[ 3 / biz / cgl / ck / diy / fa / ic / jp / lit / sci / vr / vt ] [ index / top / reports ] [ become a patron ] [ status ]
2023-11: Warosu is now out of extended maintenance.

/vt/ - Virtual Youtubers

Search:


View post   

>> No.61418330 [View]
File: 1.54 MB, 768x768, 0009--Pochiwanmaru.webm [View same] [iqdb] [saucenao] [google]
61418330

>>61389525
Good Morning anons!
>>61389975
>leaving it up to chance
Not really. You can play around with the Strength schedule values. Just take care as they seem to work in reverse compared to img2img. In img2img 0.9 gets you pretty much a new image, in deforum 0.9 is equal to 0.1.
Both methods of increasing strength or decreasing strength should be viable if you let them linger just a few frames after the transition.
It will allow the model to either interpret a new image from scratch or to take a chill moment and then reset to 0.6 or whatever strength you use.
The annoying part is I haven't figured out how to make something not be a linear transition towards the next value, so I'm manually controlling it by using the previous step as a stop gap 0: 0.6, 19: 0.6, 20: 0.9 (movement happens in this transition for 5 frames) 25: 0.9, 28: 0.6.
This means during the transition, only 4 steps are done on the image, then resetting.
Here is a quick test. The next method improvement would be to add image guidance to those steps right after the transition stops and then you have nothing left to chance anymore.
>>61395113
This second painter Ina is even better than the first.
>>61409596
I love how good the painting style came through
>>61417275
Amazing vid2vid coherence! It doesn't even need masking for the surroundings as they stay coherent anyway.

Navigation
View posts[+24][+48][+96]