Bobs Studio is the generation tool I wanted to be using, so I built it rather than keep working around the ones that existed. A frameless desktop client with a fluid panel-docking workspace, my own GPU doing local inference on one side and cloud APIs for frontier models on the other.
The native inference engine
The native engine speaks twenty-plus model architectures across image and video, in every checkpoint format the ecosystem uses (including quantized weights for cards that don't have datacenter VRAM). GPU work runs in its own isolated process and streams live latent previews back while it renders, and the app bootstraps its own Python environment on first launch because nobody should have to install CUDA by hand, me included, which is roughly why this app exists.
Built for daily production work
Distilled models get detected automatically and their constraints enforced, so one wrong slider can't silently torch a generation, and a combination that cannot fit gets refused before it loads rather than halfway through, because the budget is checked against the VRAM actually free rather than the VRAM the card advertises. Long jobs interrupt cleanly in the middle of a step instead of forcing you to wait them out, and there's a hires-fix pipeline that does tiled upscale-and-refine without seams, and the same workspace drives a Discord bot relay because the community I built still likes generating together.
Training moved into the app rather than staying in a folder of scripts, so a run gets set up and costed and watched in the same place the results land, and the language models the app talks to can now be local ones running beside the image engine instead of only cloud endpoints. Both of those exist because I kept leaving the app to do them and coming back with files that no longer matched what the app thought it had.
This is also the machine behind the Studio on this site: everything there was generated and upscaled through this app, on checkpoints I trained.