237 curated resources, searchable and source-linked
A searchable snapshot of official documentation, primary-source announcements, papers, courses, and hands-on tutorials, arranged into seven reading paths so the order is a decision rather than an accident. Links and access terms can change; inclusion means curated for this edition, not continuously monitored or independently endorsed.
What this library is, and how to get value out of it
Two hundred and thirty-seven links is not a curriculum. It is a shelf. The difference between a shelf and an education is knowing which book to open first, what it assumes you already know, and when to stop reading and go build something — and that judgement is the part a list of links cannot give you.
So this page is organised in two layers. Below this box are seven reading paths, one per module: an ordered route through that module's resources, what each one assumes, the mistake that most often derails people there, and where to go afterwards. Underneath those is the full catalogue, every resource grouped by module and section, searchable.
If you read only one thing on this page, read the path for the module closest to what you are actually trying to do. Five ordered resources you finish are worth more than forty you bookmark.
What commonly goes wrong: treating breadth as progress. The most common failure with a resource library is opening twenty tabs, feeling productive, and retaining none of it. The second most common is reading the theory forever to avoid the discomfort of building something badly. Both are addressed the same way — pick the path, do the first item, then build the smallest possible version of the thing you want.
Everything here is free to read at the source, with two exceptions that are labelled where they appear: two links sit behind Medium’s member paywall, confirmed from Medium’s own isLocked field on 30 August 2026. Nothing on this site is gated, and no link here is an endorsement — inclusion means it was curated and checked for this edition.
Seven reading paths
One per module. Each names specific resources in the order worth taking them.
27 resourcesFoundations & Prompt Engineering
Who this path is for: Anyone who has used a chat model and wants to stop guessing why some prompts work.
- Start with the picture, not the prose. 3Blue1Brown: But what is a GPT? (27 min), then Attention in transformers (26 min). Watch them in that order and watch them before you read any prompting advice. Almost every prompting rule that follows is downstream of one fact these videos make visible: the model predicts the next token from everything currently in its context, and nothing else.
- Then one mechanical detail. Prompt Engineering Guide: LLM Tokenization (15 min). Short, and it explains a whole class of confusing behaviour — why models miscount letters, why they mangle rare names, why a prompt that fits yesterday overflows today.
- Then the two official prompting references. Claude Docs: Prompting best practices (25 min) and OpenAI's prompt engineering guide (30 min). Read both, not one. Where they agree you are looking at a property of the technology; where they differ you are looking at a house style, and knowing which is which is the whole skill.
- Then structure. Claude Docs: Use XML tags to structure your prompts (15 min). This is the single highest-yield fifteen minutes in the module.
- Then the one long thing. Anthropic's Interactive Prompt Engineering Tutorial (3–5 hrs). It is the only item here that will change how you work rather than what you know, because you type into it. Do it when you have a real task in hand, not as homework.
What commonly goes wrong: Collecting prompt tricks. People arrive expecting a list of magic phrases and leave with sixty tabs. The chain-of-thought, few-shot and meta-prompting pages here are worth reading, but they are variations on one idea — put the reasoning you want in the context — and reading all four before you have used any of them is how the module gets abandoned.
33 resourcesRAG & MCP
Who this path is for: People who now want the model to work over their own documents or reach their own tools.
- Understand the problem before the plumbing. IBM: What is RAG? (15 min). RAG exists because a model's weights are fixed at training time and your documents are not. If that sentence already makes sense, skip ahead.
- Then the one concept everything rests on. Pinecone: What are Vector Embeddings (20 min), then Semantic Search with Pinecone (20 min). Embeddings are the load-bearing idea; a retrieval system you do not understand is a search engine you cannot debug.
- Then build the smallest possible version. Hugging Face: Code a Simple RAG From Scratch (30 min). Build it before you evaluate any framework. It is roughly a hundred lines and it will make every framework's documentation legible.
- Then pick one framework, not three. LlamaIndex Starter Tutorial or LangChain: Build a RAG Agent — one of them. The freeCodeCamp × LangChain course (2.5 hrs) is the deeper path if you have the afternoon.
- MCP is a separate topic; treat it that way. Anthropic: Introducing the Model Context Protocol (10 min), then modelcontextprotocol.io's intro (15 min), then build one server with the Quickstart (30–45 min). RAG is about giving a model things to read. MCP is about giving it things to do. They get bundled together constantly and confusing them will cost you a week.
What commonly goes wrong: Reaching for a vector database on day one. Pinecone, Weaviate and pgvector are all here because eventually you will need one, but a first RAG system over a few hundred documents does not need any of them, and starting there means debugging infrastructure when you should be debugging retrieval quality.
28 resourcesImage & Video Generation
Who this path is for: Two different readers: one who wants to make good images, and one who wants to know how diffusion works. The module serves both, and they should read it in opposite directions.
- If you want to make things: start at the prompting sections. The Flux prompt guide (20 min) and the SDXL realistic-prompt guide (25 min). Then open Hugging Face Spaces and use a model for free before you pay for one.
- If you want to understand things: start with the illustration. The Illustrated Stable Diffusion by Jay Alammar (30–45 min). Then Lilian Weng's What are Diffusion Models? (45–60 min), which is where the maths starts and is honest about it.
- Then the papers, only if the two above stuck. DDPM (Ho, Jain, Abbeel 2020) and the Latent Diffusion paper. These are the primary sources for everything in the two articles above.
- Editing is a separate skill from generating. The Diffusers docs on Inpainting and Image-to-Image (30–45 min each). Most professional use of image models is editing an existing image, not conjuring a new one, and it is the part tutorials skip.
- Voice and video carry a consent problem. Before ElevenLabs' cloning docs, read its safety and voice-cloning consent policy in the official-sources module. Cloning a voice you do not have permission to clone is the most common way a hobby project becomes a legal problem.
What commonly goes wrong: Judging a model by a leaderboard. The Artificial Analysis leaderboards here are useful for narrowing a shortlist and misleading as a verdict — they measure aggregate preference on generic prompts, which is not the same question as whether a model can do your specific job.
48 resourcesAutomations & Agents
Who this path is for: People who want work to happen without them, and who are one bad workflow away from learning why that is harder than it looks.
- Read the sceptical piece first. Anthropic: Building Effective Agents (25 min). Its core argument — that most tasks want a simple, fixed workflow rather than an autonomous agent — is the most useful thing in this module and it is best absorbed before you are emotionally invested in an agent.
- Then learn one platform properly. Pick n8n or Make and stay there. For n8n: the freeCodeCamp beginner's guide (45 min), then n8n Docs: Flow Logic (30 min). For Make: Make Academy's Make Basics path (3 hrs). Learning two platforms halfway is worse than learning one.
- Then learn to break it on purpose. n8n Docs: Error Handling (20 min) and How to Use Sub-Workflows (20 min). This is the step everyone skips. An automation without error handling does not fail loudly, it fails silently for three weeks.
- Then copy a template that resembles your problem. The RSS, price-monitoring, lead-management and chatbot templates here are all working starting points. Read the workflow before you run it.
- Multi-agent last, and sceptically. The IBM multi-agent explainers (15–25 min each) describe a real architecture. Most problems presented as multi-agent problems are one workflow with a bug.
What commonly goes wrong: Automating a process nobody has written down. If the manual version is not documented, the automated version will encode whatever the last person did by accident. Do the Playbooks page's manual level first — that is exactly what it is for.
28 resourcesVibe Coding & No-Code
Who this path is for: Non-engineers building working software, and engineers trying to understand what their non-engineer colleagues are shipping.
- Read the definition and the objection together. Karpathy's original tweet (5 min), then Simon Willison's Not all AI-assisted programming is vibe coding (10 min). Willison's distinction — vibe coding means accepting code you have not reviewed — is the one that matters, because it tells you exactly when the practice is fine and when it is not.
- Write down what you are building before you prompt for it. The Formlabs or Atlassian PRD templates (15–20 min). A one-page spec is the difference between a tool that generates your app and a tool that generates a plausible app.
- Then pick one builder. Lovable's Quick Start, Bolt's intro, v0's docs, or Replit's Agent docs. The Lovable Prompting Bible (20 min) is the most transferable of these — its structure works in any of the four.
- Then learn the parts you are pretending not to have. Supabase's Getting Started and its Solutions for Vibe Coders page (20 min), plus Auth.js Getting Started (20 min). Your generated app has a database and an auth system whether or not you chose them.
- Then the security minimum. GitHub Docs: Storing your secrets safely (10 min) and Finding and fixing your first code vulnerability (20 min). Non-negotiable before anything is public. A generated app will happily commit an API key.
What commonly goes wrong: Shipping something with a database that anyone can read. Generated apps routinely leave database access rules wide open, and this is not something the model will mention. If your app stores anything about anyone, the Supabase docs on access rules are the page to read before launch, not after.
28 resourcesBusiness Applications & Governance
Who this path is for: Anyone who has to defend an AI decision to someone else — a client, a regulator, a board, a licensing body.
- Start with the framework everyone else cites. NIST AI Risk Management Framework 1.0 (30 min), then the Generative AI Profile (20 min). Almost every AI policy you will be handed is derived from these; reading the original means you can tell a real policy from a rewritten one.
- Then the playbook, because the framework alone is abstract. NIST AI RMF Playbook (20 min). It converts the framework's functions into actions you can actually assign to someone.
- Then a policy you can adapt today. FRSecure's free AI Acceptable Use Policy template (15 min). Adapt it; do not adopt it unread.
- If you are in a regulated profession, read your own authority. ABA Formal Opinion 512, the State Bar of California guidance, the NYSBA task force report, and Florida Bar Opinion 24-1 are all here in full. They disagree with each other in places. That disagreement is the actual state of the field, and a summary that smooths it over is worse than useless.
- Then read the tool terms as evidence, not as marketing. The training-data and privacy pages for Claude, OpenAI, Gemini and Notion AI are collected in this module precisely so a procurement conversation can cite the vendor's own words.
What commonly goes wrong: Treating a vendor's marketing page as a data-handling commitment. The pricing and privacy pages collected here differ from the landing pages in ways that matter, and the difference is usually the answer to the question you were asked.
45 resourcesAI Coding Agents, Media & Voice Tools — Official Sources
Who this path is for: Anyone comparing tools, and anyone who needs a citation rather than an opinion.
- Use this module as a reference, not a reading list. These 45 links are official pricing, privacy, data-usage and policy pages. Nobody should read them front to back. Come here with a question.
- For a buying decision, read three pages per tool. The pricing page, the privacy or data-usage page, and — for anything generative — the content or safety policy. Those three answer what it costs, what it does with your input, and what you are allowed to do with the output.
- For a data question, the phrasing to look for. Whether your inputs train the model by default, and whether the answer differs by plan. It usually does, and the free tier is usually the permissive one.
- For media tools, check provenance. Google DeepMind's SynthID page and Adobe's Firefly IP-indemnification terms are the two documents that most often change a decision, because they answer whether output is detectable and who is liable if it infringes.
- Expect these to rot faster than anything else on the site. Pricing and policy pages are the most volatile links in the library. The OpenAI Sora discontinuation notice in this module is here as a reminder that products get withdrawn.
What commonly goes wrong: Quoting a price from memory. Every figure in this module has a date attached to it at the source, and none of them has one here — that is deliberate, and it is why the link goes to the vendor rather than to a table on this page.
The full catalogue
Every resource, grouped as it is curated. The search box below is a convenience for people with JavaScript; the complete list is on the page either way.
The ten guide pages on this site 10
Included so one search box covers everything here, not only the outbound links.
Not sure where to begin? Answer what brings you here and how much time you have, and this page routes you to the right pages — no account, no email, no sales pitch.
A sourced timeline of artificial intelligence from Turing's 1950 thought experiment through today's frontier models — 17 turning points, each with primary-source links.
Neural networks, training, attention, tokens, and multimodality explained in plain English — plus a visual map of how AI, machine learning, deep learning, and generative AI actually relate.
The real working methods: how to structure a prompt with XML tags, the magic-prompt technique, image and video prompting frameworks, agent design principles, and the vibe-coding workflow — with worked examples.
Sixteen concrete, worked use cases for applying generative AI — research synthesis, meeting notes, content drafting, learning, code help, email triage, document comparison, brainstorming, customer support, spreadsheet help, presentations, translation, performance reviews, study guides, document data extraction, and resume tailoring — each with a real example and human-review notes.
Decision support, not reviews: strengths, limitations, privacy, and pricing considerations for seven general AI assistants — ChatGPT, Claude, Gemini, Perplexity, Microsoft Copilot, Grok, and Meta AI — plus how to build a lean AI stack.
Ten full AI implementation playbooks — manual, semi-automated, and fully-implemented versions of each workflow, with step-by-step build instructions, risks, and KPIs.
Eighteen free, ungated prompt templates across Strategy, Operations, Marketing, Client Work, and Governance — each one designed to be pasted directly into ChatGPT, Claude, Gemini, or any LLM, which does the work for you.
Where AI stands in mid-2026: the closely matched multi-lab race, the real numbers from Stanford HAI's 2026 AI Index, and the genuinely unresolved debate over artificial general intelligence between named researchers on both sides.
Why Zero → Frontier exists, the sourcing standard it holds itself to, how it was built, and how to report a broken link or an error.
Foundations & Prompt Engineering 27
↑ Read the path through this module first
Core Concepts
Google's official primer on what generative models are and how to talk to them, for total beginners.
Hand-animated walkthrough of tokenization, embeddings, and next-token prediction.
How query/key/value self-attention lets tokens 'read' each other.
How text gets broken into tokens before reaching the model.
What an 'agent' is (model + instructions + tools).
Anthropic's own 9-chapter hands-on course — the best free substitute for the paid prompting module.
Anthropic's condensed reference on clarity, examples, XML structuring.
OpenAI's official prompting strategies.
Standard and zero-shot CoT with worked examples.
In-context learning via examples.
Using a model to generate/refine other prompts.
Theoretical grounding for meta-prompting.
Role, background, task, instructions, constraints as layers.
Applies the layered context framework to agentic workflows.
Anthropic's own guide to XML-tag prompt structuring.
Scoping a real problem and building the prompt to deliver it.
Tools
Official onboarding docs covering search, Focus modes, citations.
Detailed walkthrough of Focus modes and free-tier limits.
Official community beginner tutorial on scenarios and triggers.
Builds a first scenario end-to-end.
Official step-by-step for building a Custom GPT.
The instruction-writing skill underlying CustomGPT design.
Install, run a model from the terminal, call the local API.
Local LLMs, hardware requirements, Modelfiles.
Setup/usage guide including temperature and system prompts.
Build a chat prompt and test Gemini models free in-browser.
Complete tour of AI Studio.
RAG & MCP 33
↑ Read the path through this module first
RAG & Retrieval
The five-stage retrieve-augment-generate process.
Conversational episode breaking down RAG fundamentals.
IBM Research's original explainer.
How embeddings represent meaning as dense vectors.
How semantic search differs from keyword search.
Full step-by-step build of a local RAG pipeline.
Indexing, retrieval mechanics, query translation, generation.
Builds a minimal RAG system in Python using Ollama.
Series covering each RAG design decision.
Official quickstart for the API/Playground interface.
Test Gemini prompts and generate API code, free.
The free Deep Research feature that browses and synthesizes cited reports.
Claude Projects — free workspaces for uploading knowledge.
Step-by-step for creating a Project and uploading files.
Free Community Edition with full RAG pipeline and agent features.
Official open-source repo for self-hosting Dify free.
Run LLMs entirely on your own machine, private and free.
LlamaIndex's own '5 lines of code' starter tutorial.
Same tutorial wired to local models via Ollama.
Official docs tutorial on indexing and retrieval-generation.
Notebook series stripping RAG down to fundamentals.
Central free hub of articles on vector databases.
Official quickstart for the open-source Weaviate vector database.
Free Postgres extension turning Postgres into a vector store.
How a reranker improves RAG output quality.
MCP & Connecting AI to Tools
Anthropic's original announcement — primary source.
Official intro page: the USB-C analogy.
MCP architecture and transport methods.
Official quickstart building a working MCP server.
Wiring an MCP client to locally running servers.
Official catalogue of reference/community MCP servers.
Full open-source repo of MCP servers (Notion, Gmail, Slack, GitHub).
Beginner-friendly video on MCP setup.
Image & Video Generation 28
↑ Read the path through this module first
How Diffusion Models Work
Free four-unit course teaching diffusion theory from scratch.
30+ diagrams showing how CLIP, UNet, and the decoder work together.
The original Denoising Diffusion Probabilistic Models paper.
The official Stable Diffusion research paper.
Deep technical walkthrough of forward/reverse diffusion.
Build and train a small diffusion model from scratch.
Line-by-line annotated PyTorch implementation of a DDPM.
Constantly-updated Elo ranking of image models.
Image Generation & Prompting
Official guide to masked inpainting with open checkpoints.
Conditioning generation on an input image plus prompt.
SDXL's dual text-encoder architecture, and how to prompt each encoder separately.
Natural-language prompting for Flux Schnell/Dev.
Community breakdown of keyword-style prompting.
Thousands of free browser-based demos of SDXL, Flux, etc.
Video Generation & Prompting
Free intro course covering the AI filmmaking pipeline.
Updated tutorials on the latest video models.
Elo rankings of Kling, Veo, Seedance, Runway Gen.
Cinematic language: camera moves, lighting, shot type.
Turning smartphone photos/video into 3D models.
Practice text-to-video and image-to-video for free.
Voice/Avatar Cloning
Free-tier-compatible guide to cloning a voice instantly.
Instant vs. professional cloning explained.
Free-tier text-to-speech and multilingual generation.
Beginner tutorial for the free-tier cloning flow.
Open-Source Models (Hugging Face)
Hands-on unit teaching the open-source Diffusers library.
Flagship free course on using open-source models.
Selecting and deploying open Hugging Face models.
Free searchable directory of open-source models.
Automations & Agents 48
↑ Read the path through this module first
Automation Fundamentals
Vendor-neutral explainer of RPA.
Plain-language breakdown of RPA use cases.
Official tutorials and feature walkthroughs.
Install n8n, build a real workflow, deploy to production.
Official reference for triggers, branching, loops.
3-hour beginner-to-intermediate crash course.
Free self-paced academy covering modules and functions.
Official foundation course on the scenario builder.
Make basics through building working workflows.
Create cloud flows and agent flows.
Central hub linking every official Power Automate path.
The RPA/desktop-automation side of Power Automate.
Building Specific Automations
Reads RSS, summarizes with GPT, archives/notifies.
Runs offline with a local, free LLM.
Beginner-narrated build log of a first AI news workflow.
Scraping patterns for competitor/product data.
Scrapes competitor prices and alerts on changes.
Benchmarks products/prices across sources.
Automating lead capture and CRM syncing.
A lead-capture-to-CRM pipeline with notifications. Medium member-only.
Official lead-management templates to clone and adapt.
Building an AI-powered support chatbot with memory.
No-code build of a support chatbot with an LLM. Medium member-only.
Embedding a custom-branded chat widget on any website.
Build then embed and deploy a chatbot live.
Error Trigger nodes and error workflows.
Calling one workflow from another for reusable logic.
Agent Design Principles
Anthropic's canonical guide to workflows vs. agents.
Rules for designing narrow, well-documented tools.
Building checkpoints where a human approves an agent's step.
What separates agentic systems from simple automations.
Structured Anthropic-run course on MCP fundamentals.
Multi-Agent Systems & Voice Agents
MAS architectures, structures, and real use cases.
How an orchestrator agent coordinates sub-agents.
How agents share context and hand off tasks.
Building a true multi-agent system inside n8n.
A 'Jerry'-style AI employee: email, calendar, CRM.
Email, calendar, and CRM combined into one assistant.
An agent that answers calls and books calendar events.
Orchestrator + research agent + writer agent build.
Reducing token waste and structuring agent handoffs.
Official docs introducing flows and nodes.
Hands-on course with a free-credit tier.
Documentation covering agent concepts and triggers.
Building a first Lindy agent on the free plan.
Free quickstart for building a voice agent from zero.
Step-by-step video on Vapi's free trial tier.
Fields inbound calls and books calendar appointments.
Vibe Coding & No-Code 28
↑ Read the path through this module first
Concepts & Best Practices
Primary-source definition of vibe coding.
Clarifies what vibe coding is/isn't.
How the term evolved into a professional workflow.
The describe-generate-preview-refine loop, live.
Free downloadable PRD template and guide.
Structured, ready-to-copy PRD template.
Spotting and patching a real vulnerability.
Keeping API keys/secrets out of committed code.
Tools
Free Hobby plan, no card required.
What Bolt is and how to go from prompt to published app.
Official getting-started guide for the free tier.
Writing effective prompts for a prompt-to-app builder.
Official docs for v0's free-tier AI UI generator.
Hands-on tutorial building a real app with v0.
Describe an app in plain language and let Agent build it.
Free tier with unlimited Tab completions.
Agentic AI pair-programming/debugging workflows.
Setup and the core agentic terminal workflow.
Building Blocks (Frontend/Backend/DB/API/Auth)
100% free full-stack certification.
Node/Express APIs certification.
HTML/CSS/JS through Node/Express/MongoDB, end to end.
Free Postgres database + backend instantly.
How v0/Lovable/Bolt connect to Supabase.
Free, open-source authentication for Next.js.
Deployment & Version Control
Repos, branches, commits, and pull requests.
Version control with practical command examples.
Deploy globally with git push.
Shipping an app to a live URL.
Business Applications & Governance 28
↑ Read the path through this module first
Research Tools — Official Sources
Official data-use and training-opt-out terms for Perplexity's answer engine.
Confirmed free-tier notebook, source, and daily-chat caps, straight from Google.
What's free vs. paid for Elicit's 138M+ paper systematic-review tool.
Confirmed free-tier Pro messages, Deep reviews, and Study Snapshot caps.
Google's own page on the agentic multi-source report-generation feature.
Meetings & Knowledge Tools — Official Sources
Free-tier minutes, imports, and AI Chat query limits, direct from Otter.
Free-tier storage minutes, AI credits, and meeting-length caps.
Fathom's 'free forever' recording tier and its stated limits.
The only major notetaker with no bot in the call — free-tier terms.
How Notion's Enterprise Search/AI Connectors handle your data by default.
Writing Tools — Official Sources
Anthropic's own explanation of Claude's training-data defaults and opt-out.
ChatGPT's official data-use and opt-out policy.
The confirmed exact free-tier prompt cap (100/month) for Grammarly's AI features.
Default training/human-review settings for the Gemini app and Docs' 'Help me write.'
Coding Tools — Official Sources
Hobby-tier terms for Cursor's VS Code fork with agent mode.
Free-tier completions/chat allowance and March 2026 training-data policy update.
Starter-tier terms for Replit's autonomous full-stack Agent.
Confirmed exact free-tier credit and message-per-day caps.
Confirmed exact free daily/monthly credit caps for full-stack app generation.
Confirmed exact free-tier token caps for StackBlitz's in-browser app builder.
AI Governance — Primary Sources
The official, free, voluntary U.S. framework this site's policy template is built around.
NIST's own practical, actionable companion to the AI RMF.
NIST's 2024 supplement addressing generative-AI-specific risks.
A genuinely free, no-login starter policy template for small businesses.
AI Governance — Regulated Professions
The first ABA ethics opinion on generative AI: competence, confidentiality, candor, supervision.
California's official practical guidance for lawyers using generative AI.
NYSBA's April 2024 comprehensive report and recommendations on AI in legal practice.
Florida's official ethics opinion on lawyers' generative AI use.
AI Coding Agents, Media & Voice Tools — Official Sources 45
↑ Read the path through this module first
Coding Agents — Official Sources
Setup, sandboxing model, and surfaces (CLI, IDE, cloud, GitHub) for OpenAI's coding agent.
Token-based credit pricing for Codex across ChatGPT plans and the API.
Confirms API/Business/Enterprise data is not used for training by default.
Technical docs for Anthropic's agentic coding tool across terminal, IDE, and desktop.
Exact training/retention policy differences between consumer and commercial Claude Code accounts.
Official setup guide and stated limitations for Anthropic's desktop file/task agent.
Anthropic's own permissions and safety guidance for Cowork's file-editing capabilities.
Free/Pro/Teams/Enterprise terms for Cognition's agentic code editor.
Source code and documentation for the open-source, model-agnostic coding agent.
Exactly what Cline does and doesn't collect, and how to opt out.
Confirmed exact free-tier message/integration credit caps for Wix's no-code app builder.
Data handling, SOC 2/ISO 27001 certification, and EU/UK residency terms.
Free/Pro/Max/Teams terms for Cognition's autonomous software engineer.
Google's agentic development platform with Editor and Manager surfaces, launched with Gemini 3.
Confirmed exact free-tier task and concurrency caps for Google's async coding agent.
Image Generation — Official Sources
Subscription tiers and Stealth Mode terms; confirms no free tier on Discord/web.
Confirms all prompts/images train Midjourney's models by default, no opt-out.
Product overview for GPT-image conversational generation and editing.
Technical docs for 'Nano Banana' (Gemini 2.5 Flash Image) and Imagen.
How Google's invisible AI-content watermark works across image, video, and audio outputs.
Generative-credit tiers and pricing for Adobe's 'commercially safe' generative model family.
Exact scope of Adobe's IP indemnification — which features/plans it covers and excludes.
Confirmed exact free-tier daily token allowance (150/day) for Canva-owned Leonardo.
Training-use terms for public vs. private generations, and the free-tier public-content license.
Video Generation — Official Sources
Confirmed exact free-tier one-time credit allowance for Runway's Gen-4/4.5 video suite.
Google's AI filmmaking studio built on the Veo model family.
Free daily credit allowance and paid tiers for Kuaishou's video generation platform.
Confirmed exact free-tier video count/length caps for HeyGen's AI avatar platform.
States user video/voice/face data trains HeyGen's models by default, opt-out via email.
Free (Basic) tier terms and paid tiers for corporate-training avatar video generation.
Official notice and timeline for the 2026 shutdown of OpenAI's Sora video product.
Voice & Audio — Official Sources
Confirmed exact free-tier credit allowance (10,000/month) for TTS, cloning, and dubbing.
Identity-verification and consent requirements before cloning any voice.
Confirmed exact free-tier daily time cap (2 hours) for ChatGPT's Advanced Voice Mode.
Free vs. Premium tier feature comparison for Speechify's text-to-speech app.
Confirmed exact free-tier daily credit allowance for Suno's AI song generator.
Confirms user content trains Suno's models by default with no self-serve opt-out.
Confirmed exact free daily/monthly credit pool for Udio's music generator.
Google's free browser-based music generation tool, powered by the Lyria model family.
General Assistants — Official Sources
Free vs. Microsoft 365 Premium terms for Microsoft's everyday AI assistant.
Default training settings and opt-out mechanics for consumer Copilot.
Free tier and SuperGrok/X Premium terms for xAI's assistant.
Default training-use terms; confirms logged-out users cannot opt out at all.
Overview of Meta AI across WhatsApp, Instagram, Facebook, and Messenger.
Meta's disclaimers and terms governing Meta AI use across its apps.
One page left — who built this, and why?
The sourcing standard, the FAQ, and how to flag a broken link or an error.