Created by @yanokusnir on Reddit, please support the original creator!
https://www.reddit.com/r/StableDiffusion/comments/1lu7nxx/wan_21_txt2img_is_amazing/
If this is your workflow, please contact us at team@floyo.ai to claim it!
Original post from the creator:
Hello. This may not be news to some of you, but Wan 2.1 can generate beautiful cinematic images.
I was wondering how Wan would work if I generated only one frame, so to use it as a txt2img model. I am honestly shocked by the results.
All the attached images were generated in fullHD (1920x1080px) and on my RTX 4080 graphics card (16GB VRAM) it took about 42s per image. I used the GGUF model Q5_K_S, but I also tried Q3_K_S and the quality was still great.
The only postprocessing I did was adding film grain. It adds the right vibe to the images and it wouldn't be as good without it.
Last thing: For the first 5 images I used sampler euler with beta scheluder - the images are beautiful with vibrant colors. For the last three I used ddim_uniform as the scheluder and as you can see they are different, but I like the look even though it is not as striking. :) Enjoy.
