Research map / Image generation / Text-to-image diffusion
CogView3: Finer and Faster Text-to-Image Generation via Relay Diffusion
Wendi Zheng, Jiayan Teng, Zhuoyi Yang, Weihan Wang, Jidong Chen, Xiaotao Gu, Yuxiao Dong, Ming Ding, Jie Tang
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 relay diffusion (low-res base plus relaying super-resolution) instead of a single-stage model | CogView3 · DrawBench human evaluation | win rate vs SDXL: 77.0% win rate at about half the inference time | Improved |
| Choose relay starting point at the midway 500/1000 rather than earlier or later (ablation) | CogView3 super-resolution stage · DrawBench | ImageReward: 0.829 (200) / 0.812 (800) -> 0.847 (500) | Improved |
| Add prompt expansion at inference for CogView3 (ablation) | CogView3 · DrawBench human evaluation | prompt alignment | Improved |
| Add prompt expansion for SDXL and Stable Cascade (ablation) | SDXL / Stable Cascade · DrawBench human evaluation | prompt alignment | No clear effect |
| Progressive distillation combined with relay diffusion | CogView3-distill (4+1 steps) · DrawBench | ImageReward: 0.394 (LCM-SDXL) -> 0.731 | Improved |
| Distill CogView3 to 4+1 steps | CogView3 · Inference time | time cost: 10.33s -> 1.47s | Improved |
| CogView3 vs SDXL on FID | CogView3 · COCO-5k | FID: 26.29 (SDXL) vs 31.63 | Hurt |
| Tiled diffusion for iterative super-resolution to 2048x2048 (ablation) | CogView3 super-resolution · 2048x2048 generation | visual quality | No clear effect |
Closest papers by meaning
- Würstchen: An Efficient Architecture for Large-Scale Text-to-Image Diffusion Models (ICLR 2024 · similarity 0.90)
- SwiftBrush: One-Step Text-to-Image Diffusion Model with Variational Score Distillation (CVPR 2024 · similarity 0.90)
- PixArt-$\alpha$: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis (ICLR 2024 · similarity 0.89)
- Exploring the Deep Fusion of Large Language Models and Diffusion Transformers for Text-to-Image Synthesis (CVPR 2025 · similarity 0.89)
- Prompt-Free Diffusion: Taking "Text" out of Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.89)
- Any-Size-Diffusion: Toward Efficient Text-Driven Synthesis for Any-Size HD Images (AAAI 2024 · 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.
