obsxrver

/

wan2.2-t2v-scat

Like 30

anyhow, if you see this OP, I'd be willing to pay for an I2V version that does not alter faces.

I think I know why it has problems with i2v, currently training a new one with an updated dataset that should make it work.
Will upload for free, but of course I accept donations 😄
XRP raa6cT47iv1KJxjLWwdQ7bJAyEPC9CfKFE
Eth
0xfD529b09974Ee51E7b6Fd2b5c7559a2D2c490178
LTC
LV3GKSomGYw2K7uqgUVjHg1YuauPmc1ztm
XMR
887YE2tQJqGYS5w8qdmQFt8frQGcWLvnbLfdB9nnjVfGLhuqfzEB3X7eSedRNHfdFV5Lc2LvkjGge4ougRRmaNrzT1dQANG

I also made a LoRA with a similar concept, and there was a similar problem.
Since the holes are close to each other, it has to learn their exact positions, which is quite difficult. In such cases, it seems that writing a very precise prompt increases the accuracy of the holes.

I also made a LoRA with a similar concept, and there was a similar problem.
Since the holes are close to each other, it has to learn their exact positions, which is quite difficult. In such cases, it seems that writing a very precise prompt increases the accuracy of the holes.

Thanks, can you share a prompt example ?