Research map / Image generation / Text-to-image diffusion
Linear Image Generation by Synthesizing Exposure Brackets
Yuekun Dai, Zhoutong Zhang, Shangchen Zhou, Nanxuan Zhao
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 |
|---|---|---|---|
| use 3D RoPE instead of 2D RoPE (ablation) | exposure bracket model · Adobe FiveK (positional encoding ablation) | LS: 4.10 -> 23.06 | Improved |
| add layer embedding to 3D RoPE (ablation) | 3D RoPE model · Adobe FiveK (positional encoding ablation) | LS: 23.06 -> 21.07 | No clear effect |
| add layer embedding to 2D RoPE (ablation) | 2D RoPE model · Adobe FiveK (positional encoding ablation) | LS: 4.10 -> 6.21 | Improved |
| use 4 brackets instead of 2 (ablation) | exposure bracket model · Adobe FiveK (number of brackets) | LS: 4.76 -> 23.06 | Improved |
| use 5 brackets instead of 4 (ablation) | exposure bracket model · Adobe FiveK (number of brackets) | CLIP Sim.: 26.02 -> 23.92 | Hurt |
| use 5 brackets instead of 4 (ablation) | exposure bracket model · Adobe FiveK (number of brackets) | LS: 23.06 -> 25.76 | Mixed |
| add LoRA (ablation) | Flux-based bracket model · Adobe FiveK (architecture ablation) | AS: 3.183 -> 5.700 | Improved |
| add exposure modulation module (ablation) | Flux-based bracket model · Adobe FiveK (architecture ablation) | LS: 2.13 -> 23.06 | Improved |
| add exposure modulation module (ablation) | Flux-based bracket model · Adobe FiveK (architecture ablation) | AS: 6.245 -> 5.700 | Hurt |
| apply modulation to single-DiT instead of MM-DiT (ablation) | Flux-based bracket model · Adobe FiveK (architecture ablation) | AS: 4.937 -> 5.700 | Improved |
| use exposure modulation self-attention instead of EV token injection (ablation) | bracket model · Adobe FiveK (EV injection ablation) | NIQE: 4.291 -> 3.658 | Improved |
| use exposure modulation self-attention instead of AdaLN injection (ablation) | bracket model · Adobe FiveK (EV injection ablation) | LS: 14.86 -> 23.06 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- LEDiff — LEDiff: Latent Exposure Diffusion for HDR Generation (CVPR 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CoDi: Conditional Diffusion Distillation for Higher-Fidelity and Faster Image Generation (CVPR 2024 · similarity 0.85)
- EditAR: Unified Conditional Generation with Autoregressive Models (CVPR 2025 · similarity 0.85)
- R&B: Region and Boundary Aware Zero-shot Grounded Text-to-image Generation (ICLR 2024 · similarity 0.85)
- Pathways on the Image Manifold: Image Editing via Video Generation (CVPR 2025 · similarity 0.84)
- Generative Photomontage (CVPR 2025 · similarity 0.84)
- A Simple Approach to Unifying Diffusion-based Conditional Generation (ICLR 2025 · similarity 0.84)
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.
