No.3916
Give me a pic and I'll do an animation generation thingie with local "WAN Video". You need to include a "natural language" description of what will happen. There is an AI to autotag the general description of the static image.
For instance this was what I wrote for the OP video:
Himari Burg
Strong, smooth animation. Cartoon anime animation. The girl looks around. She blinks her eyes. She lowers the blanket, revealing a hamburger. She holds the hamburger to her mouth and takes a bite. She then covers herself with the blanket and hides her face.
This was the autotag for the image:
AI Tag
Anime-style drawing of a cute, young girl with light pink hair and large, expressive purple eyes. She is wearing a white hooded cloak with a hood, and is sitting on a red couch. The background is a simple, dark brown gradient. The girl's expression is neutral, and she is looking directly at the viewer. The image has a soft, pastel color palette. The style is clean and detailed, with a focus on the character's delicate features and soft shading.
I can do NSFW too since it's a local model, but that should be on the appropriate board. *cough*
I'm trying to figure out the painful installation of this 'sage attention' thing that is supposed to half generation time, but until then I'm going to limit the size and duration of things. It took me 3 minutes to generate this, which is definitely not right. Not sure how I was able to do 50 second generations a couple days ago...
No.3918
Is this the current standard for local video generation, or is there anything somewhat close to this recent "Veo 3" thingy or what it was called? I heard about that one just a day ago and it looks very impressive, but seems to be gated by G**gle.
No.3919
>>3917That's not really describing anything. She's already offering a cake.
>>3918I'm not exactly an expert at it since I just now got hardware able to do it. You can get it to like 12 seconds or something, but the processing is exponential the longer you go. The online models are certainly far better.
The upside is that you could do porn with it without working at google or otherwise having a privileged position. Presumably a higher level of control, but this ComfyUI program is god awful and I hate looking at it. I don't think I'll be able to tolerate this video stuff for very long due to how horrendous the UI experience is.
It seems like unfortunately the real life stuff is far better, so good news for people that want to do deepfakes or cause strife in the real world.
No.3920
>>3916This burger just bit this shab.
No.3922
>>3921I don't think that's possible. Well, it's certainly not with my knowledge at least. For best results I need to start with an image with most of the detail already there and then this AI video can move it around in vaguely understandable ways.
There's a text-to-video option, but I haven't tried it yet and there's no way it will compare to actual 2D models for static images.
No.3923
>>3919What hardware are you using in particular, if I may ask?
No.3924
>>3923You can ask! I did some major upgrades this year due to the economic chaos caused by retards making bad decisions.
The 5090 that I got just last weekend is doing the bulk of the work, but some of it is being offloaded to DD5 memory. I'm trying to get this thing called "sage attention" to work, but it's been very difficult due to software upgrading at different times and I quite seem to get everything synced up correctly.
"SCRIPT FAILED! You're using Version 1.294.12941 of Script Thing. You need 1.294.12939!"
WARNING! Script Thing 1.294.12939 is incompatible with Decompiler Thing 2.9291B!"
It's really a marvel at any of this AI stuff works at all due to all the dependencies and separate scripts and everything. This pic is from cleaning up a few installations when troubleshooting.
No.3936
>>3922>I don't think that's possible. Well, it's certainly not with my knowledge at least.Ok, then something simpler with that pic please.
When you're generating videos, is the input pic necessarily the first frame of the video? I recall the Sora demos of multiple videos that held the final frame constant, so different sequences of events would play out and converge to the same final scene in a way that felt very uncanny due to how improbable such a thing would be. Can you do something like that for the skibidi toilet girl?
No.3943
>>3936I'm really new to this. All I know how to do is take an image and animate it... somewhat. You're asking me to do the things that billion-dollar data centers are doing and I'm not sure that I can. Local models have some pretty heavy limitations even if there's some freedom involved with the lack of censorship.
Well, here's a vid. I just said "head bopping up and down with extremely long neck" and as I type this it's generating and I'm not sure if it will work.
In the future I don't want to do real life images like this, but you can see that it's pretty impressive technologically. But isn't as fun or cute as 2D stuff.
No.3944
>>3943also this model is supposed to be focused on more fantasy/2D stuff. I didn't download the real life model that would probably be better at real life images.
No.3945
>>3943omg I really like that result, thank you
No.3949
>>3948she looks like she is aggressively masturbating....
No.3951
how strange
No.3953
>>3950KORURI WATCH FOR THE ASTRAL DEMONS WHEN PROJECTING OR THEY'LL CATCH YOU
No.3956
It keeps trying to smoothly interpolate and fade stuff in and out. Maybe it says it's a video model intended for 2D animation but I don't buy it, I reckon even video models finetuned for 2D will still do 3D better.
No.3960
>>3943Wtf I love skibidi toilet now
No.3961
>>3959this guy draws really cute lolis