Squeeze My Prompt
All tasks

I want to generate an image with AI.

Every question is optional. The ones marked Recommended matter most.

Attach the picture you want to change in the tool.

AI tools invent labels, logos, prices and people. Before you publish, check every detail against the real product, and use only your own brand and people who agreed to appear.

What do you need?

For example: replace the green leaves with a single sprig of mint

For example: the lemons, the angle, the crop and the light

For example: Dockside Lemonade, a sparkling lemonade in a glass bottle

For example: Studio product photo

For example: Two whole lemons and one cut half, with two green leaves

For example: Transparent (cut-out)

  • Transparent (cut-out): for logos, stickers, product cutouts. Works in ChatGPT and Ideogram; others need the background removed after.
  • Plain white: easy to remove later.
  • Plain color: say which color under Setting.
  • A real place: describe it under Setting.

For example: Soft studio light from the upper left, bright and fresh

For example: Whole subject, centered with space around it

  • Close-up: the subject fills the frame.
  • Wide: the subject is small in a larger scene.
  • Low angle: looking up; feels big.

For example: A sunlit marble kitchen counter

For example: Bright lemon yellow and leaf green; fresh and clean

For example: FRESH SQUEEZED

For example: extra fruit, knives, cutting boards

  • The subject: keep its shape, colors and details.
  • The style: match its look, not its content.
  • The layout: follow its composition.

For example: Facebook or Instagram feed

For example: Leave clean space; I'll add the words afterward

  • Leave clean space; I'll add the words afterward: safest for prices, offers and brand lettering.
  • Put my exact words in the picture: best kept to a short headline.
  • No words, and no empty space: for Google's image ads, or when the picture alone does the job.

Google's rules for ad images do not allow large empty areas and restrict text laid over the picture. Choose "No words, and no empty space", and type your headline into Google's own fields.

Type your words in Headline, Offer or price, and What to do next. If you don't see them, add them under More detail.

For example: Top

On a full-screen vertical ad at Phone 9:16 the space goes at the top or the bottom: Left side and Right side are read as Top.

For example: 2 for $5 this week

For example: Find it chilled

For example: Real lemons, nothing artificial.

For example: people in their twenties and thirties who want a cold drink that is not a soda

For example: Dimpled, glossy peel; a few water droplets

For example: Everything sharp

  • Background softly blurred: subject stands out.
  • Macro: tiny things shown large.

For example: across the top in bold white sans-serif letters

More detail, if you have it

Saved on this device only

How to write a prompt for an AI image tool

Use this form for a still image from ChatGPT, Gemini, Midjourney, Firefly, Ideogram or a similar tool: pick New image or Edit my image at the top. For Midjourney, choose it under “Write for” in the settings card.

  • Kind of image. The makers advise naming the kind of image first. FLUX’s guide opens every prompt with the image type and medium, and Ideogram’s template starts the same way. Leonardo’s guide notes that without a stated style the model falls back on the style it has seen most for that subject.
  • What’s in it. xAI’s guide says to write the subject first and give exact counts, and Midjourney suggests “three cats” over “cats”. Words near the start carry more weight, Ideogram says.
  • Background. A real transparent background is a setting in the tools that have one, not a word in the prompt. Ideogram warns that asking for white gives a white area, not transparency, and both Ideogram and OpenAI say a drawn checkerboard is not transparency.
  • Light. FLUX and xAI ask for the light’s source, direction and softness.
  • Framing. Composition is part of Google’s prompt formula and of Midjourney’s list of details to consider.

Why one paragraph, and why “Avoid” is short

Most makers write their examples as plain sentences, and Google, Ideogram and Seedream’s guide say keyword lists work less well. OpenAI accepts any format. Google, Midjourney and FLUX advise describing what you want instead of what you don’t, and a 2024 study of DALL-E 3 found negation a common failure. So the form keeps exclusions to a short list of specific things.

What the form adds for you

For a new image it ends with “Avoid: text, logos, watermarks” unless you typed words for the image; OpenAI’s own examples use the same exclusions. It writes the shape as one word (square, vertical or horizontal) and puts the exact ratio in the settings card, because most tools set the shape with a control. For Midjourney it moves the ratio and the Avoid list to --ar and --no at the end, as Midjourney’s parameter list and its No page describe.

When you edit a picture

The form’s edit prompt puts the change first, then says what stays. xAI’s guide advises leading with the changes and following with one sentence that keeps the rest, and says an edit that opens with “keep everything” barely moves. OpenAI advises saying “change only X” and listing the details to preserve, and Google’s editing example tells the model not to change any other elements. If you name nothing to keep, the form asks for everything else to stay the same. An edit adds no shape and no default Avoid list.

A worked example

Lemons on a transparent background, in a 4:5 portrait shape:

Studio product photograph: two whole lemons and one lemon cut in half, showing its juicy cut face, with two fresh green lemon leaves. Dimpled, glossy peel and a few small water droplets on the peel and leaves. The subject is isolated on a fully transparent background, with no backdrop, surface, shadow or checkerboard pattern. Soft, even studio light from the upper left, with gentle highlights on the peel; bright and fresh. Bright, natural lemon yellow and leaf green; fresh and clean. The whole subject is centered with space around it, in a vertical composition. Everything sharp. Avoid: text, logos, watermarks, extra fruit, knives, cutting boards.

What not to bother with

Quality tags such as “8k, masterpiece”: FLUX’s guide says to leave them out.