Generative AI Media Pipelines

VesselAI swaps a character's body into your video while preserving pose, lighting, and scene—solving the manual rotoscoping problem for indie creators and studios. Built on ComfyUI + Alibaba's Wan 2.2 Animate (14B FP8) with custom LoRA distillation, it runs entirely local on Apple Silicon with zero cloud dependency. Processes 40+ frame sequences on M4 Pro without OOM by tuning MPS precision and attention modes—demonstrated in production viral content.

**ZorqAI** automates a two-stage video pipeline: polling Discord for user uploads, compressing them with ffmpeg to 720p H.264 for Viggle.ai submission, then monitoring ComfyUI render outputs and posting results back to the same channel. Built with Python polling loops and file-system watchers, it prevents infinite loops through message tagging and file-size validation, handling Discord's 25–100MB limits. Deployed as persistent LaunchAgents on a dedicated Mac, processing video workflows end-to-end without manual intervention.