3D modeling/printing artist, AI enthusiast, rookie Python coder.
Finland
Joined October 2017
- Tweets1.8K
- Following79
- Followers6.8K
- Likes1.9K
WanVideo sliding context window test on the 480p 14B T2V -model, worked surprisingly well but incredibly slow (50 mins on a 5090, 513 frames at 832x480)
As an impatient person, I often get obsessed with optimizations, tried implementing TeaCache for WanVideo, failed, but in the process accidentally (maybe) succeeded? Results look promising at least!
WanVideo has been lot of fun, currently playing with applying sliding context windowing similar to AnimateDiff, the 1.3B model is especially suitable for this due to it's speed, here's 1025 frames in one go, under 5GB VRAM used, but took about 10 mins on a 5090 with 30 steps.
HunhuyanVideo -model's vid2vid passes the hippo test! Very promising and versatile model, thanks to the cfg distillation and every possible optimization I could think of this clip of 101 frames at 768x432 took about 2 minutes to sample, fitting slightly under 20GB using my nodes.
I have finally pushed a bigger update to my CogVideoX ComfyUI wrapper nodes, cleaning up most of the bloat that has been accumulating with all these different models. One of the discoveries I made during this is that the orbit -LoRAs work with the "Fun" -models as well!
CogVideoX 1.5 image2video with DimensionX orbit LoRA can really be something!
#ai #GenerativeAI #cogvideox
As bit of a personal challenge I decided to get the new Mochi (github.com/genmoai/models) -text2video model running locally on 4090, this was generated under 20GB VRAM in ComfyUI.
It's not perfect as I had to use tiling to be able to decode the whole thing,
github.com/kijai/ComfyUI-Moc…
Oh hey Tora trajectory actually works with the image2video model too, this will be good...
#ai #comfyui #cogvideox
This model is so much fun!
I have implemented Tora:
github.com/alibaba/Tora
to my CogVideoX nodes:
github.com/kijai/ComfyUI-Cog…
This implementation can be run within ~13GB VRAM
#ai #comfyui
I've been deep into CogVideoX -code lately. Using the CogVideoX-Fun "pose" model allows control with variety of inputs, and applying Kosinkadink's context windowing from AnimateDiff (thanks to ) I've managed some decently stable longer generations.
#ai #cogvideox
Segment-anything-2 feels really good, inspired me to spend whole day building nodes for it, still WIP, but already works for example with Florence2 detection!
SAM2 + Florence2 combo in ComfyUI
github.com/kijai/ComfyUI-seg…
github.com/kijai/ComfyUI-Flo…
Managed to get LivePortait running near realtime within ComfyUI, now also without Insightface, using Google's MediaPipe (Apache-2.0 license) instead!
github.com/kijai/ComfyUI-Liv…
Now that @printablescom has a store, I'll be adding my paid models there too (hopefully new ones soon as I find the time). First up in honor of the year of the dragon: printables.com/model/713708-…
#3dprinting #3dmodel #dragon
Example of what AI generation can do in practice currently, from initial image (which is also generated) to a very detailed printable model, the level of detail mostly made possible with the Marigold depth estimation.
#ai #3dprinting #dragon #technology
Another #AI assisted render, this time using my "Azure Cloud Wyrm". Everything about the model itself made in #blender3d, rendered normal+depth pass and used #ComfyUI to run it through #animatediff with corresponding ControlNets.
#dragon #stablediffusion
I've been learning some #AI assisted animation lately, this is made from my #Ranni sculpt I created in #Blender, with simple matcap normal + depth render pass fed to #stablediffusion #AnimateDiff
Watching @OnlyMurdersHulu suddenly saw something familiar, while I know this is only few seconds of screen time, that's Meryl Streep and Martin Short next to my original design on the shelf: "Lunael's Dream", they even lighted it up!
#3Dprinting #OnlyMurdersInTheBuilding
