The Best AI Tools for D&D in 2026: An Honest Review
The honest 2026 guide to AI tools for D&D. Eight recommendations by job (prep, rules, voices, art, maps) with every tool's drawbacks named, ours included.
The honest 2026 guide to AI tools for D&D. Eight recommendations by job (prep, rules, voices, art, maps) with every tool's drawbacks named, ours included.
If you typed “best AI tools for D&D” to find this post, you have seen the problem. When I checked the top ten results for that exact query on July 14, 2026, every one came from a company selling one of the tools it ranked. Several vendors occupied multiple spots with near duplicate pages, while claims such as “honest,” “tested,” and “no paid placements” had become standard boilerplate. The search page looks like independent comparison shopping until you notice that nearly every guide is a sales conversation wearing a reviewer’s cloak.
Most lists of AI tools for dungeon masters are vendors talking about themselves and their competitors, and this one comes with a bias too. I am Scrydiver, I build ScryRPG, and ScryRPG sits at number three on this list. You will find its drawbacks named as plainly as everyone else’s, including the situations where another product is the better choice. When that happens, I will tell you which competitor to use by name, because disclosure means little if the recommendation bends whenever it reaches your own product.
Once the duplicated roundup pages are stripped away, the field is smaller and more practical than the search results suggest. The tools worth comparing cluster around eight entries, each suited to a particular job: campaign continuity, encounter preparation, improvisation, character work, maps, images, audio, or a narrower utility that saves time at the table. Ranking them without regard to those jobs produces nonsense. A brilliant generator can be a poor campaign manager, while a useful campaign manager may be the wrong place to create the dungeon map you need before Friday.
My contract with you is simple: every recommendation must answer what the tool does well, what it fails at, and who should skip it. By the end, you will know which tool to pick for which job, what to avoid, and where every option on the list reaches its limit. First, though, we should deal with the objection hanging over the entire category: whether using AI is ruining D&D.
Some of you arrived here suspicious of AI in tabletop games, and I understand why. Generative AI that tries to replace the human DM is a real concern, because it changes who interprets the world, responds to player choices, and gives the campaign its particular voice. That category exists, and pretending otherwise makes any recommendation less credible.
Still, “Is AI ruining D&D?” is too blunt a question to produce a useful answer. The better question is what a particular AI does at your table. A system attempting to occupy the DM’s chair deserves different scrutiny from a tool that sorts notes, recalls an NPC’s last appearance, or handles repetitive preparation so the DM can spend another hour on encounters, motives, and consequences. I make the full argument elsewhere, along with a practical look at where AI helps a working DM without taking over the work that gives a campaign its character.
The question is not whether AI belongs at your table, but which job you are allowing it to perform.
That distinction guides every review below. I have organized the tools by the job each one performs, and when a product crosses from assisting the DM into replacing one, I say so plainly.
Here is the 30-second version for skimmers: eight tools, each owning one job. A longer answer for each lives below the table, including the detail and the failure modes that matter once you bring these tools to a real campaign.
| Tool | Job | Free tier | Best for | Honest drawback |
|---|---|---|---|---|
| ChatGPT + Claude | General prep and brainstorming | Yes, both | NPC dialogue, encounter ideas, lore drafts | Can misstate rules with confidence; memory is selective, not your campaign record |
| NotebookLM | Rules lookup from your own books | Yes | Cited answers from PDFs you upload | Retrieval only, no creative help |
| ScryRPG | Party-shared AI familiar on your real campaign | Yes, generous; Discord bot free | One AI conversation your whole party shares, imports, real campaign actions | Not a VTT and not an AI DM; no native mobile app |
| Friends & Fables | AI Game Master | Yes (3 players) | AI-run 5e campaigns when no human DM is available | The AI is the DM, wrong fit for tables that have one |
| ElevenLabs | NPC voices | Yes (10k credits/mo) | Distinct voices for recurring NPCs | Credits go fast at the table; setup cost is real |
| Archivist | Session recording into queryable memory | 30-day trial only | Asking your own sessions what happened | From $10/mo per campaign; audio-first |
| Midjourney + Gemini | Character and NPC art | Gemini yes; Midjourney no | Painterly portraits (MJ), free fast iteration (Gemini) | Character consistency still drifts everywhere |
| Dungeon Alchemist | Battle maps with VTT export | No ($44.99 one-time) | Foundry and Roll20-ready maps in minutes | Procedural, not generative; maps get samey |
ChatGPT and Claude are the obvious starting point for most people exploring the best AI tools for D&D, and pretending otherwise would make this list less useful. They occupy roughly the same place in my prep: NPC dialogue, plot and encounter brainstorming, lore drafts, and summaries pulled from messy session notes. Pick the one whose answers and interface suit you, then learn how to give it context and constraints.
I prefer Claude when I want long-form prose that needs to hold a consistent voice across several pages. Its paid plans now carry a 1-million-token context window on the default model, which is enough room to work against whole sourcebooks rather than feeding material in fragments. Claude has a free tier, Pro costs $20 per month, and Max runs from $100 to $200 per month.
ChatGPT has the broader ecosystem, including custom GPTs and built-in image generation. It also has a free tier, while Plus costs $20 per month and Pro costs $200 per month. That top plan makes sense only if you use AI throughout the day and can point to a workload that justifies it. The current defaults are GPT-5.5 in ChatGPT and Claude Sonnet 5, though model names age too quickly to be a useful basis for choosing between them.
Neither belongs behind the DM screen as a rules authority. Both can misstate D&D rules with total confidence, so verify anything that touches math at the table. Their memory has improved considerably in 2026: both now offer cross-chat memory and Projects, which means the old complaint that they simply forget your campaign between sessions is no longer accurate. Recall remains selective and fallible, however, and whatever they remember lives inside one person’s private account rather than your party’s shared record. Keep inventories, gold totals, and custody of the macguffin somewhere canonical.
I have found most custom GPTs advertising D&D mastery to be thin prompt wrappers. You can usually write a better system prompt yourself in three minutes.
Rules accuracy is precisely where these general-purpose workhorses stumble, and that seam is what the next entry closes.
NotebookLM is the underrated answer to “ChatGPT made up a rule.” Load your sourcebook PDFs and homebrew documents into a notebook, then ask questions in plain language. Its answers stay grounded in those sources and include inline citations, so you can inspect the relevant passage instead of trusting a confident paraphrase assembled from who knows where.
During play, that makes rules lookup far less disruptive. While prepping, you can search across several books without remembering which chapter contains the obscure interaction you need. It works just as well on your own setting notes, house rules, and faction documents. If your prep involves asking “what is the rule for X?” more than twice a session, this changes that experience entirely.
The cult workflow combines session audio transcripts, homebrew documents, and sourcebook PDFs in one notebook. “What did the innkeeper promise in session 9?” and “How does grappling work?” can then be answered from the same place, with citations pointing back to the campaign record or rules text. NotebookLM can also pull in web sources when asked, although its grounded, inspectable answers are why it belongs among the best AI tools for D&D.
The standard tier is free, with limits of 100 notebooks, 50 sources per notebook, and up to 500,000 words per source. Higher limits come through Google AI subscriptions starting at $9.99 per month; there is no longer a standalone NotebookLM Plus product.
Its drawbacks are straightforward. NotebookLM offers no creative help, only retrieval and synthesis. You need to own the PDFs you upload, and asking it something trivial can feel slower than firing the same question at ChatGPT. Google does not market it as a TTRPG product, which is probably why so many vendor listicles miss it.
Neither ChatGPT nor NotebookLM knows what your party is doing. They cannot see your characters, inventory, or shops. That is a different category of tool, and it is the next entry.
ScryRPG is the product I build, so my bias is plain. Its strongest idea is also easy to distinguish from the other AI tools for Dungeon Masters: the whole party shares one live conversation with the same familiar. Each message identifies its speaker, permissions follow that person, and the familiar works from shared campaign context. ChatGPT and Claude give each player a private chat; Friends & Fables puts an AI DM in charge. ScryRPG gives a human-run table one familiar that everyone can talk to.
The campaign manager holds your characters, inventory, shops, loot, lore, and encounters in one shared place. You can import PDFs, session notes, spreadsheets, or an Obsidian vault, then review the characters, items, shops, lore, and relationships ScryRPG extracts before approving them. The result is structured campaign records rather than another summary buried in chat history.
That structure matters because ScryRPG’s AI familiar can act on the campaign. Once you enable changes, it can spawn loot, update inventories, edit lore, and stock shops. Routine safe changes can be pre-authorized, while destructive or bulk actions always require approval. This distinction between describing an update and performing it is central to why I think the campaign was always the category.
Discord is a first-class surface, with the same familiar, party conversation, shops, loot, and inventory available from the server where you run your game. The bot costs nothing to add, and its core commands are free. There is also a real Foundry VTT module: pair a world to a party once, and characters, inventory, shops, loot, and encounters sync into Foundry. Player and GM views are filtered server-side. Sync works on every tier, with writeback on paid tiers.
ScryRPG is not an AI DM, a VTT, or a session recorder. Friends & Fables owns the AI DM lane, Foundry remains your VTT, and Archivist owns session recording. ScryRPG lives beside the human DM and holds the campaign rather than the audio.
The drawbacks are real. There is no native mobile app; the web app installs as a PWA and Discord covers couch play, but you will not find an app-store version. The encounter tracker records initiative and HP for a deployed fight, though it does not roll for players, apply damage, or advance turns. It is not a tactical combat engine.
D&D 5e and Pathfinder 2e have the deepest support. Draw Steel includes native stat blocks plus hundreds of seeded creatures and hero abilities, but remains publicly labeled beta. Daggerheart has full character sheets without an official content library yet.
ScryRPG is free to use, with unlimited questions to your familiar and a monthly credit allowance for changes it makes. The Discord bot is free, while paid tiers add upgrades.
If a human DM runs your game, ScryRPG is the right shape of tool. If nobody at your table wants that job, the next entry is.
Friends & Fables is the most complete answer for a group that wants an AI to occupy the DM’s chair. Franz, its AI GM, runs a structured campaign with 5e-based mechanics rather than merely improvising a branching story. Checks, HP, conditions, inventory, initiative, and tactical combat live inside the web platform, alongside an integrated VTT and battlemaps. When nobody in your group wants to prep and run the game, this is the option I would choose.
The free tier supports three players. Larger groups need Starter at $19.95 per month for four players, Pro at $29.95 for five, or Legend at $39.95 for six; annual billing advertises two months free. Friends & Fables still has a legacy Discord bot, but development has moved to the web platform, so I would evaluate it on that experience rather than treating it as a Discord product.
The fit is unusually clear:
No human DM available?
Want mechanics with teeth, not loose collaborative fiction?
Have three to six players? Friends & Fables is for you.
Its strengths become liabilities when a human DM is present. Franz does not help that person run encounters or keep campaign context organized; it replaces them. Your campaign also lives inside a closed sandbox, and every turn sent to an AI carries a different social contract from declaring an action to a person across the table. Some groups will love the pace and freedom from scheduling around a DM. Others will miss the negotiation, surprise, and judgment that come from a human referee.
For solo play with a stricter rules engine, my pick is The Endlessness, launched in spring 2026 and actively updated, which enforces the 5e SRD with initiative, conditions, concentration, and death saves.
Most tables, though, want AI to voice one memorable NPC rather than run the whole campaign. In that category, one tool dominates.
ElevenLabs is the default answer when you want an NPC voice that sounds convincing enough to survive more than a single tavern exchange. The Foundry ecosystem treats it that way, too: at least three active modules, Archive of Voices Pro, Talking Actors, and ElevenLabs for Foundry, pipe its voices straight into the VTT.
There are two sensible ways to use it. You can prerecord key lines during prep, which gives you control over pacing and lets you discard awkward takes, or generate speech live through a Foundry module or soundboard. Live generation sounds tempting, but it demands more setup than most groups will tolerate once the novelty fades. Prerecording is the realistic workflow.
My underrated tip: record three lines for each major NPC. Ten minutes of prep can change the energy of an entire session.
Pricing now uses shared credits. The Free tier includes 10,000 credits per month, roughly 10,000 characters of standard speech. Starter costs $6 per month for 30,000 credits, Creator costs $22 for 121,000, and Pro costs $99 for 600,000. That free allowance sounds generous until one substantial NPC scene consumes most of it, so regular use pushes you toward a paid tier quickly.
Voice cloning is gated by tier, with instant cloning available from Starter and professional cloning from Creator. Cloning a real person without consent is out of bounds; I explain the broader standard in where I drew the lines.
You also need to commit to a voice for each recurring NPC. Changing it halfway through a campaign is jarring in the same way recasting a familiar character would be.
Recording your sessions is its own layer, and the next tool turns those recordings into something you can question.
Archivist owns the session-recording-to-campaign-memory category. It can record live sessions, including games on Discord, or ingest uploaded audio, play-by-post logs, and raw notes. From that material, it builds transcripts, recaps, timelines, quest records, NPC tracking, and a persistent memory you can question in plain language: When did the party first meet the Silver Claw, and what did they promise her?
Those questions are where Archivist stops feeling like a recap generator and starts earning its keep. Its published numbers include more than 20,000 sessions processed, 300,000 compendium entries, and 150,000 questions answered. It processes session audio without retaining the recordings afterward, which addresses one of my first concerns whenever a table invites software to listen in.
There is no ongoing free tier. The trial covers 30 days and two sessions, with no card required. After that, Casual costs $10 per month for one campaign and four sessions monthly; Seasoned costs $20 for three campaigns and ten sessions. Pro is $35, Legend is $60, and add-on passes are available. Those campaign limits can push a GM running several games into a higher plan quickly.
Archivist and ScryRPG are complementary. Archivist captures what happened at the table from the audio, while ScryRPG holds the living campaign that your familiar can act on: who has the sword now, or what the shop sells. A serious table could use Archivist for “what happened in session 14?” and ScryRPG for “what is in the party’s pack right now?”
The drawbacks follow directly from its focus. If you do not record your sessions, you lose most of its value, and it offers no creative prep help. Saga20 is a newer alternative with audio-upload summaries, wiki updates, and cross-session voice matching, though it cannot record Discord directly yet.
Visual prep is its own beast, and the next two entries handle it.
For character portraits and scene art, I see two practical lanes. Midjourney remains my aesthetic-first choice when the image needs to carry a reveal, establish a villain, or make the party stop talking for ten seconds. Its current V8.1 model produces 2K images, follows prompts better, and runs faster than its predecessors. Access starts at $10 per month for Basic, with Standard at $30 and higher plans beyond that; there is no free tier, and you can work through either the web app or Discord.
The recurring-character problem complicates that recommendation. Midjourney’s Omni Reference can help preserve a recognizable face across generations, but it only works with V7, not V8.1, and clothing, props, and small details still wander. I will not pretend there is a meaningful success percentage when results depend so heavily on the reference, prompt, and character design.
If you want free, fast iteration, the Gemini app is the stronger starting point. Its current image model is Nano Banana 2, available to signed-in users at no cost within usage limits. Paid Google AI tiers increase limits and resolution while adding Nano Banana Pro. Conversational editing, multiple reference images, and solid character consistency make it unusually practical for developing an NPC through several revisions. Calling it Imagen is outdated: the separate Imagen 4 API is deprecated, with shutdown scheduled for August 2026.
ChatGPT Images 2.0, available across free and paid ChatGPT, is a credible third choice for one-off NPC portraits, especially when you want to revise the image through conversation. For typography-heavy handouts such as posters, letters, and shop signs, Ideogram 4 and Recraft V4.1 lead the pack.
Every image tool drifts across repeated generations. Reference features reduce the problem, while a locally trained LoRA remains the power-user solution and nobody’s first evening. Test a recurring PC before committing an entire campaign’s visual identity to one tool.
The ethics conversation lives at where I drew the lines. If you publish commercially, remember that DMs Guild bans standalone AI art, asset packs, and AI-generated text, while AI art inside eligible books requires disclosure; Adobe Firefly is the rights-conscious lane.
Battle maps need a different tool entirely, because general image models still cannot draw a usable grid.
Dungeon Alchemist is my recommendation when you need a battle map that drops into your VTT without an hour of cleanup. Pick a theme and room type, and it furnishes the map in seconds; exports to Foundry, Roll20, and Fantasy Grounds preserve walls, lights, and doors. Roll20’s automatic import requires a paid Roll20 tier with API access.
Here is the correction most AI tool roundups skip: Dungeon Alchemist does not use generative AI. Its developers say plainly that it combines procedural algorithms with assets made by paid artists. Listicles keep miscategorizing it, but I am including it because people searching for the best AI tools for D&D are usually searching for a result, not an argument about taxonomy, and this produces the most usable version of that result.
It costs $44.99 once through Steam, with no subscription, and runs on Windows, macOS, and Linux. The tradeoffs follow directly from how it works. There is no web app, maps can start feeling samey across a long campaign if you never mix in other sources, and procedural generation can only arrange rooms it understands. It will not draw the peculiar, half-collapsed nautiloid shrine you can see perfectly in your head.
Pure prompt-to-image battle maps still struggle to maintain grids, exact geometry, and editability. Constrained approaches are moving faster: TT-RPG added sketch-to-map creation with VTT wall export this spring, while Roll20 shipped its procedural Instant Dungeon generator in May. Inkarnate also deserves mention because version 2.0 brought a much larger battle-map asset library, though you still place those assets by hand.
Foundry MCP Bridge is the right answer for technical GMs who live in Foundry. This free, MIT-licensed module gives an MCP-capable assistant such as Claude Desktop roughly 20 tools for reading and driving your world. I left it off the main list because it requires Foundry, a desktop AI, and a plugin chain, which puts it beyond what most readers want to assemble. Do not confuse it with the separate commercial Foundry API Bridge, which advertises 71 commands behind an API key.
Adobe Firefly occupies the rights-conscious lane, particularly if you publish commercially. Adobe trains it on licensed and public-domain material, while business plans can include conditional indemnification rather than a blanket promise. Custom Models, which can train a consistent character or style, entered public beta in March 2026. Its aesthetic tends to look conservative beside Midjourney, but that restraint may be an acceptable trade when provenance matters.
Hero Forge contains no AI, yet it remains the standard answer for posed minis, STLs, and static character portraits. The Hero Forge 2.0 Kickstarter raised $3.1 million, which is a fairly decisive vote from the tabletop audience.
Saga20 is the newer session-memory contender. Upload audio and it produces summaries, updates your wiki, and matches voices across sessions in 25 languages. It cannot record Discord directly yet, a meaningful omission for online tables that keeps Archivist ahead in this category.
Syrinscape is my answer when someone asks for soundscapes built through a path that is not AI. Its audio is curated and creator-produced, and the new browser-based Nova platform entered beta in June 2026 with remote sharing for online tables. That combination makes it especially appealing when you want authored atmosphere without generating the soundtrack yourself.
AI Dungeon. It is still alive and patched monthly, with tiers from free to $99.99 per month, but it remains freeform interactive fiction rather than D&D: no enforced sheets, checks, or combat math. Fine for solo storytelling with dice flavor, but not what you came to this list for.
The long tail of AI GM platforms. HyperWrite AI Dungeon Master, RPGGO, AI Realm, and MacerAI have not earned a slot over Friends & Fables. MacerAI added multiplayer in June and AI Realm keeps shipping, so the category has some life, but I would start with Friends & Fables for a group or The Endlessness for solo use; HyperWrite’s offering is a prompt tool inside a writing suite, not a platform.
Custom GPTs claiming “D&D mastery.” In my experience, these are thin prompt wrappers whose system prompt does nearly all the work. You can write a better one in three minutes and tune it to your campaign.
Suno for session music. Warner settled and licensed Suno in late 2025, but litigation involving Universal and Sony remained active as of a July 2026 filing. If you want table music without that asterisk, Syrinscape is the curated lane; I would revisit Suno after the remaining suits settle.
Waiting for D&D Beyond or Roll20 to ship AI. The walled garden future vendors keep hinting at does not exist. Hasbro’s CEO said this spring that generative AI is not on the D&D roadmap because players do not want it, Sigil shuts down at the end of October 2026, Roll20’s public roadmap contains no first party generative AI, and its May dungeon generator is procedural; Demiplane’s new Campaigns feature also ships without AI. Pick tools that exist.
World Anvil’s AI. There is no integrated AI product, and the team has said AI features are out of scope. I respect the stance, but you should stop waiting for it.
Technically, yes. Friends & Fables can run real 5e mechanics as an AI GM for a group, while The Endlessness offers solo play through a strict SRD engine. The more useful question is whether that experience matches what you want from D&D. AI run games have a different shape, closer to interactive fiction supported by real dice than a table guided by a human mind. That can be fun on its own terms. If you have a human DM, though, I would use the other seven tools to strengthen their prep and presentation rather than replace them.
Yes. You can assemble a capable starter stack without spending anything: ChatGPT and Claude both have free tiers, NotebookLM is free, Gemini offers free image generation, and ScryRPG has a generous free tier plus a free Discord bot. Friends & Fables also supports three players for free. Paid plans across the tools reviewed here range from $6 to $45 per month, but there is no sensible reason to subscribe to every product at once. Start with the free options, notice which recurring task keeps stealing your prep time, then pay for the tool that removes that specific nuisance.
I would start with ChatGPT or Claude alongside NotebookLM. Use the first for NPC concepts, rough encounter sketches, and the awkward brainstorming stage when you know the session needs momentum but have not found its shape. Use NotebookLM for rules questions because it answers from the books you provide rather than improvising from uncertain memory. I would skip AI GM platforms until you have run a few sessions yourself. Learning how to read the table, pace a reveal, cut an encounter short, and follow a player’s strange idea is the human craft worth developing first.
Several do, although the integrations differ. Dungeon Alchemist exports maps natively to Foundry, Roll20, and Fantasy Grounds. ElevenLabs can reach Foundry through several community modules. ScryRPG has a Foundry module in closed beta that syncs characters, inventory, shops, loot, and encounters into your world while keeping player and GM views separate. If you want AI operating inside Foundry itself, the free Foundry MCP Bridge listed in the honorable mentions is the clearest route. Friends & Fables takes another approach by providing its own VTT, so you do not connect it to an external tabletop.
The logistical half of prep compresses hard: statting NPCs, looking up rules, structuring notes, and organizing lore can move from an evening’s work to a focused pass. The creative half still belongs to you, especially the questions that give a campaign its pulse, such as what the villain wants and why the party should care. For most GMs, the right mix of tools can turn a four hour prep block into one or two, with the reclaimed time going toward motives, consequences, and table ready choices. The tools by job above show which part of that workload each product absorbs.
Path 1: You have not used AI for D&D before.
Start with the free tier of ChatGPT or Claude. Choose the prep task that swallowed your time last week, whether that was a recap, an NPC, or encounter ideas, and ask it for a first draft tonight. Give the experiment twenty minutes. You will learn more from revising one useful attempt than from comparing another dozen feature lists, and you will find where the friction in your prep lives without committing to a subscription or assembling a stack.
Path 2: You use ChatGPT for prep, but it keeps fumbling your campaign.
You have two seams to close, so choose the one that hurts. If rules answers keep sounding plausible but cannot be trusted, set up NotebookLM against your own books; it is free and fast. If the AI cannot see your characters, inventory, shops, or what the party did, stop tweaking memory settings. You have a category problem, because a private chatbot lacks the campaign context you need. That is what ScryRPG is built to solve.
Path 3: You are running an active campaign with a real party.
This is where I can be direct: use ScryRPG. It is free, the Discord bot is free, and ScryRPG’s AI familiar lives in one conversation your whole party shares, grounded in your characters, inventory, shops, and lore. Import your homebrew doc or Obsidian vault and watch it become structured campaign records, then run a shop visit from Discord on a Tuesday night between sessions. When a real party is moving through an active campaign, a shared familiar beats a private chatbot.
Pick your path, and if it leads here, ScryRPG is free to start. Bring the whole party.