1 paper
Advait Gupta, NandaKiran Velaga, Dang Nguyen +1
Text-to-image models like stable diffusion and DALLE-3 still struggle with multi-turn image editing. We decompose such a task as an agentic workflow (path) of tool use that address…