Independent AI developer focused on practical, local-first generative media systems.
- ComfyUI pipelines for image and video generation
- Reproducible Wan2.2 and Qwen-Image workflows
- Resource-aware inference on consumer hardware
- Automated quality gates for generated video, including motion, freeze, black-frame, and decode checks
- End-to-end short-form video assembly with FFmpeg, captions, loudness normalization, and delivery checks
local-ai-video-pipeline documents a local AI short-drama production pipeline, with runnable ComfyUI API orchestration, Wan2.2 action evaluation, long-job controls, and measured results from an M4 / 16 GB machine.
Keyframe-anchored video generation, motion quality evaluation, low-memory inference, reproducible benchmarks, and ROCm compatibility for AMD Instinct GPUs.