Prompts for people who ship the work.
A working library of prompt templates, organised by what you are making — text, images, video, audio, avatars, software, marketing — with separate packs for the tools you actually open.
- 07Categories
- 08Tool packs
- 73Templates
- 00Signup walls
Top 24 AI models right now
The highest-rated current models from 12 labs, ranked on independent benchmarks, plus the most popular image and video models by blind human votes. Models released in the last 60 days are tagged New. Open any model for prices, specs and head-to-head comparisons.
Language models
- New
GPT-6 AstraOpenAIECI 169.1 - New
Claude Opus 5.5AnthropicECI 168.4 - NewGPT-6.1 TerraOpenAIECI 168.1
- NewGPT-6.1 SolOpenAIECI 167.8
- NewClaude Sonnet 5.5AnthropicECI 166.7
- New
Gemini 4 ArgonGoogleECI 166.1 - New
Muse WonderMetaECI 161.7 - NewClaude Haiku 5.5AnthropicECI 160.4
- New
DeepSeek V4.1 Flash 1017DeepSeekECI 158.8 
Kimi K3Moonshot AIECI 158.3- Gemini 3.8 FlashGoogleECI 157.6
- Muse Spark 1.3MetaECI 157.4
Qwen3.8 Max 0902Alibaba (Qwen)ECI 156.2
GLM-5.3Z.ai (Zhipu)ECI 156.2- DeepSeek V4 Pro 0813DeepSeekECI 155.9
- NewMuse Spark OpenMetaECI 155.1
- New
Grok 4.7xAIECI 154.6 - GLM-5.3-FlashZ.ai (Zhipu)ECI 153.2
- Qwen3.8 27BAlibaba (Qwen)ECI 151.8
- New
Mistral Large 4Mistral AIECI 150.6 
Inkling SmallThinking MachinesECI 150.5- Kimi K2.7 CodeMoonshot AIECI 150.2
- InklingThinking MachinesECI 148.5
MiniMax-M3MiniMaxECI 146.9
Popular image models
- NewGPT Image 2.5 SunburstOpenAIArena 1,517
- NewGPT Image 2.5 FlareOpenAIArena 1,482
- Grok Imagine Image 2.0xAIArena 1,439
- New
MAI-Image-2.6MicrosoftArena 1,427 - NewNano Banana 2.1GoogleArena 1,427
Seedream 5.0 ProByteDance SeedArena 1,390- Nano Banana ProGoogleArena 1,390
- NewQwen-Image 2.1Alibaba (Qwen)Arena 1,362
Popular video models
Start from the output you need
Prompting is medium-specific. A prompt that produces a clean product photograph looks nothing like one that produces a passing test suite.
-
01
Text prompts
Prompts for the writing you actually ship: articles, landing pages, emails, summaries and rewrites.
-
02
Image prompts
Describe the shot like a photographer or art director, and the model stops guessing.
-
03
Video prompts
One shot, one movement, one action. Video prompts fail when they try to be a whole scene.
-
04
Audio prompts
Music and voice models respond to arrangement notes, tempo and mix language — not to genre names alone.
-
05
Avatar prompts
The hard part of avatars is not the first image, it is the second one looking like the same person.
-
06
Software prompts
Coding agents fail on missing context, not missing intelligence. Give them the constraints a new teammate would need.
-
07
Marketing prompts
Marketing prompts get good when you feed them your data instead of asking for best practices.
Same intent, different phrasing
Midjourney wants a subject-first shot list. Claude wants tagged sections. Kiro and Antigravity want acceptance criteria and verification commands.
-
01
Prompt for Midjourney
Midjourney rewards concrete nouns, camera language and a short parameter tail. Adjective stacking does very little.
-
02
Prompt for Gemini
Gemini is at its best when you hand it a lot of material and a precise job to do with it.
-
03
Prompt for Kiro
Kiro works from your project context. The prompt that pays off most is the one you write once as a steering file.
-
04
Prompt for Antigravity
Antigravity agents work across editor, terminal and browser. Prompt them like you would brief a contractor who has to show evidence.
-
05
Prompt for ChatGPT
Most ChatGPT problems are input problems. Set up your instructions once, then keep each request to a single job.
-
06
Prompt for Claude
Claude follows structure well. Tag your inputs, separate instructions from material, and it stays on the rails.
-
07
Prompt for Sora
Write Sora prompts like shot cards. One subject, one action, one camera move, one light setup.
-
08
Prompt for Veo
Veo is strongest when you approve the frame first and ask only for motion and sound.
Seven parts, every time
Medium and model change the details, not the skeleton. When output disappoints, one of these is missing.
Role and audience
Who produces this, and who reads it. Sets vocabulary and depth.
One task
A single deliverable. Anything with an "and then" belongs in a second prompt.
Context you supply
Source text, product facts, code, screenshots. Never rely on recall for something you can paste.
Constraints
Length, format, tone, banned words, technical limits. Most of the quality lives here.
Output shape
Headings, a table, JSON, a shot list. Name it rather than hoping for it.
Uncertainty rule
How to flag doubt, and what to do when the answer is not in the material.
Scope fence
What to leave alone. Essential once a model can edit files or run commands.
Fast diagnostic: read your prompt as a competent stranger would. Every question you would need to ask is a gap the model fills by guessing.
Learn it from the source
Two references worth reading in full before anything else. Both are vendor documentation, maintained by the people who build the models.
- What is prompt engineering? — Google Cloud cloud.google.com ↗ Plain-language overview of prompt engineering, prompt types and where it fits in a production workflow.
- Prompt engineering guide — OpenAI API docs developers.openai.com ↗ The API-side reference: message roles, instruction placement, reasoning controls and structured output.
Pick a medium and get going
No account, no paywall, no newsletter gate. Every template is copy-and-go, and every page explains why it is shaped the way it is.