AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Your Latent Mask is Wrong: Pixel-Equivalent Latent Compositing for Diffusion Models

Rowan Bradbury, Dazhi Zhong

CVPR 2026 · Image generation › Autoregressive and latent generation · approach: High-resolution diffusion synthesis · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace heuristic latent blending with DecFormer (soft masks, 1024px)FLUX VAE compositing · Halo L1Halo L1: 0.050 -> 0.018Improved
replace heuristic blending with DecFormer (binary masks)FLUX VAE compositing · PSNRdB: 28.4 -> 35.7Improved
replace heuristic blending with DecFormer inside FLUX.1-Dev inpaintingFLUX.1-Dev · COCO-2017 val inpaintingFID: 23.514 -> 20.556Improved
add inpainting LoRA on top of DecFormerFLUX.1-Dev + DecFormer · COCO-2017 val inpaintingFID: 20.556 -> 19.280Improved
DecFormer+LoRA vs fully finetuned Flux FillFLUX.1-Dev · COCO-2017 val inpaintingPSNR: 14.231 vs 16.750Hurt
add halo-weighted L1 loss (ablation)DecFormer · network ablationHalo L1: 0.0973 -> 0.0829Improved
add residual shift head alongside alpha (ablation)DecFormer · network ablationLPIPS: 0.0514 -> 0.0303Improved
add halo loss (ablation)DecFormer · network ablationLPIPS/MSE: LPIPS 0.0299 vs 0.0303No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy