This is a low vram friendly workflow.
GGUF models working with 4060 8G card could be OK.
Have a try my friends~
Description
FAQ
Comments (9)
thanks for this! works fine with my 4060 8G - but it takes ~25 minutes for an upscaled 49 frame video, compared to ~6 minutes using Hunyuan fast. Is there any way to speed it up?
It might be faster using this flow, but its quality seems not good enough.
https://civitai.com/models/1309674/simple-wan-21-low-vram-comfy-ui-workflow-gguf-4gb-vram-16gb-ram
I just did a little optimization based on this.
BTW, 6 min using Hunyuan fast on 4060? Is that real? Could u please share ur workflow and thank u so much~~~
@Diaoshini8280128 I am using "HunBasic 1.5 (T2V + Upscale)" from here: https://civitai.com/models/1007385/hunyuan-allinone with hunyuan-video-t2v-720p-Q5_0.gguf from here: https://huggingface.co/city96/HunyuanVideo-gguf/tree/main
In the workflow I activate "Fast Lora" using this Lora https://civitai.com/models/1063735?modelVersionId=1193807 - set steps to 10, FastLora weight to 1.0, width/height to 480x320 and the frame count to 81 in the workflow's setting console.
The upscaler is set to 1.5 and I end up with 720x480px videos that look really nice (most of the time).
@yesitsfine That‘s great~~ I gotta try. Thanks again!
⚡ I didn't understand how to set the DENOISE, what to set to be closer to the original image ??? ⚡
with "NVIDIA GeForce RTX 2060 - VRAM 6MB, RAM 16MB".
40mn for 418x720 good quality, on a waterfall and a house ...
first WAN tests on my PC ... I'm very satisfied even for the 40mn time .quality and time it took on servers a few months ago 👍👍
I could feel the exciting moment when the first high-quality video generated with a low-rank GPU card after seeing lots of "4090 mates" creation, even it took time.
I'm sorry but could not anwser your question about the denoise parameter. Perhaps asking GPT4 could help. ^_^
1st: PC precision ""NVIDIA GeForce RTX 2060 - VRAM 6GB, RAM 16GB"", I don't have a ZX81 😧.
good idea to give the .json workflow to an LLLM a, I use QWEN , it understands a workflow surprisingly well
How to extend the duration of the film so that the characters have more actions?
Perhaps more sampling steps and more length in ’WanlmageToVideo‘ node could help get more actions.
My GPU card is not that good and could not do more test due to OOM error.
BTW, some VFI node could help extend the length of vedio but slower actions.