Research map / Image generation / Text-to-image diffusion
Disentangling to Re-couple: Resolving the Similarity-Controllability Paradox in Subject-Driven Text-to-Image Generation
Shuang Li, Chao Deng, Hang Chen, Liqun Liu, Zhenyu Hu, Te Cao, Mengge Xue, Yuan Chen, Peng Shu, Huan Yu, Jie Jiang
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 textual-visual decoupling (ablation) | DisCo · DreamBench | CLIP-I: 0.915 -> 0.928 | Improved |
| add textual-visual decoupling (ablation) | DisCo · DreamBench | CLIP-T: 0.319 -> 0.329 | Improved |
| add textual-visual decoupling (ablation) | DisCo · DreamBench | ImageReward: 1.237 -> 1.339 | Improved |
| add GRPO re-coupling stage (ablation) | DisCo · DreamBench | ImageReward: 1.189 -> 1.339 | Improved |
| add GRPO re-coupling stage (ablation) | DisCo · DreamBench | CLIP-I: 0.922 -> 0.928 | Improved |
| add GRPO re-coupling stage (ablation) | DisCo · DreamBench | CLIP-T: 0.319 -> 0.329 | Improved |
| use trained pairwise VLM reward instead of CLIP reward (ablation) | DisCo GRPO · DreamBench | CLIP-I: 0.898 -> 0.928 | Improved |
| use trained VLM reward instead of ImageReward reward (ablation) | DisCo GRPO · DreamBench | CLIP-I: 0.914 -> 0.928 | Improved |
| use trained VLM reward instead of ImageReward reward (ablation) | DisCo GRPO · DreamBench | ImageReward: 1.404 -> 1.339 | Hurt |
| use trained VLM reward instead of untuned pretrained VLM reward (ablation) | DisCo GRPO · DreamBench | ImageReward: 1.189 -> 1.339 | Improved |
| apply naive prompt decoupling to DreamO | DreamO · DreamBench | ImageReward: 1.186 -> 0.661 | Hurt |
| apply naive prompt decoupling to DreamO | DreamO · DreamBench | DINO-I: 0.813 -> 0.771 | Hurt |
Closest papers by meaning
- Large-Scale Text-to-Image Model with Inpainting is a Zero-Shot Subject-Driven Image Generator (CVPR 2025 · similarity 0.88)
- XVerse: Consistent Multi-Subject Control of Identity and Semantic Attributes via DiT Modulation (NeurIPS 2025 · similarity 0.88)
- One-Prompt-One-Story: Free-Lunch Consistent Text-to-Image Generation Using a Single Prompt (ICLR 2025 · similarity 0.88)
- Subject-driven Text-to-Image Generation via Preference-based Reinforcement Learning (NeurIPS 2024 · similarity 0.88)
- DisenBooth: Identity-Preserving Disentangled Tuning for Subject-Driven Text-to-Image Generation (ICLR 2024 · similarity 0.88)
- Evaluating Semantic Variation in Text-to-Image Synthesis: A Causal Perspective (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.
