PLA.I
This is a brain dump of some of My AI experiments that I have conducted using ComfyUI!
Mainly uses Wan, Qwen, Flux and sometime Nano Banana Pro.
I also use houdini for 3d blockouts and simulations.
Car Stylization and Geo Creation
These cars are images found on google.
I used Flux USO with a low denoise I get an initial style transfer.
I used Qwen Image Edit for turning it into a convertible.
Converted Image to 3D model using Hunyuan 3d 2.0 directly in ComfyUI (Proxy Geo)
AI Polish to add life to blockout
Original
WAN VACE Keyframes
These are the 5 keyframes I manually batched into comfyui. the gemini logo is there because this was quick RnD so I used nano banana to edit images to make them feel more dynamic. This made the visualization of previs more elegant and was a lot more successful then I thought. I want to see if I can come up with a vace workflow that accepts keyframes along with open pose videos for more art directable motion.
Wan VACE rendering
Viewport to Render?
All videos are rendered on a 3070ti Laptop GPU.
Vellum Sim from Houdini, Image creation and editing with Qwen.
Batch Frame by Frame Style Transfer
Batch Frame by Frame style transfer using Flux Dev and 2D illustration loras with incremented seeds.
(randomized rotation and scale done in Nuke).
Really promising results
Video Editing with VACE
All videos are rendered on a 3070ti Laptop GPU.
This is mainly using SAM 2 and WAN vace control Masks to edit and target areas to change.
Nano Banana had better prompt adherence compared to Flux Kontext.
I am using heavily quantized models to fit my 8 gb of VRAM
Not perfect but learning the capabilities
