AveryJack Labs

What it is

Three browser-based tools that assemble prompts for AI image and video generators: an image prompt builder, a video prompt builder, and Aurora, a step-by-step builder for cinematic video prompts.

They run anonymously in the browser. No account, no upload, no backend — work-in-progress persists to local storage.

The part that’s actually a knowledge problem

The interesting work here isn’t the React. It’s the vocabulary.

Underneath the interface sits a curated term set of roughly 380 entries, each with a written plain-language definition: 73 artistic media and styles, 227 named artists, and 83 Native American tribes. “Impasto” isn’t just an option in a list — it carries the definition thick paint application technique creating heavy texture and dimension.

That’s a controlled vocabulary with scope notes, which is the same artifact I build for enterprise knowledge bases. The reason it matters is identical in both cases: a term is only useful if the person choosing it knows what it means and uses it the same way every time. An untagged free-text prompt box gives you inconsistent results for the same reason an untagged knowledge base gives you inconsistent search results.

Design decisions

Two prompt outputs, not one. The image builder emits a full prompt that includes MidJourney-specific parameters — OmniReference, stylize, weird, chaos, line styles — and a simple prompt that strips them. The same creative intent has to be expressible for a platform that understands those flags and for one that would choke on them.

Three output formats for video. Plain text, Markdown, or JSON. Some generators want prose, some want structure.

Subject selection is capped. Each subject group — general, flora, fauna, food, seasonal, holiday — allows a small fixed number of picks rather than unlimited. The constraint is the feature: competing subjects are the most reliable way to produce a muddy image.

Aurora sequences the decisions. Six ordered steps run story and narrative first, then characters, visual style, audio, and technical specifications, with generation and export last. It’s an opinion about order — that you should decide what the thing is about before you decide how it’s lit.

Why it’s here

It’s a small product, but it’s the clearest example of me applying knowledge management technique somewhere nobody expects it. Faceted selection, controlled vocabulary, scope notes, constrained inputs, and multiple output serializations of one underlying record are knowledge-base patterns. Here they happen to produce prompts instead of articles.