Category: AI Tools & Automation

A local model becomes useful the moment it stops being a chat window and starts running inside a script. This section is about that step: transcription, speech synthesis, image and video generation, screenshot and video understanding, and content pipelines that run unattended on a desktop machine. Everything here was built and measured on Windows 11 with an AMD Radeon PRO W7900 — a deliberately awkward combination, because most published guides assume Linux and NVIDIA. Where ROCm needed a workaround, the workaround is in the article. Where a tool was slower than its NVIDIA benchmarks suggest, the real timing is given. The workflows fall into three groups. Vision models read what is on screen — dashboards, error states, video frames — and turn it into text a script can act on. Audio models go both directions, transcribing video with Whisper and generating speech locally. Generation covers ComfyUI driven through its Python API for blog illustrations, and Wan 2.1 for text-to-video without a subscription. Most of these use the runtimes described in Local AI & LLMs, so if a model server, GPU setup or API detail is unclear, that section has the groundwork.