Seedance 2.0 release

[Viridian]

Honorary mention goes to LTX 2.3. That one launched quietly. Output is pretty good with the same prompt and source image. Price is reasonable considering it renders at 1080p at the lowest, and is VERY fast. Maintains visual cohesion well, decent sound, decent physical interaction. It's text prompt adherence is somewhat lagging behind Kling 3.0 - like with most current models, it has to be told what to do and it does it, otherwise it will faceplant into the mud like Kling 2.6 used to do.

[MadMax359]

she's lovely!
what's the status of allowing nudity?

[Viridian]

she's lovely!
what's the status of allowing nudity?

Based on what Sekani has done, it can generate breasts but won't generate nipples, so you get that weird blur censor / ghost nipples. You can't upload nude images to animation as far as I know.

[Sekani]

she's lovely!
what's the status of allowing nudity?

T2V (text to video) allows nudity if you don't explicitly ask for it, but it'll blur the naughty bits. Example, if you say your female subject is wearing a loincloth and don't specify anything up top, you'll get bare breasts about half of the time.

I2V (image to video) is significantly more censored; even some of the non-nude images created with its own T2V model will trigger the censor if used as a source.

[MadMax359]

i'm confident you guy will push the envelope!

[Sekani]

This model may be cooked.

Apparently the creators got caught up in some identity theft lawsuits (the I2V was good enough to create accurate voices of real people without a voice sample), so their I2V model is currently blocking any reference image with a human face. There are some absurd hacks to bypass this on Reddit, but with so many alternatives this hassle doesn't seem worth the effort. There are even scams where sites are advertising "unrestricted" Seedance 2.0 access but actually using a different model for generations.

At the moment the T2V model is great for people who only care about Grok-level short clips, but there's no way to maintain consistency for anything longer than 15 seconds. I2V seems to only be successful for extremely stylized animated longer videos.

Anyway, Pixverse 6 and HappyHorse will be out in a matter of weeks so this may all be forgotten.