the find
− 0.6 mTwo sibling workshops on the desktop tower. FluxStudio generates FLUX.2 images with up to three reference images steering a single composition, a curated LoRA library, and a 4K path via upscale. VideoStudio turns prompts into motion: text-to-video through a generated base frame, image-to-video, and reference-to-video that can borrow either a subject's appearance or a dancer's pose skeleton. Clips run two to eight seconds at 720p native, upscaled to 1080p or 4K.
cross-section
− 1.3 mBoth are thin, careful layers over a shared ComfyUI backend — each tool builds its workflow graph in code, submits it, and tracks progress over a websocket. The image side is fully headless-scriptable: a Python module that any agent on the machine can import to generate on demand. The video side's orchestrator is reused in-process by Creator Studio, so the same code path serves a browser click and an API job. Prompt craft is externalized into its own knowledge base — per-model prompting guides that get read before any prompt is written.
inscription
− 2.0 mField note, from the VRAM ledger: On the tower build, the GPU is shared with a resident AI companion whose brain lives on it. So every render job opens with a treaty — the pipeline pauses her brain, takes the VRAM, does the work, and restores her before reporting done. The companion literally naps so the renderer can work, and wakes with the finished frames already on disk.