Research map / Image generation / Text-to-image diffusion
UniVerse: Empower Unified Generation with Reasoning and Knowledge
Kaiyue Sun, Weiyang Jin, Chengqi Duan, Rongyao Fang, Xian Liu, Yuwei Niu, Chunwei Wang, Aoxue Li, Xihui Liu
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 CoT injection to UniVerse training (ablation) | Bagel · R2I-Bench, no thinking | overall score: 0.40 -> 0.44 | Improved |
| train on UniVerse without CoT | Bagel · WISE, no thinking | overall score: 0.52 -> 0.55 | Improved |
| add CoT injection to UniVerse training (ablation) | Bagel · WISE, no thinking | overall score: 0.55 -> 0.56 | No clear effect |
| train on UniVerse with CoT injection | Bagel · R2I-Bench, no thinking | overall score: 0.36 -> 0.44 | Improved |
| add CoT injection to UniVerse training (ablation) | Bagel · R2I-Bench Causal, no thinking | score: 0.32 -> 0.69 | Improved |
| add CoT injection to UniVerse training (ablation) | Bagel · R2I-Bench Common Sense (CS), no thinking | score: 0.42 -> 0.38 | Hurt |
| add CoT injection to UniVerse training (ablation) | Bagel · WISE, thinking enabled | overall score: 0.74 -> 0.77 | Improved |
| add CoT injection to UniVerse training (ablation) | Bagel · R2I-Bench, thinking enabled | overall score: 0.51 -> 0.54 | Improved |
| enable thinking mode at evaluation | Bagel · WISE | overall score: 0.52 -> 0.70 | Improved |
| train on all UniVerse data instead of a single category (ablation) | Bagel w/ CoT · WISE, thinking enabled | overall score: 0.74 -> 0.77 | Improved |
| train on Reasoning subset (largest category) (ablation) | Bagel w/ CoT · WISE, thinking enabled | overall score: 0.70 -> 0.74 | Improved |
| keep salvaged samples (realigned after filtering) in arithmetic data (ablation) | Bagel w/ CoT · WISE, thinking enabled | overall score: 0.71 -> 0.72 | Improved |
Closest papers by meaning
- MMMG: A Massive, Multidisciplinary, Multi-Tier Generation Benchmark for Text-to-Image Reasoning (NeurIPS 2025 · similarity 0.90)
- Thinking-while-Generating: Interleaving Textual Reasoning throughout Visual Generation (CVPR 2026 · similarity 0.90)
- GIR-Bench: Versatile Benchmark for Generating Images with Reasoning (ICLR 2026 · similarity 0.89)
- GoT-R1: Unleashing Reasoning Capability of Autoregressive Visual Generation with Reinforcement Learning (ICLR 2026 · similarity 0.89)
- Image Content Generation with Causal Reasoning (AAAI 2024 · similarity 0.89)
- ROVER: Benchmarking Reciprocal Cross-Modal Reasoning for Omnimodal Generation (ICLR 2026 · 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.
