calcuis

/

wan-gguf

like 40

hey, seems you are using the wrong model (according to your selection in Load Diffusion Model); t2v doesn't work for i2v; it can run but not the outcome you expected

Thanks for the response, i had tried both text to video and image to video.
Here is an updated image showing the settings used for text to video, with the same strange distorted circular output:

(If anyone could clue me in i would be very appreciative!)

select t5xxl_um_fp16.safetensors if you use GGUF CLIP Loader; opt to use native Load CLIP instead if you select t5xxl_um_fp8_e4m3fn_scaled.safetensors

Thanks for the response, I did try that as well but i get the same strange flickering circular output.

I also tried various samplers and schedulers. Comfy is up to date. I am using rtx 2060 6gb with 64 gb sys ram.

Terminal shows no issues:
got prompt
Requested to load WAN21
loaded completely 3353.64662437439 2706.1788330078125 True
100%|███████████████████████████████████████████████████████████████████████████████| 20/20 [00:14<00:00, 1.42it/s]
Prompt executed in 19.07 seconds```

If anyone has run into this issue and might know the solution, please let me know. Thanks!

You can see the issue and the setup below:

Still having the above issue...

:(

HI
This worked for me after a lot of struggle:
umt5_xxl_fp8_e4m3fn_scaled for clip
wan_2.1_vae for vae.
even gguf model or the same model as you use.
A collection from civit kijaj and calcuis :-)

Thanks for the response, I did try that as well but i get the same strange flickering circular output.

If anyone has run into this issue and might know the solution, please let me know. Thanks!

You can see the issue and the setup below:

not sure why you are trying to connect the ModelSamplingSD3 first before KSampler; it was designed for sd3 model probably won't work for others

Still having the above issue...

:(

just drag the sample video or the json workflow to run if you are not sure how to set it up at the very beginning