RTX 5090 Microdrama Automation Workflow

I have an RTX 5090–powered workstation and want every part of my microdrama production pipeline to run locally. The plan is to rely on ComfyUI as the core “orchestrator” and generate hyper-realistic clips in which character animation, scenery setup, and voice synchronisation happen automatically from a short prompt or script. You will need to select, install, and optimise the best local models (image-to-video or 3D motion, TTS, lip-sync, etc.) so they run smoothly on the 5090’s VRAM without calling external APIs. Once the models are in place, I expect a ComfyUI graph—or an equally efficient node-based setup—that ingests a text prompt, produces the visuals, aligns the audio, and renders a finished clip. Deliverables • Fully configured local environment with all required models and dependencies • A ready-to-use ComfyUI workflow file that chains prompt → animation → scenery → voice sync → final render • Step-by-step documentation so I can reinstall from scratch and reproduce the same results • A short sample microdrama generated by the system to confirm hyper-realistic output and perfect voice sync Everything must stay offline, be repeatable after a clean Windows install, and work with minimal user intervention once set up. Let’s discuss model choices, licensing, and any performance tweaks you recommend before you begin.