IMHO, for self hosted (open source) generative (still image) AI, the absolutely user friendliest way is the “Krita AI Diffusion” plugin for the FOSS paint program Krita. Not only is it on par with the user-friendlyness of market leaders like Photoshop. Using Black Forest Labs “Flux” models, the quality of the results is on par with Photoshop as well.
Krita AI Diffusion is doing it’s magic via the API of ComfyUI, a node based tool for AI workflows. It has nodes for pretty much every new AI tool that comes out, from still image generation, LLM integration, image Segmentation to AI Video generation etc.
What makes it especially nice for users, is that Krita AI Diffusion offers to install a local ComfyUI server, but it also offers connecting to a remote server, either self hosted or SaaS (which the author of the plugin offers as a service to earn money).
I think it would be wise if KDEnlive would research this option as a way to integrate AI workflows in the future instead of integrating those processing code itself in KDEnlive.