Context is something I have been actively studying as a crucial component of effective co-creation with generative models. It was a running theme in the Tilly project, but in that case in the context of text generation. We realised that for Tilly to be more useful, she needed to have context of different things, including of past conversations, the creative goals of the studio and so on. Now that I have been diving into diffusion models in the context of the AFR project, I have realised that a similar thing exists in their very architecture: ![[Pasted image 20240103152852.png]] In this case, context is text, which is used as the prompt. However, in my notes of the project, I also noted that an image prompt, which conditions the image generation, can also be included as context. In an iterative human-AI co-creative process, the previous step of the generation can be passed as context to enable a more iterative creative process. The idea of context is a running theme I am exploring in my thesis, and a crucial component of dialogic interaction. Other critical concepts I am exploring in this dialogic model are: interaction through and about, dialogic loops, consistency, and others.