Research map / Image generation / Autoregressive and latent generation
∞-Brush: Controllable Large Image Synthesis with Diffusion Models in Infinite Dimensions
Minh-Quan Le, Alexandros Graikos, Srikar Yellapragada, Rajarsi Gupta, Joel Saltz, Dimitris Samaras
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add cross-attention neural operator instead of vanilla grid cross-attention (ablation) | ∞-Brush · TCGA-BRCA 1.25x 4096x4096 | CLIP FID: 3.81 -> 2.63 | Improved |
| add cross-attention neural operator instead of vanilla grid cross-attention (ablation) | ∞-Brush · TCGA-BRCA 1.25x 4096x4096 | Crop FID: 16.28 -> 14.76 | Improved |
| use function-space conditional diffusion instead of patch-based generation | very large image synthesis · TCGA-BRCA 1.25x 4096x4096 | CLIP FID: 2.75 -> 2.63 | Improved |
| use function-space conditional diffusion instead of patch-based generation | very large image synthesis · TCGA-BRCA 1.25x 4096x4096 | Crop FID: 11.30 -> 14.76 | Hurt |
| use ∞-Brush instead of SDXL | large image synthesis · TCGA-BRCA 5x 1024x1024 | CLIP FID: 6.64 -> 3.74 | Improved |
| use ∞-Brush instead of SDXL | large image synthesis · TCGA-BRCA 5x 1024x1024 | Crop FID: 6.98 -> 17.87 | Hurt |
| use ∞-Brush instead of patch-based method | large image synthesis · NAIP 1024x1024 | Crop FID: 43.76 -> 48.65 | Hurt |
| add conditioning to infinite-dimensional diffusion model | ∞-Diff · CelebA-HQ | CLIP FID: 9.44 -> 8.38 | Improved |
| train on 0.4% pixel subsets of full images instead of full-resolution or patch training | ∞-Brush · training cost | epoch time (hr): 12 vs 300 (patch-based at 4096) | Improved |
Closest papers by meaning
- DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 2025 · similarity 0.89)
- FAM Diffusion: Frequency and Attention Modulation for High-Resolution Image Generation with Stable Diffusion (CVPR 2025 · similarity 0.89)
- ScaleDiff: Higher-Resolution Image Synthesis via Efficient and Model-Agnostic Diffusion (NeurIPS 2025 · similarity 0.89)
- ZoomLDM: Latent Diffusion Model for Multi-scale Image Generation (CVPR 2025 · similarity 0.89)
- Simpler Diffusion: 1.5 FID on ImageNet512 with Pixel-space Diffusion (CVPR 2025 · similarity 0.88)
- A Simple Approach to Unifying Diffusion-based Conditional Generation (ICLR 2025 · similarity 0.88)
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.
