Every Free AI Prompt Generator We Built (Index)

TL;DR: Sixty free prompt generators and template libraries live on this blog, grouped below by what you are trying to make: text, images, video, audio, code, structured data, research and business writing. Every one runs on the page, with no signup. They generate the prompt. They never generate the finished image, video, audio or app.

What counts as a free prompt generator here?

Two things, and the difference matters when you are choosing. A generator is a page with slots: you fill in subject, tone, constraints and format, and assemble a prompt. A template library is a page of finished prompts with bracketed placeholders: copy one, swap the brackets, run it. Generators suit one-off jobs. Libraries suit jobs you repeat.

Everything below is one or the other. Guides, troubleshooting pages and reference tables are excluded, apart from a short labelled section at the end.

Which free prompt generator works for plain text?

Start here if your output is prose, email, analysis or instructions. Eleven pages.

• Free prompt enhancer: ten before-and-after rewrites across ten jobs, plus the cases where enhancing a prompt makes the output worse. The right first stop when a prompt underperforms.
• 200 free ChatGPT templates: numbered 1 to 200 and grouped by use case, from writing and email through sales, code and support. The broadest page on the site.
• 100+ ChatGPT prompt templates: the older, smaller library, organised the same way with click-to-copy [variable] slots. Pick the 200-template page unless you want this one's categories.
• ChatGPT prompt templates PDF: 40 deeper fill-in-the-blank templates, printable to PDF with Ctrl or Cmd+P. Choose it over the two above when you want fewer, longer prompts you can keep offline.
• Translation prompt generator: register, locale and do-not-translate fields, plus back-translation checks and what AI translation still gets wrong.
• Claude prompt templates: 20 prompts using Anthropic's documented conventions: data at the top, XML tags, no prefill. Use it if your ChatGPT prompts feel flat in Claude.
• Claude project instructions: eighteen project-level instruction templates for coding, research, writing, analysis and ops, plus what Anthropic publishes about precedence and limits.
• Gemini Gem instructions: eighteen Gem templates and the verified rules on who can build Gems and what knowledge files cost.
• DeepSeek prompt templates: 25 templates plus five cost levers ranked by size, checked against DeepSeek's own API docs. The page to read if token spend is the constraint.
• Grok prompt templates: 28 templates for news, markets, X listening and fact-checking, with the recency lines that stop a live source becoming a confident wrong answer.
• Kimi K3 templates: 20 templates built for long documents, covering window, caching, placement and truncation.

Which image prompt generator matches your model?

Nine pages. Parameter sets diverge enough between models that a generic image prompt reliably underperforms a model-specific one, which is why these are split.

• Midjourney prompt builder: 12 cinematic presets, a 300-plus parameter list, a style modifier library, and reverse-engineering an existing image into a reusable prompt.
• Nano Banana Pro generator: fill-in-the-blanks fields, eight 4K presets, the resolution table for gemini-3-pro-image, and a converter for prompts written for Imagen.
• GPT Image 2 generator: nine fields built around readable text inside the image, plus 22 templates by job. Use it when the picture has to contain words.
• FLUX.2 generator: 25 photorealism presets, the current FLUX lineup checked against Black Forest Labs' API spec, and why there is no negative_prompt field.
• Ideogram 4.0 generator: built for posters, logos and typographic layouts. Against the GPT Image 2 page: Ideogram for layout-led design, GPT Image 2 for text inside a photographic scene.
• Negative prompt generator: 29 negative prompts, each labelled with the models that accept one, plus the positive rewrite for models that have no negative field at all.
• Free AI headshot prompts: professional, casual, creative and executive looks, tested on Midjourney v7, Ideogram and ChatGPT image.
• 50 Midjourney v7 prompt examples: tested prompts by use case with parameter breakdowns. A library, not a builder; pair it with the builder above.
• Architecture prompt generator: eight slots and value banks for building designers, with worked examples and a blunt list of what image models cannot do with a floor plan.

Which video prompt generator should you use?

Thirteen pages, and the model matters more here than anywhere else. Duration, resolution and shot syntax are model-specific, and getting them wrong usually fails quietly rather than erroring.

• Veo 3 prompt generator: refreshed for Veo 3.1, with 12 cinematic templates, JSON and timestamp modes, and audio blocks that lip-sync.
• Kling 3.0 prompt generator: turns one line into a timed multi-shot list, with four worked examples and a note on the specs Kling does not publish.
• Runway Gen-4.5 generator: 24 templates, each labelled with the model branch, endpoint, ratio and duration it needs.
• JSON video prompt templates for Veo 3: character-consistent multi-shot, dialogue scenes, ads and B-roll as structured objects. Pick it over the plain Veo generator when you need repeatable shots across many clips.
• Seedance templates: the @Image reference syntax, shot-list structure and the documented duration and 4K limits, from ByteDance's Ark docs.
• Luma Ray 3.2 templates: 30 prompts for product, cinematic, motion-transfer, keyframe and reframe work.
• Wan 2.7 templates: text-to-video, image-to-video and reference-to-video, with parameters checked against Alibaba Cloud Model Studio's docs.
• Grok Imagine Video templates: 29 templates for product, UGC, cinematic, explainer and social video, with the settings each needs.
• MiniMax Hailuo templates: character motion prompts plus the durations, resolutions and camera commands verified at MiniMax's docs.
• Pika templates: 20 prompts, what Pika 2.5 accepts, why Pikaffects takes no prompt at all, and when Pika is the wrong model.
• Vidu Q3 templates: 22 fill-in-the-blank prompts grouped by job, from product demo to food and architecture.
• LTX templates: 22 templates labelled with variant, resolution, fps and duration, plus the routing rule for 20-second jobs.
• HeyGen avatar video templates: the outlier. HeyGen's avatar endpoint takes a script, not a prompt, and the page maps 21 templates to the surface each belongs to.

What is there for audio and voice?

Two pages, and that is the whole set. A thin area of the site, not a complete one.

• Suno music prompt templates: 26 style and lyrics templates, separating what Suno documents from what users report but Suno never published.
• ElevenLabs voice and direction templates: 25 templates across model parameters, text-level direction and voice design, each labelled with the models it works on.

Which prompt generators do developers actually use?

Twelve pages. Four are generators that interrogate you before writing anything; the rest are template sets for a named tool.

• Code review prompt generator: fill-in-the-blanks, with ready prompts for Python, JavaScript, Java, Go, Rust and SQL, plus the output rules that make an AI review usable.
• Bug report generator: asks you questions first, then assembles repro steps, expected versus actual, environment and an honest severity.
• Commit message generator: paste a diff, get a Conventional Commits message. 21 prompts covering splits, breaking-change footers and PR descriptions.
• Regex prompt generator: 20 prompts for explaining and building patterns, plus 18 regexes tested in Python, JavaScript, PCRE2, RE2 and POSIX ERE.
• CLAUDE.md templates: where Claude Code looks for the file, how precedence and imports work, and five templates including a deliberately bad one.
• MCP prompt templates and slash commands: server-side prompts that clients surface as slash commands, and how to build one.
• v0 prompt templates: 21 prompts by build stage on Vercel's own published framework.
• Bolt.new templates: 21 templates by build stage, and how Bolt's billing by project-read changes what belongs in a prompt.
• Replit Agent templates: 21 prompts plus the checkpoint rules that keep scope under control when the agent can run commands and publish.
• Base44 templates: 22 templates by build stage, written around the fact that Base44 generates the backend on your first prompt.
• Lovable templates: 22 templates by build stage, and the product decisions a Lovable prompt has to carry.
• n8n AI node templates: 20 templates for prompts that run unattended, once per item, with fields that can arrive undefined.

Among the five app-builder pages, the deciding factor is where the work happens: Base44 and Lovable for prompts carrying schema and product decisions, Bolt for in-browser full-stack work, Replit when the agent runs real commands, v0 for UI you will move into your own codebase.

How do you get valid JSON out of a model?

With a schema, not an instruction. Telling a model to "reply in JSON" produces JSON-shaped text; a schema supplied through the provider's structured-output field produces something a parser can trust.

• JSON prompt generator: seven schemas by job, the exact request body for OpenAI, Anthropic and Google, and the plain-prompt fallback for chat windows where nothing is enforced.

Two references sit next to it: JSON prompts explained for the concept and when to skip it, and the JSON video templates above.

Which research prompt generator should you start with?

Five pages, best read in this order.

• Research question generator: 21 prompts that turn a vague topic into a researchable question, plus where PICO, SPIDER and FINER actually come from. Start here.
• Literature review generator: 21 prompts that synthesise sources you supply, with the citation checks that stop fabricated references. Use it after you have sources, not instead of finding them.
• NotebookLM prompt templates: 28 prompts for extraction, cross-source synthesis, contradiction-finding and citation-tracing.
• Perplexity prompt templates: 24 templates that demand a claim, a source and the sentence behind it. Perplexity for open-web scanning, NotebookLM for documents you already hold.
• Grant writing templates: specific aims, significance, broader impacts, abstracts and budget justification.

Which generator writes the business document you need?

Seven pages, all built on the same discipline: force the inputs, then refuse to invent proof.

• Landing page copy generator: one input sheet and nine section-by-section prompts, hero to FAQ.
• Ad copy generator: character limits read off Meta, Google and TikTok's own docs, plus the policy rules that get AI ad copy rejected.
• Cold email generator: six required inputs, a test for real personalisation versus merge fields, and CAN-SPAM, PECR and CASL rules checked at source.
• Client proposal generator: fill-in-the-blanks for freelancers and agencies, focused on the scope, pricing and acceptance sections that decide the outcome.
• Proposal-to-invoice templates: six templates covering the whole engagement, with per-client variables. Pick it over the proposal generator when you want the sequence, not one document.
• SOP generator: the prompt plus the interview sequence that pulls undocumented judgement calls out of whoever does the work.
• Gamma presentation templates: 22 templates by deck job, and why the prompt has to carry the narrative arc, slide count and audience or Gamma picks all three.

References, not generators

These six produce nothing. They are lookup tables, listed because the generator you are using will often send you to one.

• Negative prompt support matrix: which image APIs publish a negative_prompt parameter and which only honour negation in prose.
• Model-specific formatting cheat sheet: XML versus markdown, system versus user, and context order across eight model families.
• LLM parameter cheat sheet: ranges, defaults and the parameters your provider now rejects.
• Prompt engineering cheat sheet: frameworks, sampling parameters and the moves that lift output quality, on one printable page.
• Midjourney style modifiers: tested phrases that produce reliable looks, by genre.
• Video duration parameters by model: every duration parameter with the resolution and frame-rate couplings that break it.

How do you actually use one of these pages well?

Fill every slot, then delete the ones that turned out empty. The failure mode is leaving a placeholder in and letting the model invent a value for it.

Say you want a product photograph with legible packaging text. The GPT Image 2 generator asks for nine fields, and filled in they assemble to roughly this:

Subject: a matte black insulated water bottle, 750ml, standing upright
Text in image: the words "TRAILHEAD 750" in a condensed sans-serif,
  centred on the bottle, correctly spelled, no other text anywhere
Composition: three-quarter view, product centred, slight low angle
Lighting: soft key from camera left, subtle rim light on the right edge
Background: seamless warm grey, gentle falloff, no props
Colour: neutral, true blacks, no colour cast
Detail: visible matte texture, no reflections of the studio
Aspect: 1:1
Exclude: hands, condensation, brand logos other than the text above

Two things make that work, and both come from the generator rather than general prompting advice. The Text in image field states the exact string and forbids other text, because "add the brand name" is where garbled lettering comes from. Exclude is written as an instruction rather than a negation, because several image models have no negative field and will render whatever you told them not to.

Then keep what worked and save it somewhere you will find it again. That last step is the one everybody skips.

What is not on this list?

Three honest gaps, as of 27 August 2026.

Jobs with no generator page at all. There is no résumé or cover-letter generator, no spreadsheet-formula or SQL generator, no SEO meta-description generator, and no podcast or newsletter generator.

Audio is thin. Two pages, both template libraries, neither a builder.

Role-based prompt packs are a separate cluster. The developer, marketer and debugging prompt collections are organised by who you are rather than by what you are making. They are useful, and mixing them in here would have doubled this page without helping anyone choose.