Rethinking Global Text Conditioning in Diffusion Transformers
This work addresses a design choice in text-conditioned diffusion models, offering a training-free method to improve performance for AI content generation applications.
The paper investigates the necessity of modulation-based text conditioning in diffusion transformers, finding that conventional pooled embeddings contribute little to performance, but repurposing them as guidance yields significant gains in tasks like text-to-image generation and image editing.
Diffusion transformers typically incorporate textual information via attention layers and a modulation mechanism using a pooled text embedding. Nevertheless, recent approaches discard modulation-based text conditioning and rely exclusively on attention. In this paper, we address whether modulation-based text conditioning is necessary and whether it can provide any performance advantage. Our analysis shows that, in its conventional usage, the pooled embedding contributes little to overall performance, suggesting that attention alone is generally sufficient for faithfully propagating prompt information. However, we reveal that the pooled embedding can provide significant gains when used from a different perspective-serving as guidance and enabling controllable shifts toward more desirable properties. This approach is training-free, simple to implement, incurs negligible runtime overhead, and can be applied to various diffusion models, bringing improvements across diverse tasks, including text-to-image/video generation and image editing.