MaxFusion: Plug&Play Multi-Modal Generation in Text-to-Image Diffusion Models
This addresses the computational bottleneck for adding new tasks in generative models, offering a plug-and-play solution for multi-modal applications.
The paper tackles the problem of scaling text-to-image diffusion models to new modalities without full retraining, proposing MaxFusion, which uses variance maps of intermediate features to combine aligned features from multiple models, achieving efficient multi-modal generation with minimal compute.
Large diffusion-based Text-to-Image (T2I) models have shown impressive generative powers for text-to-image generation as well as spatially conditioned image generation. For most applications, we can train the model end-toend with paired data to obtain photorealistic generation quality. However, to add an additional task, one often needs to retrain the model from scratch using paired data across all modalities to retain good generation performance. In this paper, we tackle this issue and propose a novel strategy to scale a generative model across new tasks with minimal compute. During our experiments, we discovered that the variance maps of intermediate feature maps of diffusion models capture the intensity of conditioning. Utilizing this prior information, we propose MaxFusion, an efficient strategy to scale up text-to-image generation models to accommodate new modality conditions. Specifically, we combine aligned features of multiple models, hence bringing a compositional effect. Our fusion strategy can be integrated into off-the-shelf models to enhance their generative prowess.