https://www.linkedin.com/in/abdallah-issac/
best image to video model
***update your Comfyui to latest version the commit for this model was yesterday !
*** (VHS)use this to see your video while process (it good to cancel unwanted result and save your time)
https://civarchive.com/images/59092666
The best type of fp8 where you can use torch compile instead of e4m3 nor gguf also work with 30x0 cards ,return to this issue
(be careful torch compile work with torch 2.6 , i fixed here )
how to use
the workflow https://civarchive.com/models/1276735?modelVersionId=1440307 (any advice welcome)
always start your prompt with FPS-24,
try to choose image carfully and guess what will be happend with your words
(dont use gpts)
as you choose an image there are many must be ignored like style the scene ,,,
try less prompt which not far from image .
to use loras enable this as shown in image
all other types u can found here ,,, https://huggingface.co/Kijai/SkyReels-V1-Hunyuan_comfy/
,, any request of any type of gguf quantization will be welcome
Description
FAQ
Comments (15)
Thank you. I was trying another workflow earlier with skyreels but it was taking over 10 minutes with my 4090 with no easy way to change resolution. This one is working better for me. question. is hunyuan loras not compatible with this model? or can i use loras?
i'll add this in desc
i add to the description how to activate loras ,enable "patch model patcher order"node
Thanks for this, Abdallah. I'm a ComfyUI newbie, so this is a little confusing to me. Once I enable the node you mentioned, where does the lora go?
@pintong704 It's a good question. i am not sure myself, i tried adding lora before and after the node to test, but i get pretty poor results, i haven't had enough time though to test it further or i may be doing something wrong. I get a feeling though that this image2video method just doesn't play nice with lora's. i sorta see lora working but the quality of video diminishes greatly and coherence of the result. It could be i need to use much lower strength on lora, or i need to put it somewhere else in workflow, i don't know. hopefully Abdallah has insight. Otherwise although this is best image2video i seen so far, i am still hoping official support is out soon and has more flexibility
if u think again image here work as lora , so if image in contrast with lora u will have bad result , also all of this still under expermient plus this model has little difference than original hunyuan !
what is the inference time you got ?
I am still getting 7 minutes for 960 x 544 , 97 frames. Is that expected
The speedup is amazing, but unfortunately, i'm just getting a lot of noise in the results, just looks like a colorful mess.
Could you post a picture of your settings in the workflow? I checked everything and believe I have everything right, but I must be missing something.
can u show me urs , upload screenshot and link it here plz
what is the inference time you got ?
I am still getting 7 minutes for 960 x 544 , 97 frames. Is that expected
@yashlanjewar20 its depend on your gpu , mention ur gpu type to compare with who have same gpu
a single 4090.
Also can you mention your gpu with inference time and specs
plz add ur video , number of frames , time , ur gpu! to help other members
also work in swarm-ui ;)
Unfortunately, this is not for my hardware. 640x480x41 frames - 20 minutes . 4070 12 Gb
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.