馃搨Files :
Recommendation :
>24 gb Vram: base or Q8_0
16 gb Vram: Q5_K_S
<12 gb Vram: Q4_K_S
For base version
VACE Model: wan2.1_vace_14B_fp8_e4m3fn.safetensors or wan2.1_vace_1.3B_fp16.safetensors
In models/diffusion_models
CLIP: umt5_xxl_fp8_e4m3fn_scaled.safetensors
in models/clip
For GGUF version
VACE Quant Model: Wan2.1-VACE-14B-QX_0.gguf
In models/diffusion_models
Quant CLIP: umt5-xxl-encoder-QX.gguf
in models/clip
VAE: wan_2.1_vae.safetensors
in models/vae
ANY upscale model (depreciated):
Realistic : RealESRGAN_x4plus.pth
Anime : RealESRGAN_x4plus_anime_6B.pth
in models/upscale_models
馃摝Custom Nodes :
Description
New text encoder loader, fix on seed node.
FAQ
Comments (28)
The Image Saver-Node is missing in the Description and sadly for some reason i can't use it. The Import always fails even though comfyui is on the newest version and i'm using the newest version of the node :/
It's easy enough to swap out the seed node with just the default generic one so it was easy to fix :P
But maybe others will face the same issue.
And sadly when trying to generate with it, i'm getting this Error:
WanVideoSEImageClipEncode shape '[1, 19, 4, 96, 64]' is invalid for input of size 485376<comfy.clip_vision.ClipVisionModel object at 0x000001D5A9BBBCD0> !!! Exception during processing !!! shape '[1, 19, 4, 96, 64]' is invalid for input of size 485376 Traceback (most recent call last): File "G:\pinokio\api\comfy.git\app\execution.py", line 327, in execute output_data, output_ui, has_subgraph = get_output_data(obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) File "G:\pinokio\api\comfy.git\app\execution.py", line 202, in get_output_data return_values = _map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) File "G:\pinokio\api\comfy.git\app\execution.py", line 174, in _map_node_over_list process_inputs(input_dict, i) File "G:\pinokio\api\comfy.git\app\execution.py", line 163, in process_inputs results.append(getattr(obj, func)(**inputs)) File "G:\pinokio\api\comfy.git\app\custom_nodes\ComfyUI-WanVideoStartEndFrames\nodes.py", line 586, in process mask = mask.view(1, mask.shape[1] // 4, 4, lat_h, lat_w) RuntimeError: shape '[1, 19, 4, 96, 64]' is invalid for input of size 485376Image i'm trying to generate is at 512x768 Resolution while the imput images are 4096x6144
This does not generate any issues in the usual workflows. The issue happens inside the "Clip Encode" Node :/
Thanks for the test, I modified the node for the seed and changed the text encoder loader.
@UmeAiRT聽It works great now! I've just posted my first results from 3 different Start & End-Frame videos.
Sometimes it takes 2-3 Videos to get a good result, sometimes the videos don't end with the correct end-frame but i'm sure that's entirely on the technology and not the workflow 馃
The only thing that's weird, is that the previews for the videos do not work. I can wait until the whole process is done and the videos will be in my output folder, but the videos do not show as previews in the nodes like they do in the other workflows, maybe one of the outputs is not connected or something 馃槀
Have you managed to get triton/sageattention to work with this workflow?
If i set the attention_mode ot sageattn and un-bypass the Triton node in the "Model preperation"-Group i'm getting this Error in the WanVideoSEModelLoader-Node:
must be called with a dataclass type or instance
2025-03-23T22:35:12.078157 - !!! Exception during processing !!! must be called with a dataclass type or instance 2025-03-23T22:35:12.086157 - Traceback (most recent call last): File "G:\pinokio\api\comfy.git\app\execution.py", line 327, in execute output_data, output_ui, has_subgraph = get_output_data(obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) File "G:\pinokio\api\comfy.git\app\execution.py", line 202, in get_output_data return_values = _map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) File "G:\pinokio\api\comfy.git\app\execution.py", line 174, in _map_node_over_list process_inputs(input_dict, i) File "G:\pinokio\api\comfy.git\app\execution.py", line 163, in process_inputs results.append(getattr(obj, func)(**inputs)) File "G:\pinokio\api\comfy.git\app\custom_nodes\ComfyUI-WanVideoStartEndFrames\nodes.py", line 362, in loadmodel patcher.model.diffusion_model.blocks[i] = torch.compile(block, fullgraph=compile_args["fullgraph"], dynamic=compile_args["dynamic"], backend=compile_args["backend"], mode=compile_args["mode"]) File "G:\pinokio\api\comfy.git\app\env\lib\site-packages\torch\__init__.py", line 2447, in compile return torch._dynamo.optimize( File "G:\pinokio\api\comfy.git\app\env\lib\site-packages\torch\_dynamo\eval_frame.py", line 716, in optimize return _optimize(rebuild_ctx, *args, **kwargs) File "G:\pinokio\api\comfy.git\app\env\lib\site-packages\torch\_dynamo\eval_frame.py", line 790, in _optimize@vslinx聽For the preview I had disabled it to save VRAM and I forgot to put it back. You can right click on each recording node and choose "show preview"
@UmeAiRT聽Aaaaaaah! that makes sense! Thank you 馃鈾ワ笍
Getting the following error while trying to use this workflow:
WanVideoSEImageClipEncode
shape '[1, 7, 4, 72, 72]' is invalid for input of size 160704
Thanks for the test, I changed the text encoder loader in 1.1.
@UmeAiRT聽I get a similar error in 1.1:
WanVideoSEImageClipEncode
shape '[1, 13, 4, 80, 80]' is invalid for input of size 352000
Looks like an interesting workflow though as I guess you could do looping animations with it?
@UmeAiRT聽Still getting the same error unfortunately
@UmeAiRT聽shape '[1, 25, 4, 75, 75]' is invalid for input of size 579375
聽I get a same error in 2.1
@UmeAiRT聽any update on this error?
Just one small bug, seems like, not showing ressults on workflow . Otherwise working like a charm.. Tutti Frutii .. Thanx you !!! GPU 16 GB + RAM 32 GB.. Model I2V-14B-480P_fp8_e4m3fn.safetensors with quantization - fp8_e4m3fn + sageattn.. Model to offload, Text encoder and Clip to main_device, with my PC working this configuration.. I am use win 10 + comfyui portable
and there is maybe, better swap to order with upscale and interpolation , first upscale and only after interpolate videos, result is better
For the preview I had disabled it to save VRAM and I forgot to put it back. You can right click on each recording node and choose "show preview"
@xxxtembel聽I inverted it at the request of users because the interpolation after upscale requires a lot of VRAM. But I also published the interpolation and upscale in separate workflows
@UmeAiRT聽thanx fixed
@UmeAiRT聽I understand, I'll will swap it for myself, I don't use upscale to more than 0.4
Works great, thanks!
Thank you! Are you using the exact models suggested or have you made any changes?
@UmeAiRT聽I downloaded the suggested ones. Got errors with other models.
@NoneofYourBusiness2聽Ok thanks
I don't think the WanVideoWrapper node will support gguf in a short time.
My configuration is 4070tis 16g and I can't run fp16 so I'm using fp8 to generate video that's blurry in the middle and indistinct but the first and last frames are clear and I can't control the exact frame rate in the workflow just by the yellow nodes
Is this compatible with GGUF?
Not at the moment, I have to update it for that
If it's not compatible with GGUF why does it say GGUF in the title ?
Don't get me wrong, you're doing a great job, just fix this few things.
The first version was GGUF but the results were very bad and I forgot to remove GGUF from the title sorry for that
Please update the GGUF version, thank you.