i burned through way too many credits yesterday trying to fix a stupid little head turn.
not a fight scene. not a full short film. just a character looking over their shoulder without the jaw sliding sideways or the hair turning into neck soup.
i used to care a lot more about model rankings. after sora stopped being the obvious thing to compare everything against, i kept checking leaderboards like they were going to tell me what to use next.
they don't, really.
a model can have an insane demo and still make you pay for five dead runs before one clip is even close. face drift, hands going feral, motion that either does nothing or suddenly invents a new skull shape. all of that still costs credits.
and time.
i'm starting to think "cost per usable clip" is the only number i actually care about. not the listed price, not the prettiest launch video, not the benchmark screenshot. how many bad generations do i have to eat before i get one thing i can actually use?
i've been bouncing between runway, kling, and a few others. runway is where i usually test the messier motion passes, but i burn credits chasing the one clean take. kling has been better for face/skin stuff in a few runs, especially expressions, but the second i need one exact boring movement it turns into retries.
the thing with PixVerse is that it's not really one model. it feels more like a place to bounce between different options without restarting the whole search. having the same credits work across models makes low-res checks less annoying, especially when i'm trying to kill bad prompt ideas before they turn into expensive mistakes on a pricier tool.
still exhausting, though. every tool has its own way of making you pay for being slightly too specific.
how are people here measuring this now? do you count failed generations as part of the real price, or only the clips that survive?
[link] [comments]