🧩 Wan2.1-FusionX — Official ComfyUI Workflows- WIP
📢 7/1/2025 Update!
New: FusionX Lightning Workflows
Looking for faster video generations with WAN2.1? Check out the new FusionX_Lightning_Workflows — optimized with LightX LoRA to render videos in as little as 70 seconds (4 steps, 1024x576)!
🧩 Available in:
• Native • Native GGUF • Wrapper
(VACE & Phantom coming soon)
🎞️ Image-to-Video just got a major upgrade!!!!!!
Better prompt adherence, more motion, and smoother dynamics.
⚖️ FusionX vs Lightning?
Original = max realism.
Lightning = speed + low VRAM, with similar quality using smart prompts.
☕ Like what I do? Support me here: Buy Me A Coffee 💜
Every coffee helps fuel more free LoRAs & workflows!
📢 Did you know you can now use FusionX as a LoRA instead of a full base model?
Perfect if you want more control while sticking with your own WAN2.1 + SkyReels setup.
🔗 Grab the FusionX LoRAs HERE
🔗 Or Check out the Lightning Workflows HERE for a huge speed boost.
This is the official workflow hub for the Wan2.1_14B_FusionX models that can be found HERE
⚠️ NOTE: The workflow's are embedded inside the png file. Just drag it into ComfyUI and it will load up.
⚠️ NOTE: Each workflow has detailed notes and links to the correct models. Please read the notes carefully before using.
And right here, you’ll find a full set of workflows designed to unlock the model’s potential across a range of generation types, including:
🎬 Text-to-Video (T2V) – Available now. Just drag and drop the PNG file into ComfyUI. (I’ve included a sample video created with the current settings in the folder.)
🖼️ Image-to-Video (I2V) – Available now. Drag and drop png into comfyUI. I Included the start frame from the example video if you want to test it. (Please note: The Wrapper version supports start AND end frame. Native only supports start frame.)
🎬 VACE Wrapper and VACE gguf Native - Use a control video and/or a ref image with your text prompt to have more control of your output video.
🖼️ VACE non-gguf Native - Coming soon
🎬 Phantom Wrapper - Mix up to 5 images into a video for full character and scene control.
🖼️Phantom Native and gguf - Coming soon.
⚠️ NOTE: for image to video, to get up to a 50% increate in overall motion in the video, set your frame count to 121 and FPS to 24!!! After some testing this really helps!
⚠️ NOTE: please read the note boxes in the workflows because there are important details that will help you overcome some error's you may encounter.
Each category will include both:
✅ Native Workflows – Built directly with WAN components for full control and customization.
🚀 Wrapper Workflows – Uses the Kijai Wrapper for optimized generation speed.
These are the same workflows used in all demo videos on the model's main page — no extra LoRAs, upscaling, or interpolation. Just clean, raw model outputs with the right settings.
⚠️ All required components (e.g., CausVid, AccVideo, MPS LoRAs) are already baked into the model. Do not re-add them unless you know what you're doing.
Whether you're looking to create cinematic text-to-video scenes, stylized image-driven sequences, or combine multiple references into a single shot — these workflows are your starting point.
## 📢 Join The Community!
We're building a friendly space to chat, share creations, and get support. I am also adding a channel to include some good motion lora's to help to get more motion in your i2v video's and I'll be adding other goodies here so please join us :)
👉 Click here to join the Discord!
Come say hi in #welcome, check out the rules, and show off your creations! 🎨🧠
Description
gguf node added
FAQ
Comments (53)
Awesome work!! Have you shared a GGUF version of the VACET@V_FusionX_WF? I'm trying to run it on my home PC as opposed to my work PC and I keep getting OOM.
If u scroll to the bottom of the description there are links to them.
Sorry I thought u meant model. Im still working on the wf
is it easy to attach a lora to this? using the T2V it works very for me
It depends on what workflow you are using, wrapper or native.
I am using the wrapper version
I’ve been blown away so far, everything looks so clean
Is there a I2V 720p version too?
The one in here is 720p version
yeah found it. do you have any further tricks on how to tame the beast? What to do if loras dont react as they should? i know this combination was tricky before your merge. that merge seemed to made it more stable and predictable. But there is still room for improvement quality wise compared to the base model. Not so much in resolution but more in captuing little details in movement and moods(character expressions) as well as it struggles to do 2 things at the same time. and then a 3rd. Maybe this is the downside of the acceleration.
In wf vace, you need to add an image loader for the reference. Please add
This workflow uses a text prompt and a control video; a reference image is not required. I've included notes within the workflow explaining its usage.
@vrgamedevgirl Then how does it differ from the wf image2image? Do you have a wf with a motion control video?
@_RUST_ The VACE workflow I shared takes a input video which then extracts the POSE from it. I don't have an image to image workflow. There is an image to video if that is what you mean? If you look at some of the example video's you can see that you can control the video with your body so you can "act out" the movements. does this make sense?
@vrgamedevgirl Yes, I meant image2video. You have a video plus text, and I would like to have my picture in motion, not a random text. But your model is great! Especially t2b)
@_RUST_ Got it! So you're asking for a new workflow that lets you use a reference image. No problem! I’d recommend joining the Discord server—there’s a request channel where you can drop that in. Click here to join the Discord!
WanVideoModelLoader
Can't import SageAttention: No module named 'sageattention'. Someone got an idea ?
If you read the notes in the Workflow it tells you what you need to do if you don't have sagattn installed.
You need to bypass the torch compile node, change the attention_mode in the model loader to SDPA. You may need to change the base precision to fp16. Try this and it should work. It will be a bit slower if you don't have sagattn.
I get the same error, which is strange - I have sageattention installed and it works great in other workflows. But not this one for some reason
great work, the consistency and output quality are amazing , thanks, this is saving so much time waiting for generations, I have found with my setup (Quadro RTX 5000, 16gb vram) generating at 768x384 then up-scaling x2 is 3 times faster than 1024x512 with no upscale, with only a very slight reduction in final quality
In i2v regarding loras, is it better to just rely on prompts or can loras be used for specific required actions
You can and should use lora's for extra motion! :) You just need to bring over a lora node. I'm in the process of updating the workflows to include an optional lora node.
@vrgamedevgirl power lora loader node by rgthree seems to work well in the native workflow
@mofo69 what sort of seconds/iteration are you seeing with that setup?
@imbrium201 at 8 steps and 768x384 then up-scaling x2 81 frames takes an average 500 seconds, as opposed to 1024x512 with no upscale takes 1500 seconds for the same 81 frames. Thats using the wan2114BFusionx_fusionxImage2video.safetensors model., its generating 3 times faster than all previous WAN models and giving better sharper clearer outputs
Wrapper works perfectly. Is it normal that i have to use around 35 blocks for 81 frames with 24gb vram?
U should not have to with that much vram. What res? Can u join the discord and we can chat? Discord link is in the description on the model page
572x1024. Wrapper start to end frame. All other settings are default
Nice work! Is there an I2V GGUF workflow?
Waiting for the same. I had a go at adapting the T2V flow for I2V but not sure I did it correctly.
@jonk999 yeah, i haven't had a chance to dig into it much but it seems the wrapper nodes dont inherently utilize gguf? Can you quickly share how you did it to maybe give an idea incase @vrgamedevgirl is focused on other things?
Sorry, still working on this. Lots of requests. I do know, that you take the native image to video workflow you just need to swap out the model loader with the gguf one.. very simple. I'll post it soon though.
@vrgamedevgirl ah that answered my question, so has to be native. ty!
If also adding a Lora node, it would go before the model shift or after? Or doesn't matter? I notice on the T2V GGUF you have it before the model shift, but other flows I've seen it's been after the model shift.
Just replace the "Load Diffusion Model" node with a "UNet Loader (GGUF)" node, as in the T2V GGUF example workflow.
I'm getting the following error when using the default settings for the Wan_FusionX_i2v_wrapper_WF.json workflow.
!!! Exception during processing !!! Given groups=1, weight of size [5120, 16, 1, 2, 2], expected input[1, 36, 21, 72, 128] to have 16 channels, but got 36 channels instead
I thought it was due to the resolution of my source image, but it happens with an image with the same dimenions, so clearly that's not it. Has anyone seen this?
For me to better assist you can you join the discord sever? when you can share a snippet of your WF. 👉 Click here to join the Discord!
@vrgamedevgirl I can do that, but I'm literally using your workflow with no changes other than the image I'm loading.
@_degenerativeai_ You would have needed to point the models to your own. Its possible your pointing to the wrong model. It happens all the time and is one of the top reasons for these types of errors :)
@vrgamedevgirl You are indeed correct. I downloaded Wan14BT2VFusioniX_Phantom_fp16.safetensors and I saw you give advice to someone on your Discord if they downloaded the models from your links in the workflow. I decided to try that before bothering you on Discord. Thanks! It works great now.
@_degenerativeai_ Perfect!!! But you can reach out on discord and say hi if you want and always reach out if you need help! Thats what its there for :)
@vrgamedevgirl I really appreciate it. I'll definitely stick around. It looks like you have a good community going there.
Hi, is there a solution to this problem? I have the same problem.
Can't find the one that uses controlnets
It just the normal text to video model for now along with this workflow https://civitai.com/models/1663553?modelVersionId=1886466
there are notes that have links to the right models as well in there
Heya, are you planning on making a Phantom GGUF workflow? Having hard time tinkering around that video loader, can't just replace it with GGUF loader (or don't know how)
I am. I'm working on the VACE one now (VACE for Native and GGUF) after that I'll work on the phantom one :) -- To stay updated and get access to new features before they are published i would recommend joining the discord server. See link in description.
I definitely will stay updated, thanks! Love your work, revolutionary <3
Amazing model! Help me please. How can i add Lora to native wf?
You just bring in a lora loader. I'm working on updating the WF's so it has as optional Lora node. come to discord and we can show you
@vrgamedevgirl ty, got it.