4 ms·

Show HN: ComfyUI extension for the new Stable Video Diffusion model

3 points by subtech 3y ago

Hey HN! Just shipped a new ComfyUI extension to add support for the new Stable Video Diffusion model! Now, you can plug this into any existing ComfyUI workflow to do some really cool things! Here are a few examples: Image to video workflow: https://comfyworkflows.com/workflows/5a4cd9fd-9685-4985-adb8-7be84e8636ad https://comfyworkflows.com/workflows/5a4cd9fd-9685-4985-adb8... Image to video workflow (w/ high FPS using RIFE frame interpolation): https://comfyworkflows.com/workflows/bf3b455d-ba13-4063-9ab7-ff1de0c9fa75 https://comfyworkflows.com/workflows/bf3b455d-ba13-4063-9ab7... Would love to hear any feedback! :)

Kevin09210 3y ago
Well, I can't load the workflow. Installing the missing nodes requires restarting comfyui and it seems persistence on comfyworkflow's cloud isn't guaranteed anymore. Also this website is pure single-page garbage. Everything is super slow, can't ctrl+click on a link to open it in another tab, etc ... An UX nightmare. Can you recommend a competitor ? Looking at: - https://www.thinkdiffusion.com/ https://www.thinkdiffusion.com/ - https://rundiffusion.com/comfyui-workflows https://rundiffusion.com/comfyui-workflows - https://flowt.ai/ https://flowt.ai/
- Kevin09210 3y ago
  I get similar results on thinkdiffusion
gus_massa 3y ago
In the first example https://comfyworkflows.com/workflows/5a4cd9fd-9685-4985-adb8-7be84e8636ad https://comfyworkflows.com/workflows/5a4cd9fd-9685-4985-adb8... Which are the initial images? How many? Only two or a few?
- subtech 3y ago
  The initial image is just 1 -- the very first frame of the video.
  - gus_massa 3y ago
    I don't understand how you tell the rocket is going up and the floor is not moving. Is that in the workflow [1] or the program firere that without help. [1] I don't understand what the worflow says. Sorry if my question is obvious.
tinytera 3y ago
This is great! I use ComfyUI extensively and will try this out. Its hard to setup workflow correctly for these Video diffusion models. I recently built a set of docker images to simplify GenAI workflows. Check it out at https://github.com/jaideep2/apinaio https://github.com/jaideep2/apinaio Only works on CUDA hardware for now, but contributions welcome :)

3 points by subtech 3y ago

Kevin09210 3y ago

Well, I can't load the workflow. Installing the missing nodes requires restarting comfyui and it seems persistence on comfyworkflow's cloud isn't guaranteed anymore. Also this website is pure single-page garbage. Everything is super slow, can't ctrl+click on a link to open it in another tab, etc ... An UX nightmare. Can you recommend a competitor ? Looking at: - https://www.thinkdiffusion.com/ https://www.thinkdiffusion.com/ - https://rundiffusion.com/comfyui-workflows https://rundiffusion.com/comfyui-workflows - https://flowt.ai/ https://flowt.ai/

Kevin09210 3y ago
I get similar results on thinkdiffusion

gus_massa 3y ago

In the first example https://comfyworkflows.com/workflows/5a4cd9fd-9685-4985-adb8-7be84e8636ad https://comfyworkflows.com/workflows/5a4cd9fd-9685-4985-adb8... Which are the initial images? How many? Only two or a few?

subtech 3y ago
The initial image is just 1 -- the very first frame of the video.
- gus_massa 3y ago
  I don't understand how you tell the rocket is going up and the floor is not moving. Is that in the workflow [1] or the program firere that without help. [1] I don't understand what the worflow says. Sorry if my question is obvious.

tinytera 3y ago

This is great! I use ComfyUI extensively and will try this out. Its hard to setup workflow correctly for these Video diffusion models. I recently built a set of docker images to simplify GenAI workflows. Check it out at https://github.com/jaideep2/apinaio https://github.com/jaideep2/apinaio Only works on CUDA hardware for now, but contributions welcome :)