AI Paper Atlas

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

ECCV 2026 · Image generation › Text-to-image diffusion · approach: Compositional text-to-image guidance · 4 citations (OpenAlex)

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 changedWhereMeasuredEffect
add learnable layer weights with continuous training to Norm-Avg (ablation)Qwen3-VL-8B-Instruct · T2I GenAI-BenchGenAI-Bench / UnifiedReward: 76.17 -> 75.94; 3.028 -> 3.021Hurt
add two-step training (freeze learned layer weights) to learnable weights (ablation)Qwen3-VL-8B-Instruct · T2I GenAI-BenchGenAI-Bench / UnifiedReward: 76.17 -> 77.01; 3.028 -> 3.035Improved
fine-tune Qwen3-VL-8B-Instruct on VQA/caption data (GRAN-TED) with learnable weightingQwen3-VL-8B-Instruct · T2I GenAI-BenchGenAI-Bench: 76.17 -> 77.41Improved
use GRAN-TED instead of Qwen3-VL-8B Norm-Avg for T2VWan2.1-T2V-1.3B · T2V GenAI-BenchGenAI-Bench: 77.94 -> 80.33Improved
use GRAN-TED instead of Qwen3-VL-8B Norm-Avg for T2VWan2.1-T2V-1.3B · T2V VideoAlignVideoAlign reward: 1.288 -> 1.424Improved
add layer-wise weighting on top of fine-tuned MLLM (ablation)GRAN-TED · TED-6K (Norm-Avg features)accuracy: 57.22 -> 57.42Improved
use Norm-Avg instead of last-layer features (ablation)Qwen3-VL-8B-Instruct · TED-6Kaccuracy: 55.37 -> 56.81Improved
use thinking variant with Norm-Avg instead of instructQwen3-VL-8B · TED-6K (Norm-Avg features)accuracy: 56.81 -> 56.51Hurt
use thinking variant with last-layer features instead of instructQwen3-VL-8B · TED-6K (last-layer features)accuracy: 55.37 -> 55.67Improved
use multimodal MLLM instead of LLM backboneQwen3-8B vs Qwen3-VL-8B · TED-6K (Norm-Avg features)accuracy: 55.77 -> 56.81Improved
scale LLM from 8B to 32B with Norm-AvgQwen3 Instruct · TED-6K (Norm-Avg features)accuracy: 55.77 -> 56.98Improved
add fine-tuning to Qwen3-VL on adjective and adverb sub-tasksGRAN-TED · TED-6K sub-tasks (Norm-Avg)accuracy: Adj. -1.17; Adv. -1.98Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy