This is smallest workflow I could run on my RTX 3060 under Ubuntu 24.04.1 and 32gb RAM. Let me know if that worked for you.
Description
minimal workflow to create a video.
FAQ
Comments (7)
Wanted to test it out and it took 16mins to generate on a 3090 (24gb vram) and with 64gb ram. 2mins to upscale. (I should of specified realistic, rip my cartoon output).
VAE Decode (Tiled) into a Upscale Image (using Model) was an awesome idea!
Unfortunately, I'll continue using this workflow as it optimized many of HunYuan issues https://civitai.com/models/1092466?modelVersionId=1294744
I can't take credit for the tiled approach. I just took existing workflows and reduced them to the minimum. I'll have a look at the link you share though.
It definitely works on the 3060ti! But is there any way to add lora to the workflow?
Connect a "Hunyuan Video Lora Loader" right after the model loader, so the lora loader is between it and the connector piece.
You are really my god, this workflow perfectly matches my computer configuration, I am a novice, just do not know how to reduce the resolution to improve the speed of the graph, thank you
Any idea how to get this working for img2video instead of txt2video?
easiest way, use a node to convert the image to latent and then feed that latent to the sampler