Duranovsky

Full time digital tinkerer, work includes The Lord of the Rings (Rings of Power), Love Death and Robots, The boys, Just Dance

[Duranovsky @the_duranovsky]

We are Emmy nominees!

[Duranovsky @the_duranovsky]

#ClaireAIContest
"Back to School: (Re)learn to struggle"
In this piece, I aimed to portray the struggles of learning a new art form, and be creative with it. It's a journey filled with overwhelming feelings, self-doubt, and those magical eureka moments. #comfyui #animatediff

[Duranovsky @the_duranovsky]

@clairesilver @Ai4Conferences For this I used a series of carefully crafted 3D simulations that would serve as a guide for Animatediff on ComfyUI to generate human shapes as much as possible, but failing most of the time.

[Duranovsky @the_duranovsky]

@clairesilver @Ai4Conferences #ClaireAIContest
"Back to School: (Re)learn to struggle"

In this piece, I aimed to portray the struggles of learning a new art form, and be creative with it. It's a journey filled with overwhelming feelings, self-doubt, and those magical eureka moments.

[Duranovsky @the_duranovsky]

@Mr_AllenT Great results! Have you managed to make it work on Comfy portable? I managed to correct most of the errors by manually installing (instead of using the manager) but some new errors popped out

[Duranovsky @the_duranovsky]

FAST FASHION
A new experiment combining #3d simulations (#tyflow) and #AnimateDiff on #Comfyui

[Duranovsky @the_duranovsky]

#tyflow + #Redshift+ #AnimateDiff + #Stablediffusion
Experiments mixing 3D softbody simulations and #AI postprocessing

[Pika @pika_labs]

Introducing Pika 1.0, the idea-to-video platform that brings your creativity to life.

Create and edit your videos with AI.

Rolling out to new users on web and discord, starting today. Sign up at https://t.co/JHRrinsIwx

[Duranovsky @the_duranovsky]

@dreamingtulpa I hope AI doesn't evolve into making 3D. 3D only exists because it solves a problem: creating something impossible using other means. AI is better than 3D because it will eventually let you bypass modeling altogether and get a realistic result without the constraints of 3D

[Duranovsky @the_duranovsky]

AI morphing, aka Prompt Travel
#ai #stablediffusion #AnimateDiff

[Duranovsky @the_duranovsky]

@amli_art I tried to use it in Auto as well but it was very cumbersome. Changing the model every time you want to refine gets tedious fast. I decided to try comfyui and you use a SDXL preset that loads each model for you, and now it's literally one-click solution. And it's 2X faster! 😱

[Duranovsky @the_duranovsky]

@clairesilver @StabilityAI If you haven't already, try ComfyUI to run SDXL locally, much faster generations, node-based UI and a very streamlined workflow, professional, reusable (and recoverable!) setups

[Duranovsky @the_duranovsky]

@EhteshamHayat @iamneubert I couldn't wait either. 3 books, 1500 pages, but worth it!

[Duranovsky @the_duranovsky]

@paultrillo @CoffeeVectors @DJIGlobal @dronedudes @hokutokonishi Great result! I was thinking an AI-powered alternative would be to create Nerfs for each pose and then track a clean shot and place each pose using Unreal luma ai plugin.

[Duranovsky @the_duranovsky]

@key_vfx Great result! Is it currently possible in Embergen to turn the smoke into a shape? I'm thinking an operator like a "find target" or attractor that makes particles go to it but with some space for nice dynamics

[Duranovsky @the_duranovsky]

@5agado However, the mesh created by icon has incorrect mesh topology for skinning. You would need to retopo them first, reproject the textures and then skin them. It does not look like smth that can happen without supervision

[Duranovsky @the_duranovsky]

@5agado There's a free tool named accurig that you can run locally and possibly batch-rig a bunch of characters and spitout fbxs that can be imported into unreal/blender. I'm pretty sure Houdini could do the whole process since I've seen it being connected to chatgpt directly

[Duranovsky @the_duranovsky]

@CoffeeVectors IMO, a good raw mocap will help you get a clean output in less time. Getting bad initial data will force to waste more time cleaning it up... and unless you are an experienced animator/tweaker, it will have a big impact in the final quality

[Duranovsky @the_duranovsky]

@CoffeeVectors The best solution I've seen is https://t.co/7LcUI3bfvI. Suitless, can mocap several subjects at once and the output is very accurate. Downside is you need at least 4 cams (they can be phones)