This is the workflow I use for video generation (I2V)
image unrelated
Description
FAQ
Comments (13)
tried to download but windows says the zip is empty when i go to extract it. looking forward to checking out your process
weird, downloaded it and it works for me. try opening it with winrar
@eldingu actually just tried it normally again and it worked for me. the file size displayed was also different, so I assume it was something not loading correctly on civit's end. anyway thanks!
Absolutely amazing flow.
It required to install a thing or two needed by some features to work, but it went well and works perfectly.
Thanks for the share!
you're welcome :)
thanks so much for sharing, your vids are amazing
could you send an example of one of your prompts?
Thans so much for sharing, it's absolutely amazing!
Could you please tell me what kind of image you send to the workflow? I can't generate a good image for my creation. Or maybe you can share the workflow about your image? Thans!!!
glad you like the workflow :)
this is what i use for image generation: https://civitai.com/models/1386234?modelVersionId=2709612
i also recommend this model for 3d cg like image generation: https://civitai.com/models/715287?modelVersionId=2744564
use the embeddings from here for a better result for illustrious based models: https://civitai.com/models/1062439?modelVersionId=1192334
@eldingu
I have a workstation with 8 A5000. Maybe It's great to transfer the gguf into unet? And what do you think can help me to improve the video? And I'm studying how to use ComfyUI-MultiGPU now. Thanks for your reply!
@bintaoyu29100 simply raising the resolution and steps should raise the quality of the output in the workflow. beyond that, i'm afraid my knowledge is limited
Great workflow, thank you for sharing!
But recently, I’ve been getting only black screens/NaNs when generating. I’ve already tried forcing bfloat16, using the bf16-vae flag, and bypassing SageAttention/LoRAs, but the math still seems to blow up on the newer ComfyUI backend.
Have you encountered this recently, and do you have a plan to update the node logic or a 'V2' of the workflow that handles the current version's precision better? Thanks again!