The improvised session
The party ignored the plot hook and went to the abandoned mill instead. A dnd map of the mill in two minutes is the difference between running the scene and hand-waving it.
One sentence · Top-down · Grid you can move minis on
Session starts in forty minutes and the party has gone somewhere you did not prep. Describe the place in a sentence and get a top-down battle map with rooms, corridors and a grid you can count squares on.
Both generated on this page's engine, from a sentence each. They are shown side by side rather than as a before and after because a dnd map has no before — the point is the range between an underground crawl and a settlement. Every output is AI-generated, so yours will differ.

Asked for a flooded ritual chamber with side rooms off it. What came back has the things that make a room playable rather than decorative: the water reads as a distinct surface you can rule on, the rubble gives cover positions, the braziers explain the light level without anyone writing it down, and the spiral stair in the corner is a second exit. Count the squares across the flooded chamber and you can price a dash across it before the session starts.
Put them in the flooded chamber
The harder of the two, because a settlement map has to show inside and outside at once. The roof is cut away over the common room so the bar, hearth, tables and stair are all placeable, while the stable, well and cart outside stay roofed. That cut is what makes a tavern brawl runnable — the party can go through the window, and everyone can see where the window is.
Give them a tavern to wreckThe bar is not beauty. It is whether four players can agree where everyone is standing without arguing about it.
Search for dungeon art and you get a thousand gorgeous images shot from a heroic low angle, none of which you can play on. A battle map is a different object with a different job: it is a shared coordinate system that happens to be pretty. Everything that makes concept art good — dramatic perspective, atmospheric depth, a focal point that pulls the eye — actively breaks a map, because a map has to be read the same way from four sides of a table.
So the constraints this dnd map generator works under are unusually specific. The camera sits directly overhead and never tilts. Walls read as walls at a glance, with enough thickness that a corridor and a room are visually different things. Furniture sits flat rather than in perspective, so a table occupies squares rather than floating over them.
And the grid stays faint — present enough to count, quiet enough that it does not fight the art underneath it.
Two honest limits before you rely on it. What comes back is a flat image, not a virtual tabletop module: no separate wall layer, no dynamic lighting data, no fog-of-war objects. You can drop it into Roll20, Foundry or Owlbear as a background and set your own grid alignment, but nothing is pre-tagged for you.
And it invents the floor plan rather than reproducing one — if a room in your campaign has to match something you drew six sessions ago, describe that layout explicitly rather than hoping. For characters rather than places, the trainer card maker is the closer neighbour.
Creative engine
Describe the location, choose a shape, and check the corridors before you commit to it.
"A flooded ritual chamber" produces a better dnd map than "a big room", because function implies furniture, wear and where the entrances would sensibly be.
Two doors and a stair means the party can be flanked. Say how many ways in and out you want and the plan will hold together around them.
A prompt for a battle map is a spec, not a mood board. These are the four things worth spelling out.
"A smugglers' cellar under a fishmonger" beats "a cellar". Function fills the room with the right clutter — crates, a winch, a hatch — and clutter is what gives players something to hide behind.

Three chambers off a central hall behaves very differently from three chambers in a line. Connection topology is the encounter design; state it rather than letting the model decide it for you.
Say you want a faint square grid across the whole map. Without it you get an illustration, and aligning a virtual tabletop grid to art that was not drawn for one is where most of the evening disappears.
Open the result full size and look at the narrow parts. If a passage is one square wide where you wanted two, that is a five-second reroll now and an argument mid-combat later.

Four situations, and all of them are about time rather than art.
The party ignored the plot hook and went to the abandoned mill instead. A dnd map of the mill in two minutes is the difference between running the scene and hand-waving it.
Published adventures ship with maps; the world you invented does not. Every location you write is a map you have to make, and that is the recurring cost that quietly stops people homebrewing.
A remote game has no theatre of the mind fallback that works for combat — everyone needs the same picture. Drop the image in as a scene background and align the grid once.
A map the party finds in-world is a different object from the one you run combat on — smudged, partial, drawn by a character. Ask for that look and it becomes prop rather than tool, the same way a stylised portrait turns a photo into an artefact.
What DMs check before dropping a generated map on the table.
The grid is drawn as an even square overlay, and the art underneath is composed to sit sensibly against it, but nothing is measured to a standard the way a hand-built virtual tabletop scene is. Treat the squares as the scale — a door is one square, a corridor one or two — and align your VTT grid to the image once when you import it.
Outputs are yours to use, including commercially. The caveat is a practical one rather than a licensing one: a map with generated art in a published product invites the same questions any AI asset does, so decide that on your own terms before you build a Kickstarter around one.
Because generated lettering is unreliable, and a room labelled with a misspelled word is worse than a room with no label. Every dnd map here is drawn deliberately without text so you can add your own numbers in whatever tool you prep in.
Not closely. Each run invents a plan from your description, so two runs of the same sentence give you two different buildings. If continuity matters, describe the layout you already established — room count, shape, where the stair is — and expect to reroll until it lines up.
Yes, just leave the grid out of your description. A gridless top-down map is the right choice for a hex-crawl overview or a handout, and for anything you plan to use in a VTT that draws its own grid on top.
Both work, and the tavern above is the mixed case — an interior cut away inside an exterior yard. Forests, roads, riverbanks and camps all render fine from directly overhead; the thing to watch is that large outdoor spaces need landmarks, or the map becomes an undifferentiated green field nobody can navigate.
One credit pool covers Nano Banana images and Seedance and Veo video. The cost shows before every run, and failed jobs are not charged. Use a subscription for ongoing work, or a one-time pack when you just need to top up.
What's included
Secure checkout by Stripe. Card details never touch our servers.
What's included
Secure checkout by Stripe. Card details never touch our servers.
What's included
Secure checkout by Stripe. Card details never touch our servers.
What's included
Secure checkout by Stripe. Card details never touch our servers.
One-time top-ups — buy extra credits any time you run low.
What's included
Secure checkout by Stripe. Card details never touch our servers.
What's included
Secure checkout by Stripe. Card details never touch our servers.
What's included
Secure checkout by Stripe. Card details never touch our servers.
Charges appear as “KAVEL AI” on your card statement. You can cancel any time from Settings → Billing; cancellation takes effect at the end of the period you already paid for. Operator details, the full model list, and the refund window are on the about page.
Powered by
Use one balance across every supported image and video model — all of them listed below — and check the credit cost before each request.
Video models
Image models
Model credit guide
You're only charged for successful generations. The exact estimate in the generator varies by model, length, resolution, audio, and number of images.
| Type | Model | Credit cost |
|---|---|---|
| Video | Seedance 2.5 | The long-take tier, priced per second: 5s 480p ≈ 211 credits, 5s 720p ≈ 473. A full 30s take runs ≈ 1,260 at 480p and ≈ 2,835 at 720p. Supplying a reference clip lowers the per-second rate. |
| Video | Seedance 2.0 | 5s 720p image-to-video ≈ 188 credits; text-to-video ≈ 308 credits. Scales with resolution and length. |
| Video | Seedance 2 Fast | Faster and lower cost. 5s 720p text-to-video ≈ 248 credits. |
| Video | Seedance 2 Mini | The cheapest Seedance tier. 5s 720p ≈ 154 credits, 5s 480p ≈ 72 credits. |
| Video | Seedance 1.5 Pro | Audio doubles the rate. 5s 720p ≈ 27 credits silent, ≈ 53 with audio; 1080p ≈ 57 and ≈ 113. |
| Video | Veo 3.1 | Billed per video, not per second (Lite tier). About 45 credits at 720p and 53 at 1080p. |
| Video | Kling 3.0 | Audio raises the rate. 5s 720p ≈ 105 credits silent, ≈ 150 with audio; 1080p ≈ 135 and ≈ 203. |
| Video | MiniMax H3 | Fixed 2K, no resolution ladder. Priced per second — a 5s clip ≈ 158 credits. |
| Image | Nano Banana 2 | Generate or edit from text and images. 20 credits per 1K image, 30 at 2K, 45 at 4K. |
| Image | Nano Banana Pro | Consistent run times across generations. 30 credits per 1K or 2K image, 50 at 4K. |
| Image | Nano Banana 2 Lite | Faster, simpler variant, and the everyday editing price: a flat 8 credits per image at every resolution. |
| Image | GPT Image 2 | The lowest-cost premium image model here. 10 credits per 1K image, 15 at 2K, 30 at 4K. |
| Image | Seedream 5.0 Lite | Flat pricing — 20 credits per image whether you generate or edit. |
Describe the place, check the corridors are the width you meant, and take it to the table — or design what lives in it first.