New Skyreels I2V workflow working better

[slmonker]

much better then the 1st version

HunyuanVideo_skyreel_I2V_00068.mp4

[slmonker]

HunyuanVideo_skyreel_I2V_00071.mp4

[RhapsodyHayden]

I'm trying to use this but I keep getting this error.

Prompt outputs failed validation
HyVideoSampler:

• Return type mismatch between linked nodes: stg_args, received_type(LATENT) mismatch input_type(STGARGS)
• Return type mismatch between linked nodes: teacache_args, received_type(FETAARGS) mismatch input_type(TEACACHEARGS)

[slmonker]

I'm trying to use this but I keep getting this error.

Prompt outputs failed validation HyVideoSampler:

• Return type mismatch between linked nodes: stg_args, received_type(LATENT) mismatch input_type(STGARGS)
• Return type mismatch between linked nodes: teacache_args, received_type(FETAARGS) mismatch input_type(TEACACHEARGS)

Have u upadate the nodes and new workflow? I dun think the STG node connected in the workflow..............

[RhapsodyHayden]

I'll try updating, hopefully I don't lose ReActor lmao.

[slmonker]

I'll try updating, hopefully I don't lose ReActor lmao.

the new I2V workflow pretty good，i think it's really close to Kling 1.0 or even better.....even the Vram using lower to 12gb.

[tanh609]

@slmonker are you using this workflow?

https://github.com/kijai/ComfyUI-HunyuanVideoWrapper/blob/main/example_workflows/hyvideo_skyreel_img2vid_example_01.json

I kept getting out of vram issue, despite of having 24 GB available.

[slmonker]

@slmonker are you using this workflow?

https://github.com/kijai/ComfyUI-HunyuanVideoWrapper/blob/main/example_workflows/hyvideo_skyreel_img2vid_example_01.json

I kept getting out of vram issue, despite of having 24 GB available.

yep，i m using this ,and u can turn on the Auto cpu offload, it took me abt 12 -14g vram to run it

[RhapsodyHayden]

Oh. I'll give it a shot and adjust what I need to but I'm using 12GB VRAM. Looks like the Text Encoder is eating ALL my VRAM when it loads to it.

[slmonker]

Oh. I'll give it a shot and adjust what I need to but I'm using 12GB VRAM. Looks like the Text Encoder is eating ALL my VRAM when it loads to it.

then u may turn Auto Cpu offload on,it s working

[tanh609]

Wow @slmonker thank you for the quick response.

How do I turn on "Auto cpu offload"?

Oh wait I found it. I'm trying it now.

[slmonker]

Wow @slmonker thank you for the quick response.

How do I turn on "Auto cpu offload"?

Oh wait I found it. I'm trying it now.

here bro

[ronbere]

DownloadAndLoadHyVideoTextEncoder

Failed to import transformers.models.timm_wrapper.configuration_timm_wrapper because of the following error (look up to see its traceback):
cannot import name 'ImageNetInfo' from 'timm.data' (D:\ComfyUI_windows_portable\venv\Lib\site-packages\timm\data_init_.py)

[slmonker]

DownloadAndLoadHyVideoTextEncoder

Failed to import transformers.models.timm_wrapper.configuration_timm_wrapper because of the following error (look up to see its traceback): cannot import name 'ImageNetInfo' from 'timm.data' (D:\ComfyUI_windows_portable\venv\Lib\site-packages\timm\data__init__.py)

feels like smthing wrong about ur environment

[gxground]

可灵最好用 但太贵了 开源的和可灵比感觉还是有点差距

[tanh609]

Hello @slmonker,

How did you get your video to match with the original image? My video does not match with my original image.

I put this on the positive prompt

The cat extends its front paw upward, claws hooking onto the surface as its hind legs push off. Its body stretches and contracts fluidly, muscles tensing with each pull. The tail sways for balance, adjusting with every shift in weight. Its back legs find quick, sure footing, propelling it higher in a smooth, determined rhythm. With each movement, it climbs steadily, pausing only briefly before reaching for the next hold.

Original Image

Result

failed_render.mp4

[slmonker]

Hello @slmonker,

How did you get your video to match with the original image? My video does not match with my original image.

I put this on the positive prompt

The cat extends its front paw upward, claws hooking onto the surface as its hind legs push off. Its body stretches and contracts fluidly, muscles tensing with each pull. The tail sways for balance, adjusting with every shift in weight. Its back legs find quick, sure footing, propelling it higher in a smooth, determined rhythm. With each movement, it climbs steadily, pausing only briefly before reaching for the next hold.

Original Image

Result

failed_render.mp4

Have u upadted the nodes and workflow? or u put FPS-24 infront of ur prompt？

[DCVirtualCosmos]

@slmonker are you using this workflow?
https://github.com/kijai/ComfyUI-HunyuanVideoWrapper/blob/main/example_workflows/hyvideo_skyreel_img2vid_example_01.json
I kept getting out of vram issue, despite of having 24 GB available.

yep，i m using this ,and u can turn on the Auto cpu offload, it took me abt 12 -14g vram to run it

I'm trying to use that workflow but I keep getting errors related to the model's dimension. "Given groups=1, weight of size [3072, 32, 1, 2, 2], expected input[2, 16, 25, 64, 64] to have 32 channels, but got 16 channels instead" I got this on the HunyuanVideo Sampler Node.
I have updated the hunyuan wrapped and the kijai nodes, which I had working just fine with the original hunyuan model. But for some reason, it is not working with the SkyReels

[DCVirtualCosmos]

lel I did the most basic workflow with this Skyreels models (FP8 and BF16) using only comfyui native nodes (Load Diffusion, KSampler, EmptyHunyuanLatentVideo, and DualClipLoader), I still got the same error with both SkyReels models ("Given groups=1, weight of size [3072, 32, 1, 2, 2], expected input[1, 16, 24, 64, 64] to have 32 channels, but got 16 channels instead") but works fine with the normal HunyuanVideo model

[slmonker]

Hello @slmonker,

How did you get your video to match with the original image? My video does not match with my original image.

I put this on the positive prompt

The cat extends its front paw upward, claws hooking onto the surface as its hind legs push off. Its body stretches and contracts fluidly, muscles tensing with each pull. The tail sways for balance, adjusting with every shift in weight. Its back legs find quick, sure footing, propelling it higher in a smooth, determined rhythm. With each movement, it climbs steadily, pausing only briefly before reaching for the next hold.

Original Image

Result

failed_render.mp4

i tired on my pc and the prompt is“FPS-24, a cat climbs on a steep mountain”https://github.com/user-attachments/assets/5408bdb1-bc57-42a3-b35f-18b8011a336f

[niftyflora]

is it supposed to take 50 minutes to generate a 2 second video?

[slmonker]

is it supposed to take 50 minutes to generate a 2 second video?

base on ur gpu level

[pftq]

Hello @slmonker,

How did you get your video to match with the original image? My video does not match with my original image.

This happens to me sometimes if the steps are too low - the default settings for the templates (CFG 6 and 30 steps) are a bit too low in general imo. I usually have to push steps to 50-100 for something decent on CFG 6 or lower the CFG to 3 (at cost of prompt adherence).