Why does AI art keep giving people extra fingers, and is it getting fixed?

AI image generators give people extra fingers because they don't understand what a hand is — they only understand patterns of pixels that usually appear near wrists. It's a counting problem, not a drawing problem. The model has seen millions of images of hands, but from every possible angle, partially obscured, holding objects, or blurred in motion. It learns that fingers are long, skin-colored shapes that cluster together. Sometimes the cluster has four shapes, sometimes six. The AI isn't counting; it's predicting the most statistically likely arrangement of pixels, and statistically, that blob near the palm could be one finger or two. Midjourney's V6 and DALL-E 3 have gotten much better at this. In my tests, V6 nails hands about 8 times out of 10 now, up from maybe 3 out of 10 a year ago. The fix came from better training data labeling — humans specifically tagged hands in training images so the model could isolate that pattern. But here's a tip most people miss: if you're getting mangled hands, try adding "hands resting on lap" or "hands clasped" to your prompt. Giving the AI a clear hand pose reduces the ambiguity that causes the extra-finger glitch. It won't always work, but it helps. The deeper issue — AI not truly understanding objects as 3D things — will take years to solve fully. **Related**: Will AI image generators ever understand anatomy? | What's the best AI image generator for realistic people?