dawnstream on 15/12/2024 at 22:53
I've been playing with this for the past 1-2 days and wanted to share it here.
I haven't figured out how to overcome the continuity issues in the longer scenes, so the environment changes every second.
I look forward to your comments :D
Attempt #01
[video=youtube_share;6qbZDaeICaI]https://youtu.be/6qbZDaeICaI[/video]
Attempt #02
[video=youtube_share;X4osP61GFug]https://youtu.be/X4osP61GFug[/video]
downwinder on 16/12/2024 at 00:35
yes a guard blocking a entrance is the most deadly fight you can have in thief
amazing lighting
Aemanyl on 16/12/2024 at 17:21
Garrett travels to India? Ehm, no, thanks.
demagogue on 16/12/2024 at 19:21
If you use a system that you can script like Comfy, then I know there are workflows people have released that maintain continuity across different images. Well to begin with you could do one long take so you don't have to use short clips to begin with. But you could cut in other scenes afterwards and keep continuity for, e.g., the faces and outfits and I'm sure the scenery if you set it up right. You'd have to dig through their database to find the right template workflow and read up on how to script everything though. It's not as easy as an online system where most of it is set up for you.
ZylonBane on 16/12/2024 at 19:54
Quote Posted by dawnstream
I've been playing with this for the past 1-2 days and wanted to share it here.
Not everything is worth sharing.
dawnstream on 16/12/2024 at 23:23
Quote Posted by demagogue
If you use a system that you can script like Comfy, then I know there are workflows people have released that maintain continuity across different images. Well to begin with you could do one long take so you don't have to use short clips to begin with. But you could cut in other scenes afterwards and keep continuity for, e.g., the faces and outfits and I'm sure the scenery if you set it up right. You'd have to dig through their database to find the right template workflow and read up on how to script everything though. It's not as easy as an online system where most of it is set up for you.
Hello. Yes, I am using the Comfy system and the CogVideoX model. Processing more than 60 frames at a time causes crashes, but there is a node called Meta Batch Manager as a solution. This node splits the video into chunks and applies the same process to all parts. However, even though I use a fixed seed, it produces different results in each iteration. I will do some more research to figure out how to resolve this issue. Thank you for your suggestions!
demagogue on 17/12/2024 at 00:19
That makes sense. I haven't played with video generation above a few seconds, but even with that I can see why you'd want to break it up.
The workflow I remember maintains continuity of characters, where you'll have your previous image as an input and it will force continuity on to the new generation. I was playing with making a comic where that's a really important thing. So I imagine a workflow like that for scenery could be made in a similar way, but it may not have been made yet, and I also imagine it would be quite a big task to make it oneself! Also the video component adds another layer of complexity to the problem.
kostyabud2009 on 17/12/2024 at 02:08
No offense, but this looks ugly. Probably because of the AI being in development
dawnstream on 19/12/2024 at 12:22
Attempt #02
[video=youtube_share;X4osP61GFug]https://youtu.be/X4osP61GFug[/video]
marbleman on 19/12/2024 at 21:22
I misread the title as Thief Ravaged by AI, and I think it's more appropriate.
