What is charsheet?

Charsheet is a small web app that turns one picture of a character into a full character reference sheet showing that character from eight different angles: front, back, left, right, and the four diagonals in between. You upload a single image and the app uses an image editing AI model called Qwen-Image-Edit to generate each new view, then combines all eight into one sheet you can download as a ZIP file alongside the sheet image itself. The app is built with FastAPI as the backend and serves a Japanese language web interface. Behind the scenes it depends on Diffusers, a library for running image generation and editing models, plus two smaller helper models: one that speeds up generation and one trained to handle multiple viewing angles. If you already run ComfyUI, the app reuses its downloaded model files, otherwise it fetches them automatically from Hugging Face the first time you run it, which can take a while depending on your connection since the main model file is several gigabytes. Beyond the basic eight view generation, the app can detect and separate multiple characters if you upload a sheet that already contains several of them, letting you pick one to process. Individual views can also be corrected afterward: you type a plain instruction like remove the hair and the app redraws just that one image, with an undo option to switch back. There is also a one click background removal feature powered by a separate model. Running this requires real hardware. The README describes VRAM requirements ranging from under 24 gigabytes up to 65 gigabytes or more, with the app automatically choosing a slower offloading strategy on smaller GPUs. Generation itself, once the model is loaded, takes roughly forty seconds for a full set of eight views on a high end card. Only one generation or edit job can run at a time. Job state is kept in memory, so restarting the server clears the current job list, though the actual image files remain on disk.