After writing about infinite crop/zoom using semantic context awareness, I kept working on the workflow. Then I decided I wanted to actually use it without navigating a small industrial facility made of nodes.So, with Codex, I built an app around it. Dystalgia.ComfyUI still runs underneath. The models, custom nodes and generation pipeline are still there. But now I can load an image, select a detail and explore it through an interface built specifically for that process.The underlying idea remains the same. A crop needs context. Zoom far enough into an eagle’s eye and eventually you’re giving the model an ambiguous patch of color. Showing the vision model both the crop and the complete image gives it a way to describe what that patch actually belongs to. That description guides the next generation.
The resulting detail is invented, not recovered from some hidden photographic information. The aim is to make the invention stay connected to the subject as you move deeper.
Building the application was its own experiment. I described what I wanted, Codex implemented it, I tried it, and then discovered all the things I hadn’t thought of. Repeat.
It now also generates images, edits them with Flux Klein, and supports painting masks for inpainting and extending individual edges for outpainting. Those results become the next image you can work on.
The video shows the app in use. What started as a ComfyUI experiment is becoming something I can open and use directly. Which, after spending enough time connecting nodes, feels like progress 🙂
her is the main papp on google drive. passcod for zip “dystalgia”. you just go there and enter by and batch file is there.in the folder Dystalgia.
This is the workflow that the app runs on :

