Using this lora with multishot prompts forces the model to alucinate from one shot to another.
Not sure why, but if you use this lora with a minimax h3 prompt that setups multiple shots with cut's between then, the model tries to solve the movement of one shot to the next, without cutting.
It's a shame, because it does make everything looks much more real... but it can't be used with a director node, for example. It only works with simple one shot videos.
Doing the opposite for me here - I want a continuous shot but sometimes it adds a cut very early (depending on the prompt I guess - but without the LoRA the cut is NOT there, with the same prompt which specifically asks for no cuts, fixed camera etc).
Also for higher resolutions, it sometimes ruins the generations - adding a tone change/yellow tint and/or blurriness (close-up shots).
Very strange how it affects the prompt following though - I wish it only enhanced detail/realism instead of breaking other stuff
Using it with a turbo lora. I'm having trouble with it as well. It seems to stop following prompt instructions correctly, especially during shot changes, framing shifts, or character actions. When I disable it, I get much better adherence to the prompt. When it's on, it's a hit or miss. In some situations it wins, but so far most of the time it's a troublemaker. I will try with 0.5/0.6 strength to see if this improves.