Testing of the latest commit: 4781156bc87599dc07c926034a6a3cb89b3e2bae

[zadokov]

Thanks for the dynamic Lora loading! It Works!!!
You can select which Loras to load + the weight for each one. Perfect!
Currently my main problem is having the resulting clip adhering to the prompt.
Tried with timed prompts or a single prompt but it appears that usually the prompt is not really being followed.
For example - having a person with his head and feet out of the frame and asking to zoom out or pull out (etc.) is not causing the general frame to change. Prompting the move of hands (that are in the frame) for example, works OK.
I think it could be that the end frame is by far different than the starting frame so the move is problematic.
One way to solve would be to add end-frame support (several PRs support it - like #167) although the whole idea for me is to have the model create the end frame...
Any ideas how to improve this and have a very dynamic and changing clip?

[colinurbs]

In general a lack of camera control seems to be a Framepack problem. I'm working on bugfixes and utility stuff this week but plan to experiment with generations more next week. Please let me know if you learn any techniques that work. Don't be afraid of bumping up the distilled CFG a little bit too. You can push it fairly high before things start getting weird.