Under the name Diffusae, a small Canadian team called BSKL is developing a plugin that should allow direct access to generative Stable Diffusion models within After Effects.
The operation is currently still quite rudimentary, because Stable Diffusion is "only" available as an embedded layer effect. There, the prompt can be injected by an invisible additional text layer.
Furthermore, the used model, the guidance and strength of the prompt as well as the quality and the seed can be set. Thus, one has far fewer options than when using Stable Diffusion with a WebUI.
On the other hand, the not always trivial installation is supposed to be automated and the update to current models like Stable Diffusion 2.1 can also be done without user intervention.
The biggest advantage, however, is certainly the direct application in image sequences as well as the support of masks - which allows Stable Diffusion to be used in single image sections without overloading the GPU with too high resolutions.
The big question mark, however, remains temporal coherence. Currently, the random changes between the individual images are relatively large due to the principle, which regularly leads to a very psychedelic impression of the effects. The creators hope for significant improvements in this area in the future through further research in the large stable diffusion community.
Currently Diffusae only works under Windows with an Nvidia card. According to the roadmap, a macOS version for Apple Silicon models should be released within the next few days (June 2023).
Currently, the After Effects plugin costs at aescripts, but you can register for a free beta test, which might be the best way for interested users to try out the program more intensively. Additional contact to the developers is possible via a Discord server.