The prompts in Prompt Gallery are written in plain sentences, so they work in almost any image tool. What differs between tools is where the extra parts go: the negative prompt, the reference photo and the aspect ratio. This lesson shows how to move a prompt into the tool you use without losing any of it.
See which models a prompt was written for
Each prompt lists the models it was written and tested for, such as Nano Banana, Midjourney, Flux, Imagen, Ideogram or GPT Image. Open a prompt and you find them under Written for, above its tags; click one to see every prompt for that model. The list is also part of the search: type a model name to see only the prompts written for it.
nano banana prompts for Nano Banana
flux kontext prompts for Flux Kontext
midjourney portrait Midjourney prompts about portraitsA prompt not listed for your model often still works. Start with the listed ones when you want the most predictable result.
Put the negative prompt in the right place
Every prompt in the gallery has a Negative prompt: what to leave out. It has its own Copy button. Where it goes depends on the tool:
| Your tool has | Do this |
|---|---|
| A negative prompt field | Paste it there |
A parameter for exclusions, like Midjourney's --no | Add it after that parameter |
| Neither, as in many chat-style tools | Add one sentence at the end: "Avoid: " followed by the list |
In Midjourney, keep --no short: pick the most important exclusions rather than the whole list.
Attach a reference photo first
Prompts that start from "the subject from the reference photo" need your photo attached as an image reference, before you send the prompt. Tools that edit or generate from a reference image (such as Nano Banana, Flux Kontext or GPT Image) handle these best. In a tool that has no image input, those prompts will invent a person instead.
Set the shape twice
Many prompts end with a ratio, like "Aspect ratio 16:9". Some tools follow that sentence; others only listen to their own setting. To be safe:
- Set the ratio in Customize, so the prompt says it
- Also set it in your tool: a size or ratio menu, or a parameter like Midjourney's
--ar 16:9
Shorten for tools with a length limit
Some tools accept only a certain length of prompt, or pay less attention to the end of a long one. The gallery's prompts are grouped into labelled parts (for example Layout, Identity, Look and Output), which makes them easy to trim:
- Keep the first line (what to make) and the part that matters most to you
- Keep the Identity part in photo prompts, and the text rule in prompts with words
- Drop descriptive extras before you drop instructions
When the model ignores part of the prompt
Every model has blind spots. When a result misses something:
- Generate again first. Results vary a lot from run to run
- Move the missed instruction earlier in the prompt; many models weigh the beginning more
- Say it once, plainly. Repeating an instruction in three ways rarely helps
- Try another model listed on the prompt. The gallery's example image shows it can be done
The last lesson turns the patterns you have seen into prompts of your own.