Research map / Image generation / Text-to-image diffusion
Exploring Multimodal Diffusion Transformers for Enhanced Prompt-based Image Editing
Joonghyuk Shin, Alchan Hwang, Yujin Kim, Daneul Kim, Jaesik Park
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 |
|---|---|---|---|
| edit via image-projection replacement instead of fixed-seed regeneration | MM-DiT editing (SD3-M) · PARTI 60 prompts | LPIPS: 0.594 -> 0.380 | Improved |
| edit via image-projection replacement instead of fixed-seed regeneration | MM-DiT editing (Flux.1-schnell) · PARTI 60 prompts | LPIPS: 0.508 -> 0.279 | Improved |
| edit via image-projection replacement instead of prompt change at 20% timesteps | MM-DiT editing (Flux.1-dev) · PARTI 60 prompts | CLIP score: 0.311 -> 0.339 | Improved |
| edit via image-projection replacement instead of prompt change at 20% timesteps | MM-DiT editing (Flux.1-schnell) · PARTI 60 prompts | CLIP score: 0.289 -> 0.333 | Improved |
| edit via image-projection replacement instead of prompt change | MM-DiT editing (SD3-M) · PARTI 60 prompts | LPIPS: 0.325 -> 0.380 | Hurt |
| replace only q_i,k_i image projections instead of full attention map (ablation) | MM-DiT editing (Flux.1-dev) · text-misaligned prompt pairs | edit stability | Improved |
| use I2I block replacement instead of q_i,k_i projection replacement (ablation) | MM-DiT editing · computation time | speed: up to 3x | Hurt |
| add top-5 block selection for attention masks (ablation) | MM-DiT local blending (Flux.1-dev) · local editing | mask noise | Improved |
| add Gaussian smoothing to attention masks (ablation) | MM-DiT local blending · block-wise BCE, Soft mIoU, MSE | mask quality | Improved |
| use T2I blocks instead of I2T blocks for masks (ablation) | MM-DiT local blending · local editing | mask quality | Improved |
| add local blending with top-5 block masks (ablation) | MM-DiT editing (Flux.1-dev) · inference time | seconds: 55.9s vs. 53.7s | No clear effect |
| raise masking threshold theta from 0.2 to 0.5 (ablation) | Ours with RF inversion (Flux.1-dev) · PIE-Bench | PSNR: 19.92 -> 24.79 | Improved |
Closest papers by meaning
- Rare Text Semantics Were Always There in Your Diffusion Transformer (NeurIPS 2025 · similarity 0.90)
- Stable Flow: Vital Layers for Training-Free Image Editing (CVPR 2025 · similarity 0.90)
- Towards Efficient Diffusion-Based Image Editing with Instant Attention Masks (AAAI 2024 · similarity 0.90)
- Towards Understanding Cross and Self-Attention in Stable Diffusion for Text-Guided Image Editing (CVPR 2024 · similarity 0.89)
- Textualize Visual Prompt for Image Editing via Diffusion Bridge (AAAI 2025 · similarity 0.89)
- An Item Is Worth a Prompt: Versatile Image Editing with Disentangled Control (AAAI 2025 · similarity 0.89)
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.
