i'd actually push back a bit on relying too heavily on ai for anatomy learning. the thing is, ai generators are still pretty hit-or-miss with hands, joint angles, and weight distribution - exactly the stuff you need to nail down early. they're great for sparking ideas or getting unstuck when you're bored of the same poses, but they shouldn't replace actual reference work. that said, tools like Midjourney and Stable Diffusion can churn out human figures way faster than scrolling through pose sites, and sometimes you get genuinely usable silhouettes. just expect to spend time sifting through the weird proportions and anatomically impossible limbs.
if you do go this route, be really critical. look at the feet, the spine alignment, how the weight actually transfers through the legs. save the decent ones and compare them side-by-side with real anatomy references or photos. i've found that combining ai quick-sketches with actual photo references (pinterest boards, life drawing sites, even just poses you do yourself in a mirror) trains your eye way better than either alone. the ai stuff is faster for brainstorming new scenarios - like "person leaning against a wall" or "sitting on stairs" - but then you validate it against reality. that hybrid approach has helped way more than just trusting what the algorithm spits out.