The prompt toolkit for AI images: 100 prompts and 4 reference cards

You know the drill: you type a prompt, get a usable image, and the next day you want a second one in the same look. You type the same line, and something completely different comes out. Or you need a blank laptop screen for a mockup and get an invented user interface on it for the fifth time. This toolkit covers exactly these cases. 100 prompts, written according to a fixed blueprint, each in six versions: Midjourney with a complete parameter line, Adobe Firefly with the matching menu setting, and ChatGPT as a sentence, each in German and English. Plus 196 placeholders for your own subjects and 210 adjustment switches for targeted rework. Here's how it works: double-click START.html, and it opens in your browser. No internet needed, no installation, no account. Every field already contains example values when you first open it, so you can copy right away and see what happens. After that, you enter your own product, your own scene, your own colors. The five categories: Product photography for shops and catalogs. Mockups for laptops, screens, print, packaging, and textiles. Illustrations from flat vector to isometric. Backgrounds with gradients, textures, and genuine text-free space. And people, from business portraits to craftsmanship reportage. 20 prompts per category. The adjustment switches are the real trick. Each prompt has two or three prepared changes that you apply with one click and undo just as easily. The difference from a quick manual edit: a switch touches everything that belongs together. If you turn on the shadow for a packshot, the exclusion "no shadow" disappears at the same time, and the "clean cutout" becomes a "clean product photo." Otherwise the finished prompt would contain an instruction and right next to it its opposite. That's exactly where most self-built prompts fail. If no prompt fits, you build one yourself. The toolkit has ten building blocks, eighty options, and six rules that warn you about contradictions. The result comes out ready in all three tool versions. Plus an eleven-page guide: blueprint, fourteen before-and-after comparisons, fifteen typical mistakes, and six recipes for entire image series. Honesty up front: All information on parameters, value ranges, and menu paths comes from the official manufacturer documentation, checked on 13.09.2026. What nobody can credibly promise you is a specific image. Image models roll the dice anew with every run. Plan for two to four passes, especially with empty areas, free text zones, and hands. The package contains no sample images. It shows you how to get to your own.

You're sitting in front of Midjourney and can't remember whether Stylize goes up to 500 or up to 1000. Or whether --sref now transfers the style or the person. This card is made for exactly those moments. Front side for quick reference: All documented parameters in one table, namely --ar, --s, --no, --sref, --sw, --tile, and --oref, each with its value range, default value, and one sentence on what it really does. Also six Stylize corridors for six image types, from the packshot at 0 to 50 to the clay look at 250 to 400. And at the bottom, a finished prompt broken down into its building blocks, so you can see where each element belongs. Back side for putting it into practice: Four prompts with a complete parameter line, ready to type out or modify. A packshot on white, an exposed concrete texture with --tile, an isometric clay scene, and the series workflow with a reference image. Plus five tips, including the difference between --sref and --oref, which many people only grasp after twenty wasted runs. Every parameter, every value range, and every version note comes from the official Midjourney documentation, checked on 13.09.2026. The Stylize corridors are marked as empirical values, because the documentation says nothing on this. The card is double-sided in A4 landscape format and comes in two versions: dark for the second monitor, light for printing and pinning up. Both sides are also included as JPG.

Firefly works differently from Midjourney, and the two differences that cost the most are right at the top of this card. There is no negative parameter. Whatever you don't want in the image, you have to phrase positively. "No retouching" becomes "visible skin texture, no retouched look." If you don't know this, you write exclusions into the prompt that Firefly reads as image content. There is no 4:5. The selection offers five aspect ratios, and 4:5, 3:2, and 2:3 are not among them. The card tells you which format to use instead and how to crop. Front side for reference: The five aspect ratios with their menu names from Landscape to Vertical. An overview of where each function sits in the interface, from the reference under Styles to the curated style images. And a finished prompt broken down into its building blocks. Back side for application: Four prompts to type out, each with the matching menu setting. A laptop mockup in Widescreen, a business portrait via the detour of Portrait 3:4, a flat vector illustration with a fixed palette, and a product shot in Square. Plus five tips from practice. The information comes from the official Adobe help pages, checked on 13.09.2026. Where the help provides nothing, for example on a strength slider for reference images, it says "not described" instead of a guess. The card is double-sided in A4 landscape format and comes in two versions: dark for the screen, light for printing. Both sides are also included as JPG.

With ChatGPT, the strength lies in the second sentence. The first prompt gets you close, and the follow-up correction gets you to the goal. That is exactly what this card is about. Six follow-up sentences to type out. Display too dark? "Make the display a touch brighter and perfectly even." Text on the blank business card after all? Mark the area, then "This area completely empty and white." Portrait looks like it's made of plastic? "Less perfect. More skin texture, slightly uneven complexion, smile more asymmetrical." Everything can be typed out word for word. Front side for reference: What sets ChatGPT apart here, namely a free aspect ratio without a fixed list, the selection tool for targeted corrections, and the conversation instead of the single prompt. Plus a finished prompt broken down into its building blocks. Back side for putting it into practice: Four prompts to type out, namely a business card mockup in top view, a website header with a defined text zone, a craftsmanship reportage, and an isometric scene. Plus five tips, foremost the most important one: one chat per image series. As long as you stay in the same chat, ChatGPT keeps light, palette, and mood surprisingly stable. A new chat resets everything. The information comes from the OpenAI Help Center, checked on 13.09.2026, including the note that, according to the help page, markings made with the selection tool are not always placed precisely. The card is double-sided in A4 landscape format and comes in two versions: dark for the screen, light for printing. Both sides additionally as JPG.

This card answers the question that comes before choosing a tool: What does a good image prompt actually consist of? Eight building blocks, in this order: image type, subject, arrangement, background, camera, light, mood, exclusion. Each comes with an example to copy and a sentence on what happens if you leave it out. Plus the assembled example prompt, so you can see how the eight parts become one sentence. The exclusion groups: Four groups with the English terms that work in every tool, namely text and brand, people and body parts, image defects, and stage and props. The standard sentence that catches eighty percent of the typical slip-ups. The tool comparison: A table covering seven tasks and three tools. How you set the aspect ratio, where the exclusion goes, how you transfer a style, how you keep a series together. And an honest note on where the reviewed documentation simply says nothing. The six most common mistakes: Too many adjectives and too few nouns. Exclusion in the wrong channel. Contradictions in the prompt. Forgetting the aspect ratio. One run and then judging. Series without an anchor. On the back you will also find the reviewed sources with where they were found and the date they were checked. This is the card that the three tool cards refer to. If you are just starting out with AI images, this is the card to begin with. Double-sided in A4 landscape format, in two versions: dark for the screen, light for printing. Both sides additionally as JPG.