Is v10 WAN 2.1 or WAN 2.2?
That's all. Thanks for your time and reply.
The T2V is based completely off of the WAN 2.2 "low" model. I2V is mostly WAN 2.2, but some of 2.1 had to be mixed in due to how WAN 2.2 I2V works. "Mega" is based solely on WAN 2.2 "low" for all use cases.
The T2V is based completely off of the WAN 2.2 "low" model. I2V is mostly WAN 2.2, but some of 2.1 had to be mixed in due to how WAN 2.2 I2V works. "Mega" is based solely on WAN 2.2 "low" for all use cases.
How do you implement accelerators using only the low part also does using 2.2 in low only works?? i thought you needed high for the movement but can you skip it?
The T2V is based completely off of the WAN 2.2 "low" model. I2V is mostly WAN 2.2, but some of 2.1 had to be mixed in due to how WAN 2.2 I2V works. "Mega" is based solely on WAN 2.2 "low" for all use cases.
How do you implement accelerators using only the low part also does using 2.2 in low only works?? i thought you needed high for the movement but can you skip it?
The "low" model can do movement without the "high" model. However, the "low" model movement likes to have guidance on bigger motion which is typically what the "high" model provides. This is why using both models does give you bigger motion (at the expense of some speed and complexity). However, you can provide that guidance by using "first and last" frames instead, which can not only give you significant motion, but also predictable results. Using a "control video" can provide that bigger motion too, granted you are using the "mega" model. So, lots of options, tradeoffs and yes, it can be skipped.