Research map / Image generation / Text-to-image diffusion
IMG: Calibrating Diffusion Models via Implicit Multimodal Guidance
Jiayi Guo, Chuanhao Yan, Xingqian Xu, Yulin Wang, Kai Wang, Gao Huang, Humphrey Shi
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 IMG re-generation to base diffusion model | SDXL · HPD average | HPS: 28.03 -> 28.56 | Improved |
| add IMG on top of the finetuned model | SDXL-DPO · HPD average | HPS: 28.39 -> 28.79 | Improved |
| add IMG | FLUX · HPD average | HPS: 29.21 -> 29.48 | Improved |
| add SLD editing-based alignment | SDXL · HPD average | HPS: 28.03 -> 27.35 | Hurt |
| add IMG on top of FLUX for compositional prompts | FLUX · T2I-CompBench Color | score: 0.7852 -> 0.8060 | Improved |
| use text embedding instead of MLLM features to guide the aligner (ablation) | IMG Implicit Aligner · Pick-a-Pic test | Pick Score: 22.34 -> 22.38 (vs base model) | No clear effect |
| use original MLLM features rather than no guidance (ablation) | IMG Implicit Aligner · Pick-a-Pic test | Pick Score: 22.22 -> 22.59 | Improved |
| finetune the MLLM on Instruct-Pix2Pix editing data (ablation) | IMG MLLM · Pick-a-Pic test | Pick Score: 22.59 -> 22.74 | Improved |
| finetune the MLLM on editing instruction task (ablation) | LLaVA 1.5-13b · SEED-Data-Edit 300 cases | yes rate: 42.3% -> 76.3% | Improved |
| set loss ratio lambda to 1 rather than 0, 0.5 or 2 (ablation) | IMG objective · Pick-a-Pic test | Pick Score: 22.56 -> 22.74 (lambda 0 vs 1); 2 gives 22.50 | Improved |
| add DPO component then SPIN component to preference loss (ablation) | IMG objective · Pick-a-Pic test | Pick Score: 22.56 -> 22.65 -> 22.74 | Improved |
| swap LLaVA for Qwen2.5-VL-7B as MLLM (ablation) | IMG · HPD average | HPS: SDXL 28.56 -> 28.48; FLUX 29.48 -> 29.52 | No clear effect |
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.
- SDXL base — SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis (ICLR 2024) · 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
- I Think, Therefore I Diffuse: Enabling Multimodal In-Context Reasoning in Diffusion Models (ICML 2025 · similarity 0.89)
- DyMO: Training-Free Diffusion Model Alignment with Dynamic Multi-Objective Scheduling (CVPR 2025 · similarity 0.88)
- Towards Transformer-Based Aligned Generation with Self-Coherence Guidance (CVPR 2025 · similarity 0.88)
- Aligning Diffusion Models by Optimizing Human Utility (NeurIPS 2024 · similarity 0.88)
- Towards Better Alignment: Training Diffusion Models with Reinforcement Learning Against Sparse Rewards (CVPR 2025 · similarity 0.88)
- Noise Diffusion for Enhancing Semantic Faithfulness in Text-to-Image Synthesis (CVPR 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.
