A guidance prompt is a short, specific instruction attached to a single generation to steer one detail, without needing the person writing it to know anything about how AI image models actually work.
The distinction matters because of what it's not: it's not a long, carefully engineered prompt string built out of technique, weighting particular words, chaining conditions, working around known model quirks, the way prompt engineering usually requires. A guidance prompt looks more like a note you'd leave a photo retoucher: "make the sleeve length match the reference," "same backdrop grey as the last batch," "she should be looking slightly off-camera." Plain instructions, scoped to a specific fix or a specific creative choice, not a technical incantation.
What one actually looks like
A few representative examples: "keep the collar open, not buttoned," "match the shadow direction to the hero image," "warmer skin tone, this looks washed out," "crop tighter on the shoulder detail." Each one targets a single, concrete thing. None of them require knowing what the underlying model is called, how it was trained, or which phrasing patterns it responds to best. That's the point: a guidance prompt reads like feedback you'd give a person, because giving feedback to a person is a skill most stylists and merchandisers already have.
How this differs from a negative prompt
Readers who've used other AI image tools may know the term negative prompt, a list of things to exclude from a generation ("no watermark," "no extra limbs"), usually written in the same technical, keyword-heavy register as the main prompt. A guidance prompt isn't a negative prompt. It's not a subtractive list of exclusions bolted onto a technical instruction, it's a short positive direction about what a specific detail should be, written the way you'd actually talk about a shot.
Why this exists as a separate concept
Most AI image tools put the entire burden of quality on the prompt. Get the wording right, and the model behaves. Get it slightly wrong, and you're re-rolling generations trying to figure out which phrase was the problem. That's a real skill, and it's also a real barrier: a stylist or a merchandiser who knows exactly what a shot needs shouldn't have to also become fluent in prompt engineering to get it. A guidance prompt exists to separate those two things. The system already carries the load of translating a plain instruction into whatever the underlying model actually needs to hear, so the person giving direction can talk about the shot the way they'd talk to a human photographer, not the way they'd talk to an API.
How it's used in practice
Guidance prompts typically attach to one generation or one correction at a time, rather than governing an entire session. That's deliberate: you can generate a strong overall shot and then add a single guidance prompt just to fix the exact detail that's off, an errant hand position, a fabric drape, a background color that's a shade too warm, without re-specifying everything else about the shot from scratch. This is also where hallucination correction and guidance prompts meet: when Brandmachine's quality scoring flags a likely hallucination in a generated image, the fix is often a targeted guidance prompt pointed at just that region, rather than discarding the whole image and starting over.