImage21 (the VL encoder itself is the reference mechanism, no ReferenceLatent), and the sampler latent must come from InpaintModelConditioning. Prompting is the bigger change — 2.1 is an instruction follower, so "replace the masked object with X" beats Flux-style descriptions. Also covers negative prompts (live at CFG >1), crop-and-stitch params (maskblend 32, context factor 1.25), addressable multi-image references ("match the face in image 2"), bf16 fitting on 24GB, a research/non-commercial license, and an attached JSON workflow.
Related event: Migrating Flux Inpainting Workflows to Qwen Image 2.1: A Hands-On Guide(2 posts)→
More from Multimodal
- MiniMax H3 hits 100 days open-source: 1,472 derivative models, 148M+ downloads — linoy_tsaban · 2026-11-11
- MiniMax H3-Image explodes an espresso machine photo into 15 labeled parts — Charlietest123 · 2026-11-11
- Indie dev built a tool that clips best moments from 4-hour streams in 30 minutes — kissaadams · 2026-11-11
- Non-Blender user makes a video in 30 minutes with Claude and Seedance — FinanceYF5 · 2026-11-11
- Whimsical Woodland: A Watercolor AI Art Prompt Template for Dreamy Forest Scenes — LudovicCreator · 2026-11-11
- Clean point clouds: Tersus GNSS MVP S1 scans turned into high-quality splats in LichtFeld Studio — janusch_patas · 2026-11-11