This July ranking of domestic big model platforms is pretty interesting, especially the image generation section — definitely worth talking about.

资讯 / AI 快讯
1

Just saw the July update for the domestic big model platform rankings. Rankings like this are just for fun, but the trends are worth noting—over the past year, the iteration pace for image and video generation among domestic players has clearly picked up compared to last year.

What I’m more curious about is how they’re ranking these: is it based on user numbers, capability benchmarks, or overall experience? The results can vary wildly depending on the metric. For those of us picking models for product use, looking at the overall leaderboard alone isn’t that helpful—you gotta break it down by specific capabilities: text-to-image, image-to-image, video, controllability—who’s strong in each area.

Anyone here using two or three platforms at the same time? Does the actual experience match the rankings?

2

Honestly, leaderboards and real-world experience are two totally different things. Just because a model scores high on benchmarks doesn’t mean you’ll get a good-looking image on a random roll.

3

When we’re picking a model, we only care about whether it’s stable for img2img. That overall leaderboard is totally useless for reference.

4

You hit the nail on the head with the aperture thing. A lot of those rankings are based on API call volume, which has nothing to do with actual quality.

5

Anyone got a breakdown of the different skill categories?