Research map / Image generation / Text-to-image diffusion
Democratizing Text-to-Image Masked Generative Models with Compact Text-Aware One-Dimensional Tokens
Dongwon Kim, Ju He, Qihang Yu, Chenglin Yang, Xiaohui Shen, Suha Kwak, Liang-Chieh Chen
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 improved one-stage training recipe (ablation) | TiTok VQ 64 tokens · ImageNet | rFID: 5.15 -> 2.43 | Improved |
| add text-aware de-tokenization (ablation) | TiTok VQ 32 tokens · ImageNet val zero-shot | rFID: 7.72 -> 3.95 | Improved |
| add text-aware de-tokenization (ablation) | TiTok VQ 128 tokens · ImageNet val zero-shot | rFID: 2.63 -> 1.53 | Improved |
| add text-aware de-tokenization (ablation) | TiTok KL 128 tokens · ImageNet val zero-shot | rFID: 1.02 -> 0.90 | Improved |
| use continuous KL tokens instead of VQ (ablation) | TiTok 32 tokens · ImageNet val zero-shot | rFID: 7.72 -> 2.56 | Improved |
| add text-aware tokenizer TA-TiTok (ablation) | MaskGen-L KL · MJHQ-30K | FID: 9.82 -> 8.13 | Improved |
| add aesthetic score conditioning (ablation) | MaskGen-L KL with TA-TiTok · MJHQ-30K | FID: 8.13 -> 7.24 | Improved |
| add aesthetic score conditioning (ablation) | MaskGen-L KL with TiTok · GenEval | Overall: 0.47 -> 0.49 | Improved |
| add text guidance to encoder as well as decoder (ablation) | TA-TiTok KL 32 tokens · ImageNet val zero-shot | rFID: 1.53 -> 1.65 | No clear effect |
| use CLIP embeddings instead of token IDs for text guidance (ablation) | TA-TiTok KL 32 tokens · ImageNet val zero-shot | rFID: 1.62 -> 1.53 | Improved |
| increase token count from 32 to 128 (ablation) | MaskGen-L VQ · MJHQ-30K | FID: 9.11 -> 7.74 | Improved |
| increase token count from 32 to 128 (ablation) | MaskGen-L VQ · GenEval | Overall: 0.43 -> 0.53 | Improved |
Closest papers by meaning
- SoftVQ-VAE: Efficient 1-Dimensional Continuous Tokenizer (CVPR 2025 · similarity 0.89)
- Language-Guided Image Tokenization for Generation (CVPR 2025 · similarity 0.89)
- Highly Compressed Tokenizer Can Generate Without Training (ICML 2025 · similarity 0.88)
- An Image is Worth 32 Tokens for Reconstruction and Generation (NeurIPS 2024 · similarity 0.87)
- FlexTok: Resampling Images into 1D Token Sequences of Flexible Length (ICML 2025 · similarity 0.87)
- UniTok: a Unified Tokenizer for Visual Generation and Understanding (NeurIPS 2025 · similarity 0.87)
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.
