AI Prompt Library

Why fill-in-the-blank prompts work

Prompts that work tend to share a skeleton: what to show, in which style, with what light and camera, at what aspect ratio. Copying someone else’s prompt gives you their subject and their colours, and editing it by hand makes it easy to delete an instruction that mattered. Here the skeleton stays fixed and only the parts you should change are pulled out as blanks.

Filling in image and video prompts

• Be specific about the subject. “A corgi running across a lawn” beats “a dog”. Describe what can be seen: age, clothing, hair, materials. Google's image generation guide (2026) lists “Be hyper-specific” as its first best practice: “The more detail you provide, the more control you have.”
• Lighting and lens change the mood more than anything else. Real photography terms such as “soft window light” or “85mm lens” push the result toward a believable photo. The same guide tells you to “use photographic and cinematic language to control the composition”, with terms such as wide-angle shot, macro shot and low-angle perspective.
• Character sheets depend on the sentence that says the face, hair and outfit stay identical in every panel. Leave that sentence alone and change only the character description and the list of expressions or outfits. Upload the finished sheet as a reference and use the “Same character, new scene” card to keep the face steady in later images. Google's image models accept up to 14 reference images per request, and keep the likeness of up to 4 or 5 characters depending on the model (documentation of October 2026).
• Video is most reliable with one camera move per clip. Set the length and aspect ratio in your tool’s own settings as well, because each tool offers fixed values: Google's Veo 3.1 documentation (2026) lists clip lengths of 4, 6 or 8 seconds only, and 1080p only at 8 seconds. It asks for a subject, an action and a style in every prompt, and treats camera position and motion, composition, focus and lens effects, and ambiance as optional additions.
• For thumbnails and logos with lettering, keep the text short; fewer letters means fewer spelling errors. You can also generate the image with an empty area and add the text in an editor. Google notes that for text in an image the model “works best if you first generate the text and then ask for an image with the text”.

You can fill the blanks in other languages too. Google's guide names 15 languages for best performance, from English to German, Japanese, Korean and Brazilian Portuguese. Support differs from tool to tool, so the fixed part of each card stays in English, which every generator accepts.

Filling in chat prompts

The writing, summarising, coding, study and marketing prompts all follow the same order: role, task, constraints, output format. Changing the audience, length and tone blanks to match your situation makes a visible difference to the answer. Prompts that end with “Text:” or “Code:” expect you to paste your material underneath after copying. Each chat prompt has an answer-language blank if you want the reply in another language.

That order follows what the model makers publish. OpenAI's help centre says to “put instructions at the beginning of the prompt and use ### or """ to separate the instruction and context”, and to say what to do instead of only what not to do. Anthropic's guide says that setting a role makes a difference even in a single sentence, and recommends 3–5 examples. Google's Gemini prompting guide goes further: “We recommend to always include few-shot examples in your prompts”, while warning that too many can make the model overfit to them. If an answer misses, paste in three or four examples of the output you want.

For several long documents, reverse the order. Anthropic reports that placing long material at the top and the question at the end improved response quality by up to 30 percent in its tests. There is far more technique than this: The Prompt Report (2024) catalogues 58 text prompting techniques. The cards here stay with the basics of a role and a fixed output format.

What is in the library

Thirteen categories: motion graphics and video, character sheets (turnaround, expression, outfit), product shots, thumbnails, logos and icons, illustration styles, photo realism, UI mockups, plus chat prompts for writing, summarising, coding, study and marketing copy. The cards are text only, so the page loads quickly and works one-handed on a phone.

Sources (checked 9 October 2026): Google, Nano Banana image generation · Google, Generate videos with Veo 3.1 · Google, Prompt design strategies · OpenAI, Best practices for prompt engineering with the OpenAI API · Anthropic, Prompting best practices · Schulhoff et al., The Prompt Report (2024)