AI prompts by model
The visual prompt library for AI creators, grouped by the model each prompt was made for. Pick a model to see its prompts next to the image or video they produced.
Nano Banana prompts
30 image prompts · 3 versions
Nano Banana is Google's Gemini image family, and Nano Banana Pro is its high-end tier. It renders legible text and takes studio direction well: the lighting setup, camera angle, depth of field and colour grade. It also edits an existing image from a plain instruction. Prompts that name the light and the lens get closer than lists of adjectives.
Seedance prompts
30 video prompts · 3 versions
Seedance is ByteDance's video model family. It reads camera language literally (dolly, push-in, rack focus), and the newest versions hold a cast, props and a room across several cuts in one clip, cutting where the prompt says. Write it like a shot list: one camera move per beat, the light, then what changes.
FLUX prompts
12 image prompts · 2 versions
FLUX is Black Forest Labs' image family, known for photorealism and for following long prompts closely; FLUX.2 adds generation from several reference images at once. Describe the scene in full sentences (subject, setting, light, lens) rather than a pile of comma-separated tags.
GPT Image prompts
11 image prompts · 2 versions
GPT Image is OpenAI's image model, the one inside ChatGPT. Its headline strength is typography: titles, labels, posters and layouts come out legible, which makes it the pick for graphic design and key art. Put the exact words in quotes and say where they sit on the image.
Midjourney prompts
9 image prompts
Midjourney is the long-running favourite for stylised, art-directed images. It rewards short, evocative prompts and reacts strongly to medium, style and mood words, plus its own parameters such as aspect ratio and stylize. Every example here shows the exact wording next to the image it produced.
Seedream prompts
6 image prompts
Seedream is ByteDance's image family, built around editing: change one part of an image, combine several reference images, or set dense text in many languages. Describe props precisely; a prop described loosely drifts from one take to the next.