RPG knowledge
Classes, armour, equipment, level progression, spell mechanics and encounter language.
Fantasy AI Model Leaderboard for D&D and TTRPGs
See which AI models understand the work Game Masters actually need. Every ranking comes from a blind A/B vote on fantasy images, video, audio or writing. Try one now: pick the better result, then see which models made them.
Finding a blind battle
Image board · liveScore · votes
| Rank | Model | Score | Votes |
|---|---|---|---|
| 1 | GPT Image 2 | 1,227 | 125 |
| 2 | Seedream 5.0 Pro | 1,174 | 116 |
| 3 | Nano Banana 2 Lite | 1,126 | 118 |
| 4 | Krea 2 Medium | 1,086 | 34 |
| 5 | Nano Banana 2 | 1,072 | 92 |
| 6 | Flux 2 Max | 1,042 | 49 |
605 counted votes · 16 rankedFull image board
Choose a specialist leaderboard
Compare models on one medium at a time. Each board uses the same blind-vote method, scores centred on 1,000 with a 95% range, and ranks a model from 20 counted votes.
01ImageCharacter, monster, battlemap and handout accuracy90 models605 battlesCurrent leaderGPT Image 202VideoMotion, continuity and readable fantasy action30 models23 battlesCurrent leaderNot ranked yet03AudioMusic, background scores, sound and spoken delivery22 models63 battlesCurrent leaderNot ranked yet04WritingAdventure, dialogue, lore and RPG knowledge23 models106 battlesCurrent leaderGPT-5.4 Nano
Built around RPG knowledge, not stock photography
A polished image can still fail the task. The Arena tests whether a model understands fantasy terms, tabletop constraints and the details that make an output usable in play.
Read the full methodFantasy accuracy
Gnolls, illithids and displacer beasts must look and behave like the creature in the prompt.
Table-ready structure
Battlemap geometry, cover, grids, handout legibility, scene continuity, music and TTS.
Creative quality
Adventure design, dialogue and lore are judged alongside style, clarity and prompt fidelity.
Vote first. Reveal second.
Two outputs receive the same prompt. You choose A, B, a tie or both fail before seeing the model names. Counted preferences update the standings.
Add a prompt of your own
Pick two models or choose a random match within your Gold limit. Completed viable outputs can enter future blind battles.