Practical playbooks
Step-by-step notes on workflows, constraints, review loops, and implementation choices for AI systems and generative-media production.
About // Builder behind the notes
Hi, I’m li qunnan.
I’m an AI engineer, technical writer, and independent builder focused on turning fast-moving AI capabilities into useful, maintainable systems.
What I do
Building practical systems around models, creative tools, and the open web.
My work sits between AI engineering and creative production. I explore local model inference, agent workflows, browser automation, text-to-image, text-to-video, ComfyUI, structured content, and technical publishing.
I am most interested in the step after a demo works: making it repeatable, observable, easier to review, and useful for real people. That can mean a reliable browser workflow, a reusable visual-generation pipeline, or a website that makes its knowledge easy for people and machines to find.
For readers
What you can expect from this site
Useful technical context
Clear explanations of the infrastructure, tools, and trade-offs behind a result—without treating novelty as proof of usefulness.
Reusable patterns
Templates and decision frameworks you can adapt for local inference, browser automation, image and video workflows, or content operations.
How I publish
Notes that explain the “why,” not only the tool.
Most articles begin with a real production question: how to keep a visual series consistent, how to make an AI workflow reviewable, how to reduce inference cost, or how to publish technical knowledge in a form that remains useful over time.
I try to separate stable principles from fast-changing product details. This makes the writing useful whether you are evaluating a new model, building a small internal tool, or designing a workflow for a content team.
Current focus
Generative media with stronger creative operations.
Right now, I am especially focused on the systems around text-to-image and text-to-video: prompt architecture, reference-driven consistency, ComfyUI graphs, keyframe-led video pipelines, asset review, and human-in-the-loop automation.
The goal is not fully unattended creation. It is giving creative people better controls, faster feedback loops, and a dependable record of how an asset was made.
Principles
How I approach the work
Start with evidence
Measure the real workload, capture the actual constraint, and separate observed facts from assumptions before choosing a solution.
Design the loop
Make success observable and failure recoverable. AI systems become more useful when generation, verification, review, and revision have clear roles.
Publish clearly
Write for people first, then provide enough structure, provenance, and context for machines, collaborators, and future readers to understand the work.
Work together
Have a workflow worth making more reliable?
I am interested in thoughtful conversations around AI engineering, generative-media operations, technical publishing, and small systems that create leverage for teams. If you are exploring a practical project, you can reach me by email.