Research map / Image generation / Text-to-image diffusion
GRAN-TED: Generating Robust, Aligned, and Nuanced Text Embedding for Diffusion Models
Bozhou Li, Sihan Yang, Yushuo Guan, Ruichuan An, Xinlong Chen, Yang Shi, Pengfei Wan, Wentao Zhang, Yuanxing Zhang
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 learnable layer weights with continuous training to Norm-Avg (ablation) | Qwen3-VL-8B-Instruct · T2I GenAI-Bench | GenAI-Bench / UnifiedReward: 76.17 -> 75.94; 3.028 -> 3.021 | Hurt |
| add two-step training (freeze learned layer weights) to learnable weights (ablation) | Qwen3-VL-8B-Instruct · T2I GenAI-Bench | GenAI-Bench / UnifiedReward: 76.17 -> 77.01; 3.028 -> 3.035 | Improved |
| fine-tune Qwen3-VL-8B-Instruct on VQA/caption data (GRAN-TED) with learnable weighting | Qwen3-VL-8B-Instruct · T2I GenAI-Bench | GenAI-Bench: 76.17 -> 77.41 | Improved |
| use GRAN-TED instead of Qwen3-VL-8B Norm-Avg for T2V | Wan2.1-T2V-1.3B · T2V GenAI-Bench | GenAI-Bench: 77.94 -> 80.33 | Improved |
| use GRAN-TED instead of Qwen3-VL-8B Norm-Avg for T2V | Wan2.1-T2V-1.3B · T2V VideoAlign | VideoAlign reward: 1.288 -> 1.424 | Improved |
| add layer-wise weighting on top of fine-tuned MLLM (ablation) | GRAN-TED · TED-6K (Norm-Avg features) | accuracy: 57.22 -> 57.42 | Improved |
| use Norm-Avg instead of last-layer features (ablation) | Qwen3-VL-8B-Instruct · TED-6K | accuracy: 55.37 -> 56.81 | Improved |
| use thinking variant with Norm-Avg instead of instruct | Qwen3-VL-8B · TED-6K (Norm-Avg features) | accuracy: 56.81 -> 56.51 | Hurt |
| use thinking variant with last-layer features instead of instruct | Qwen3-VL-8B · TED-6K (last-layer features) | accuracy: 55.37 -> 55.67 | Improved |
| use multimodal MLLM instead of LLM backbone | Qwen3-8B vs Qwen3-VL-8B · TED-6K (Norm-Avg features) | accuracy: 55.77 -> 56.81 | Improved |
| scale LLM from 8B to 32B with Norm-Avg | Qwen3 Instruct · TED-6K (Norm-Avg features) | accuracy: 55.77 -> 56.98 | Improved |
| add fine-tuning to Qwen3-VL on adjective and adverb sub-tasks | GRAN-TED · TED-6K sub-tasks (Norm-Avg) | accuracy: Adj. -1.17; Adv. -1.98 | Hurt |
Closest papers by meaning
- Diffusion Adaptive Text Embedding for Text-to-Image Diffusion Models (NeurIPS 2025 · similarity 0.90)
- Würstchen: An Efficient Architecture for Large-Scale Text-to-Image Diffusion Models (ICLR 2024 · similarity 0.90)
- LightFair: Towards an Efficient Alternative for Fair T2I Diffusion via Debiasing Pre-trained Text Encoders (NeurIPS 2025 · similarity 0.90)
- Rare Text Semantics Were Always There in Your Diffusion Transformer (NeurIPS 2025 · similarity 0.89)
- Scaling Down Text Encoders of Text-to-Image Diffusion Models (CVPR 2025 · similarity 0.89)
- Diffusion-Inspired Truncated Sampler for Text-Video Retrieval (NeurIPS 2024 · 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.
