This market assesses whether a general-purpose AI image-generation model can consistently produce a chess board with pieces in their correct starting positions by July 31, 2025.
Resolution criteria:
Success is defined as generating a proper chess board with all pieces in their correct starting positions 9 out of 10 times using a single prompt
Prompt engineering is allowed, but must be a single prompt fed directly into the model with the first output being considered one attempt
No other layers or post-processing allowed
The model must be a general-purpose AI image generation or multimodal model (not specifically designed for chess)
References:
1,000People are also trading
Claude 3.7 sonnet, with the prompt: Generate an image of a chess board, with all the pieces in their starting positions
@VedangManerikar Tbf generating a SVG seems pretty different from generating an image as a pixel raster
@spiderduckpig Especially because SVGs can just define a shape like a pawn in the text-representation format and repeatedly display it, solving the issue of consistency, and it can also just calculate the locations of all the pieces mathematically instead of implementing that in the image generation layer
https://sora.com/g/gen_01jq7fmm26eg2v1gfx29yt4rh1
prompt:
chess board initial position, view from above
no numeration
queen and king pieces clearly distinguishable
chess.com style illustration
@AndreiVlasenko yes, very close but imo the queen still isn’t clearly a queen given that it looks closer to the king than the white queen
gpt-4o native image generation (https://sora.com/explore/images) is extremely close
prompt: chess board initial position, view from above
https://sora.com/g/gen_01jq7e3g8vew6baekjxd4zt0gg