Write an image generation prompt that actually gets you what you pictured
Translate the picture in your head into the specific, unambiguous language an image model needs.
- Use it for
- Anyone using an image generator who keeps getting results that are close but not it.
A vague image request produces a vague image, because "a photo of a coffee shop" leaves every meaningful decision — angle, lighting, style, mood, what's actually in frame — up to the model's most statistically average guess. The picture in your head has all of those decisions already made; the gap is that you didn't say them out loud. Image models respond dramatically to specificity in a way text models often don't, which means the ten seconds spent making the implicit explicit is disproportionately valuable here.
This prompt uses a text model to do that translation — turning a vague description into the specific, image-model-ready language that actually captures what you pictured.
When not to use this
If you're intentionally exploring and want to be surprised, a loose prompt is the right tool and this just gets in the way. This is specifically for when you have a real picture in your head and the outputs keep missing it.
Did this work?
The generated image matches your actual mental picture on the specific things you cared about — framing, mood, subject — not just "the same general topic."
Tested on claude-opus-5. Evidence status is draft; it moves to battle-tested only on recorded runs, never by hand.