Simple method controls image detail levels in diffusion transformers
Flat Features, Detailed Images? Simple Detail Control for Diffusion Transformers
Computer Vision and Pattern Recognition
Summary
Text-to-image models often struggle to reliably control how detailed their pictures look. The authors found that making the internal features more evenly spread out (flatter) leads to more detailed images. They created an easy way to adjust image detail continuously by changing the feature distributions during generation without extra training. This method keeps the picture clear and accurate and adds almost no delay when making images.
What this means in practice
- •For image generation engineers: Add fine-grained detail control to diffusion transformer models without retraining or extra computation.
- •For digital artists using ai tools: Provide adjustable image detail settings to create images that fit aesthetic or project needs.
Authors
Qingtao Yu, Zhen Wang, Jiyang Zheng, Minghao Sun, Hongdong Li, Dylan Campbell
Abstract
Prompt engineering is commonly used to encourage text-to-image models to generate more detailed images. However, it does not provide reliable, continuous, or disentangled control. We analyze feature tokens produced with and without detail-enhancing prompts across six diffusion transformers and observe a common pattern: these prompts tend to increase the entropy of each token's energy distribution while largely preserving its ordering. That is, flattening the distribution increases the level of detail. Motivated by this counter-intuitive finding, we introduce a simple training-free approach that continuously adjusts the level of detail according to a scalar. Specifically, it shifts the squared channel magnitudes of each token toward a higher entropy distribution with a time-aware scheduler to concentrate the intervention at intermediate sampling steps. Across extensive experiments on different types of diffusion transformers, we show that our approach provides continuous control over detail metrics while largely preserving image quality and semantic structure. It requires only a few lines of code and adds negligible inference time overhead.