Welcome to Inkbunny...
Allowed ratings
To view member-only content, create an account. ( Hide )
LITTLEFisky

AI animations

Hey AI directors! Can anybody point me to a guide to ai-animations? I'm interested in offline generation. I have Forge and ComfyUI.

My PC specs: Ryzen 4600G, 32GiB RAM, RTX3060ti 8GiB

Also note: I really like these models and would like my animations in their styles:
https://civitai.com/models/1162518?modelVersionId=1714002 Plant Milk 🌿 - Model Suite
https://civitai.com/models/1143097/willys-noob-realism?... Willy's Noob Realism
https://civitai.com/models/784543/nova-animal-xl?modelV... Nova Animal XL

Any help is appreciated <3
Viewed: 203 times
Added: 1 month, 3 weeks ago
 
DragonSoul2875
1 month, 3 weeks ago
There are a few ways with comfyui.

Mochi (I heard it was good): https://stable-diffusion-art.com/mochi-comfyui/#Use_Moc...
WAN 2.2 (I used personally with a couple animations I made that I have some uploaded here): https://docs.comfy.org/tutorials/video/wan/wan2_2
Arkunus
1 month, 3 weeks ago
I hope you succeed
thecooler
1 month, 3 weeks ago
Please don't disappoint me. its ok you want to know how it "works" or learn about it, but I seriously do hope you do not use AI nor post that at all here or any where else. I do respect you as an actual animator. plus the bubble is been multiple times predicted to burst eventually.
LITTLEFisky
1 month, 3 weeks ago
I just want new animated wallpapers :D
thecooler
1 month, 3 weeks ago
Oooohhh! for fuck sake you scared me TTwTT
PipTheOtter
1 month, 3 weeks ago
Unless you plan to stich a bunch together, it likely wouldn't look that good. Most AI generated videos (local) are 720p at best, and anything over that is upscaled past support of the models or using non video based  upscalers. So if you're on a 1080P monitor, it may work ok..... but if you're on anything over that it'll look like a blurry mess at full screen :(
arfinator2
1 month, 3 weeks ago
The wan 2.2 image 2 video and first/last frame 2 video workflows are pretty good, but you're going to struggle with 8gigabytes of vram I think. Comfyui can run them with it, but you'll be down at 640x640 resolution and pretty short animations and long iteration times. Possibly workable but it'll be frustrating.

I think most of the decent animations are doing something along those lines. Work in automatic1111 or a comfyui image in-painting workflow until you get a decent starting image, then use image2video to get new frames or finished video if you're lucky. That's what I have been doing anyways, but with an rtx4090 and a lot of free time. I'm also a Blender artist so posing characters then using that to do image2image or in-painting to get characters in the poses I want is helpful.

The AI hate is understandable, but to me it's just another tool. AI is a terrible painter, but one hell of a paint brush.
DownThePipes
1 month, 3 weeks ago
going say don't mess with it all
ShamanSquirrel
1 month, 3 weeks ago
I only use online AI animators. For porno making, I use Wan 2.5, which you can find on http://venice.ai. For clean stuff, I like to use whisk: https://labs.google/fx/tools/whisk
Lunafan1k
1 month, 3 weeks ago
Your best bet for making a wallpaper would be the wan 2.2 image to video using start and end frames. In comfy theres a model template you can just install and go from there. Use a simple LORA if you need but its only 5 seconds so just keep the animation simple and itll turn out fine.

Wan 2.2 text to vid is a bit harder and youll need a ton of LORA and specific checkpoints to get what you want as the base model is for humans

Looking at the links you shared, these are for Illustrious and NoobAI models. Illustrious requires a subscription or at least money and i personally wont support that. Running locally you want stable diffusion for images and wan for video, i use SD 3.5 its fast and compatible with a lot of checkpoints, like Pony and other stable diffusion models.

I downloaded the Willy's Noob Realism checkpoint to test it in comfy, and looks like it works great in SD 3.5! the base model itself its pretty well tuned and you can definitely use various LORAs for greater fine tuning.

I run a 4090 and it takes about 5 seconds to make an image size of 960x540, then i upscale it using a series of upscale nodes to 4k.
KodiCoyoteOfficial
1 month, 3 weeks ago
Framepack is a good place to start. While Wan is very accurate, it’s very computationally expensive and may not work with your set up with only having 8GB of VRAM. Framepack is simple: just tell it what you want the subject(s) to do and how long you want it to be and it’ll render away.


https://github.com/lllyasviel/FramePack/releases/tag/wi...
LITTLEFisky
1 month, 1 week ago
Causing "MEMORY_MANAGEMENT" BSOD or just crashing with OOM
KodiCoyoteOfficial
1 month, 1 week ago
How much system memory do you have? Are you running it overclocked or an XMP profile? Framepack uses a lot of system RAM and that error usually indicates a problem somewhere with RAM.


Edit: okay wait I see you posted you have 32GB of VRAM. Windows should automatically manage your memory. What resolution are you trying to render at? I always start at 1024x1024.


You could also try bumping up your VRAM voltages up a little bit.
LITTLEFisky
1 month, 1 week ago
32GB of RAM. Not VRAM. I have only 8 of it. And FramePack says it's only 4.8GB available
KodiCoyoteOfficial
1 month, 1 week ago
That’s strange. Is there anything taking up resources? It should automatically allocate system RAM. And you’re using Windows?
LITTLEFisky
1 month, 1 week ago
Nope. Nothing was running except this tool. It's Win10 x64
KodiCoyoteOfficial
1 month, 1 week ago
VRAM usage should be basically zero unless there’s something else loading itself into memory. And memory management is a system RAM error. Are you using an XMP profile?
LITTLEFisky
1 month, 1 week ago
Nope. No XMPs, overclocking or anything like that
KodiCoyoteOfficial
1 month, 1 week ago
Try running the default XMP profile #1
New Comment:
Move reply box to top
Log in or create an account to comment.