**Don't forget to Like 👍 the model. ;)
GGUF: https://huggingface.co/city96/Wan2.1-I2V-14B-480P-gguf/tree/main
https://huggingface.co/city96/Wan2.1-T2V-14B-gguf/tree/main
This is an Image-to-Video workflow using the new Wan-AI Img2Vid model!
Very consistent and generates videos at 16 FPS. I also added the "Extend a Video" function to this workflow—simple to use.
As always, instructions and links are included in the workflow. Just make sure to update ComfyUI and the required nodes before running it! You will need at least 16GB VRam to run it descently.
Worth it? Is it better than Hunyuan? Well... I think it is.
That’s it. Leave a like and have fun! 🎥✨
Description
Added Interpolation, changed the Sampler for a better version of Split sampler, and other small changes (I don't really remember because it was made as I use the workflow... it's just better now.)
FAQ
Comments (22)
Looks great, but I keep getting this error :(
Missing Node:
Int To Number (mtb)
I found this info, but it did not fix it:
Open the ComfyUI Manager.
Click on Custom Nodes Manager.
Search for comfyui-easyapi-nodes
Can anyone help please?
Do you have manager installed?
Try downloading the URL from the github page to ComfyUI manger
I got the same error when using Comfy manager. It kept saying import failed. I ended up deleting the folder in custom_nodes, doing a 'git pull' to pull the repo into the custom_nodes folder and running pip to install the requirements. When I fully closed Comfy and relaunched, all was good and the error was gone.
@Sam_A Sorry for the late reply, my 5090 died, so I have bought a different model.
Yes, I do have Manager installed, it showed nothing missing.
I'll load it up tomorrow as I've been busy sorting the PC and forgot where I was up to :p
@Sam_A Yes, I do
look like it only load the first lora
I add 2 lora below but it skip it
[rgthree-comfy][Power Lora Loader] Matched Lora input "Wan\tool\Wan21_CausVid_14B_T2V_lora_rank32.safetensors" to "Wan\Wan21_CausVid_14B_T2V_lora_rank32.safetensors".
Requested to load CLIPVisionModelProjection
The "Matched Lora input" input message doesn't mean it's not loading all Loras. It only means it's fixing the path of an existent Lora in a smart way. I keep my fast lora in a "tool" folder, for better organization. You probably just droped it in your general Loras folder (as you can see in the message...). But it loads all the loras unless if the Lora have a problem itself. I use 3, 4 Loras at once. About the CLIPVisionModelProjection, which node asked for it? Did you check the paths to all models in the workflow? In each Model Loader node (Load Diffusion Model, VAE, CLIP, etc.) you need to check if the path to the model is right. I use folders for organization, so the original path might be different from the one you use. Click on the models names and select it in your own Comfy. Ok?
@Sam_A thanks it just confuse other workflow will always show which lora is load but this one just show only this
the clip request it just another line the generate work fine
@pasunnazacrifa Great! It's probably the Lora loader node different. I like to use this one because you can have everything in one place without add/remove/enable/disable stuff. I'm happy you could to make it work!
No module named 'sageattention' how can i fix this?
sage attention is a pain to install, try this https://youtu.be/ZBgfRlzZ7cw?si=RX8NxqgbBWPhffiP
I keep getting a black output from the workflow, what do I do? I am using:
Models:
Diffusion Model: wan2.1_i2v_720p_14B_fp8_scaled.safetensors (720p version, fp8 precision)
VAE: wan_2.1_vae.safetensors
CLIP: umt5_xxl_fp8_e4m3fn_scaled.safetensors (UMT5-XXL with fp8 quantization)
CLIP Vision: clip_vision_h.safetensors
GGUF Fallback: wan2.1-i2v-14b-480p-Q4_K_M.gguf (Q4_K_M quantization)
Workflow Settings:
Steps: 20 (tried increasing to 50-100)
Video Length: 5 seconds
Resolution: 720p long side
Sampler: uni_pc
CFG: 2.7
Environment:
Docker container with CUDA 12.2
RTX 4090 24GB VRAM
All NVIDIA libraries properly linked (libcuda, libcudnn, libnccl)
Hummm... It usually happens when a model is downloaded with errors. Do this settings works in any other wan workflow? I mean, the models...
this one is only 2 months old, and already... :
Missing Node Types, no record in comfyui to download from.
number Multiple Of
Number to Int
Context (rgthree)
Display Any (rgthree)
Power Lora Loader (rgthree)
easy imageSize
Label (rgthree)
Seed (rgthree)
Fast Groups Bypasser (rgthree)
and yes, I have the current rgthree package, it's in my nodes as a category.
I installed it 2 days ago in a new comfy using manager... Nothing wrong with the workflow. Install manager, and manager will have the option to install missing custom nodes.
It's working fine here and all those nodes you're missing exist and are retrieavable from the install missing nodes.
Are you working on a version for wan 2.2? I've really enjoyed your workflows they seem the most comprehensive and always have everything I need.
I'm testing my version of this one. I just didn't have enough time to stop, organize and post. But I will, yeah.
Sam_A I'm so glad. Thank you for the work you do. I'll be sure to tip once the new version is here!
int2string and JWImageResizeByLongerSide nodes missing. Manager doesn't even give an option to install these missing nodes. How can I fix this?
i think updating my comfyui and updating nodes used in here fixed it