Upcycling Text-to-Image Diffusion Models for Multi-Task Capabilities

Text-to-image synthesis has witnessed remarkable advancements in recent years. Many attempts have been made to adopt text-to-image models to support multiple tasks. However, existing approaches typically require resource-intensive re-training or additional parameters to accommodate for the new tasks, which makes the model inefficient for on-device deployment. We propose Multi-Task Upcycling (MTU), a simple yet effective recipe that extends the capabilities of a pre-trained text-to-image diffusion model to support a variety of image-to-image generation tasks. MTU replaces Feed-Forward Network (FFN) layers in the diffusion model with smaller FFNs, referred to as experts, and combines them with a dynamic routing mechanism. To the best of our knowledge, MTU is the first multi-task diffusion modeling approach that seamlessly blends multi-tasking with on-device compatibility, by mitigating the issue of parameter inflation. We show that the performance of MTU is on par with the single-task fine-tuned diffusion models across several tasks including image editing, super-resolution, and inpainting, while maintaining similar latency and computational load (GFLOPs) as the single-task fine-tuned models.

Imagine having a powerful, personalised AI assistant right on your phone—one that not only chats with you but can also edit, enhance, or fill in parts of your photos. The problem is, today’s image generation models are huge and too heavy to run on a phone, which has limited computing power. Shrinking these models without losing their abilities is a tough challenge. So, we asked: how can we take a compact, on-device model and train it to handle multiple image generation tasks? Inspired by how the brain assigns jobs to different areas, we explored pre-trained image models and found a key component responsible for solving image generation tasks. We split this part into smaller, specialised units called "experts." Each expert focuses on certain types of image tasks, and depending on what the model is doing, the most relevant experts get more say in the final result. With this setup, we turned a single lightweight model into a flexible multitasker—capable of doing four different image generation jobs—all while staying efficient enough to run on your phone.