NokiMo
Aitrepreneur
Aitrepreneur

patreon


WAN IMAGE-TO-VIDEO & TEXT-TO-VIDEO WORKFLOW!

Hey everyone! I've just released a new video on my "THE NEW OPEN-SOURCE IMAGE TO VIDEO KING! ", called "WAN". Simply THE BEST AND MOST POWERFUL AI VIDEO MODEL TO DATE!!!! It's just amazing!

You can check out the video right here: https://youtu.be/H8Ky0LhRaGs

I also created a 1-click installer for the Webui for my Patreon supporters for Local AND Runpod, that you can find right here: 1-Click INSTALL WAN IMAGE-TO-VIDEO & TEXT-TO-VIDEO ComfyUI WebUI! | Patreon

Here the link to the manual install guide: https://pastebin.com/NNrgKW68


You can also checkout my troubleshooting video if you are not sure what to do: Exclusive Patreon Video: Troubleshooting Flux Workflow Comfyui!


Down below you will also find the IMAGE_TO_VIDEO-TEXT_TO_VIDEO-ALL & WAN_ONLY-TXT+IMG WORKFLOWS for free for you to use!

As always, supporting me on Patreon allows me to keep creating helpful resources like this for the community. Thank you for your support - now go have some fun😉!

WAN IMAGE-TO-VIDEO & TEXT-TO-VIDEO WORKFLOW!

Comments

Cannot join private Discord for workflow errors: "You need to verify your account in order to prerform this action" Which Account?

rc

he just doesnt respond

哈哈 实在

Says youtube video is restricted?

Teresa Greasley

Does anyone find the missing node types error?

Marco

everything is working in each and every workflow shared with just updating all through Comfy manager

sudhir madan

As I wrote in the other comment, there is a problem since the FF8 WAN IMAGE TO VIDEO doesn't work, that's because the Load diffusion model only accept safetensors model (t2v) while the i2v are in GGUF format, we need to address it someway. Edit. I figured out that I was using the wrong node for the operation, meaning that using the WAN for the GGUF does the work, but then I don't understand how can the other note be useful since it only load tensors (I've downloaded all the models).

Ryuke

I came here to ask the same thing. I'm too new to this to try and add it myself (yet) but I see a lot of how people 1/2 their generation time while still maintaining decent quality. Was wondering what thought Aitrepeneur has about it.

Ryan Prosser

same here!

Chris Troughton

someone else with problem "ffmpeg is required for video outputs and could not be found." i tried a lot of things, nad doesnt work

Whatareck

any plans to add teacache and sage attention to the Wan workflow? I've been looking everywhere for info on this! Thanks.

James

That worked, thank you!

Sagah

Done it. Just followed this tutorial: https://www.youtube.com/watch?v=6sim9aF3g2c

babybatista

Having the same issue

Sagah

same problem here

babybatista

hi ! what is the best practice interpolation before upscale or upscale before interpolation ? thanks

Cédric Morvan

I have this error : VHS_VideoCombine ffmpeg is required for video outputs and could not be found.

Li

Click the "Fit View" icon at bottom right.

J. Lehmann

Same, no errors, everything's updated but it's just empty and not showing any nodes. The V3 Workflow works for me.

Yoho

Anyone got this error ? "UnetLoaderGGUF Unexpected architecture type in GGUF file, expected one of flux, sd1, sdxl, t5encoder but got 'wan'"

93e Chambre

what models are you using because with 16gig of Vram you should be using a quantised model. Also what dimensions are you using? It is always quicker to reduce the resolution then upscale later. DO not try generating 1k, 2k etc straight off the bat

Boyo's Medicine

please list all of the models and the precision, your prompt, the size of theimage.

Boyo's Medicine

I created club lora with specific colours. That gave me some consistency. I then generated the different characters in the club. When you do generations you then use the last image from the render as the first frame of the next render. You just join them together then. It it quicker to generate two runs of 100 frames than try to generate 200. You can keep the coherence on the render as well because if you push the model to high in terms of frame generation stuff will fall apart.

Boyo's Medicine

I set 'length' in WANIMAGEPARAM node to 49 and then set 'frame_rate' to 10 in Video Combine node. 49 frames at 10 fps = 5 seconds.

ChudJones

how do you create such long video?

Swarls

how do you get 5 second clips? I only get 2s clips

Swarls

í keep getting this error when trying the image 2 video . "The size of tensor a (128) must match the size of tensor b (16) at non-singleton dimension 1 " any idea on how to fix it?

StormHawk

i downloaded both JSON files and "dropped" them in and got blank screen. nothing on the screen. anyone know whats going on?

Dat Nguyen

It takes me 2500 seconds (40 minutes) to generate one vid on GGUF WAN IMAGE TO VIDEO with the default settings. i'm on a 3080 ti 16gb vram. normal? any tweaks I could try to get better speeds?

Lagartija

Thank you for this. I was up and running very quickly. FWIW, I've had great success generating 720x720 I2V clips at 10fps and interpolating to 30fps. On a 4090, that's 11 minutes to generate a 5 second clip with 49 frames, then 1 minute to interpolate. If you use the 480 model to generate 640x640 and interpolate, it's 6-8 minutes and still pretty good quality. My clips don't have a lot of fast motion, so YMMV.

ChudJones

Nice work! I didn't have my normal machine handy, so I installed on an 8 year old laptop with a GTX1050. Everything ran as expected (slow). IT took about an hour to run the samples. So, I grabbed a few year old threadripper with an A6000 and it was a much more enjoyable experience. But, I had to give that one back. I'm now in the process of loading on a couple year old i9 with an a4500 and an a4000 in it. We'll see how that does, since I can use this machine for a while. The only issue I had was one of the models(?) said it was no longer available. Nice script, really simplifies things.

MonkeyDanceParty

Same, can't find the missing nodes when searching.

Martin Thoburn

This is something I knocked up using AIpreneurs workflow https://www.youtube.com/shorts/3zUewB9lPa4

Boyo's Medicine

There is Lora support for Wan but not in native comfyui. There are nodes that allow you to use loras should you find any. Just search the comfyui manager.

Boyo's Medicine

No LORA support for this yet, right? This model seems to be really bad at rendering *certain details*.

Rick

i got the same Problem... andy Help? :)

Brian Abey

That's very helpful thank you!

Kubilay göze

By default comfyui manager is set to normal security which prevents lots of things updating or using github via the manager. You need to modify the ini file. The ini file can be found in this path: ComfyUI\user\default\ComfyUI-Manager Open the ini file in notepad. Then look for security. Change normal to weak. Do your updates then put the security level back to normal. This security is for you protection so do not be lazy and not flip it back.

Boyo's Medicine

Anybody getting "workflow>WANIMAGEPARAM loading error" : 1. Make sure you updated ComfyUI to the latest version (right now 0.3.18) and drag the workflow again. 2. If you updated from Manager and you're still getting the error; MOST LIKELY your update didn't happen and you didn't realize. Go to the "ComfyUI_windows_portable\update" folder and click "update_comfyui.bat" and if it still doesn't work click "update_comfyui_and_python_dependencies.bat" Go to Settings-About and check if the version number is at least 0.3.18

Kubilay göze

Cog does this with the inbuilt workflow called interpolate. You just put the start end frame and the prompt for what you want. It is far superior to animate diff. Other Models do this as well but for low end GPUs and for better results than animate diff this is a solid option.

Boyo's Medicine

Same issue, Update did not fix issue. Needed to do a total install in a different location. if there is a brute force update option. that would be great!

NewBe2

Actually there is a way to do that, might add it to a future workflow, unless we get something better soon, seeing at how fast everything is going nowadays

Aitrepreneur

Lol :) ... love the video !!! Ok thx !

ElecMat

How difficult would it be to do a img2vid with 2 inputs, where it generates the animation between the two? Kinda like animate diff does.

Euchale

The native comfyui implementation does not allow you to apply attention which will speed things up. I can run 98 frames in 6 minutes but my card is a bit bigger than yours. Use these nodes as they are far superior to the standard implementation. The workflow is in the nodes. https://github.com/kijai/ComfyUI-WanVideoWrapper You will need to download different models though and put them in a folder in diffusion models to not mix them up with the comfyui standard implementation models This is an example of a Wan Video I made with the wrapper: https://www.youtube.com/shorts/Jt8d8HKZVPU

Boyo's Medicine

the "workflowWANIMAGEPARAM" missing error resolves itself after you drop the "IMAGE_TO_VIDEO-TEXT_TO_VIDEO-ALL.json" file into the workflow area again after a restart. At least it did for me. You do need the latest version of ComfyUI (0.3.18), so update that first. Have fun.

LW

I crank the Length in the Ksampler to 81. I use the WAN GGUF i2V 480p Q5. It took 1:08:01, 136.05s/it for a 5 sec clip on a RTX 3060 12 GB. Not sure if there's a way to make it run faster but the quality is good.

ElecMat

Just increase the frame amount.(look in the sampler) This will be GPU limited then. If you use KJs wrapper for text to video he has been banging over 1 minute long videos from a single prompt. I am finding that the native Comfyui implementation is not as good as KJs wrapper due to the fact that using the wrapper generation times are 33% faster and even faster still if you compile the model which is an option on the wrappers workflow.

Boyo's Medicine

Is it possible to make videos longer than 2 seconds with the Wan 2.1 model?

ElecMat

I said in the video (and in the Patreon post installer) that if you don't know what to do, watch the troubleshooting video, I explain everything step by step: https://www.patreon.com/posts/exclusive-video-116007628

Aitrepreneur

cant seem to get these fixed - can you point to the nodes git plz for me ! Missing Node Types: workflowskyreelsparam workflowskyreelslogic

Kusacubari

ah gotta update to latest comfyui version

Occavango

Have you tried a different image of 1024x1024 pixels and set the video size (width + height in WANIMAGEPARAM) to 480x480?

Klaus The.

I used your full comfyui + models installer in local Windows 11 and the img2vid workflow runs like a charm. Thank you very much! (I'm using the WAN2.1-480p-Q5 GGUF on a 4090; ~4 min per video)

Klaus The.

Hello what logic nodes are you using as I am getting missing nodes with regards to the skyreels section. I have my own sky reels workflows so I know it works on my comfy but you have a group that the manager cannot unpick to find the required nodes. I am getting this argument error Int-🔬 (In group node 'workflow>skyreelslogic') workflow>skyreelslogic Compare-🔬 (In group node 'workflow>skyreelslogic') If ANY return A else B-🔬 (In group node 'workflow>skyreelslogic') workflowskyreelslogic

Boyo's Medicine

Update - I had ComfyUI already installed, so just loaded the models. The issue, the version I had was 0.3.13 and the wan option was only added, as I understand, after 0.3.15. I tried to update comfyui but it would not move past 0.3.13 even when trying to switch via the Manager panel (If there is a way to force this, it would be of great help to know. An absolute fresh install using the TEXT-IMAGE-TO_VIDEO-COMFYUI-MANAGER_AUTO_INSTALL.bat in a new location seems to have fixed the issue as I now see the ComfyUI version as 0.3.18 and the 'wan' option is there within the Load Clip node. Also, no node errors after running 'Install Missing Nodes' so essentially it works well..but I'd rather be able to do this on my old installation.

Walker4k

Everything I tested works fine except the model "wan2.1-i2v-14b-480p-Q5_K_S.gguf". I did not change anythin in your workflow uploaded an image into of course "GGUF WAN IMAGE TO VIDEO" and I get the Error "UnetLoaderGGUF cannot reshape array of size 5369813 into shape (5120,3520)". I also tried 2 other workflows from other creators and get the same error only from "wan2.1-i2v-14b-480p-Q5_K_S.gguf"

Adaptalabor

Yeah, tried that 'Update ComfyUI / Update ALL' but still no joy

Walker4k

Wan is missing from drop down in Load Clip

Occavango

Update ComfyUI itself in the Manager, and then try to reimport the workflow.

Greg

Weird error. Any ideas? - Missing Node Types When loading the graph, the following node types were not found WanImageToVideo (In group node 'workflow>WANIMAGEPARAM') workflow>WANIMAGEPARAM Int-🔬 (In group node 'workflow>skyreelslogic') workflow>skyreelslogic Compare-🔬 (In group node 'workflow>skyreelslogic') If ANY return A else B-🔬 (In group node 'workflow>skyreelslogic') workflowskyreelslogic workflowWANIMAGEPARAM

Walker4k

Can you please help me resolve these error messages?WanImageToVideo (In group node 'workflow>WANIMAGEPARAM') workflow>WANIMAGEPARAM Int-🔬 (In group node 'workflow>skyreelslogic') workflow>skyreelslogic Compare-🔬 (In group node 'workflow>skyreelslogic') If ANY return A else B-🔬 (In group node 'workflow>skyreelslogic') workflowskyreelslogic workflowWANIMAGEPARAM

Occavango

Turns out i had a Wan vae that does not work with your workflow - the ones that didn't work are labelled bf16 and bf32 - only vae that works is the "wan_2.1_vae.safetensors"

Orpheus_Reup

Any idea how to solve 'VAE' object has no attribute 'vae_dtype' I have checked all the models and they are in the right place. Have updated comfy and nodes. Not the first time i've had this error with vid models but can find little about what it means or how to fix

Orpheus_Reup

Umm what is this BS? Followed everything in your video. Came here and became a subscriber. Setup runpod and can see comfyui. NO workflow there. So I guessed that needed to be uploaded to the /default/workflow folder in Jupiter. OK so did that and refreshed the comfyui..The worfkflow is a mass of red boxes with nodes that cannot be found? What am I doing wrong? EDIT i installed missing nodes..but still have these not found!?? I have terminated my runpod as I am getting charged and it's not working. So once you tell me what is wrong and how to get it working I will try again. LTXAttnOverride (In group node 'workflow>LTXPARAMTEXTOVIDGGUF') workflow>LTXPARAMTEXTOVIDGGUF ModifyLTXModel (In group node 'workflow>LTXPARAMTEXTOVIDGGUF') LTXPerturbedAttention (In group node 'workflow>LTXPARAMTEXTOVIDGGUF') workflow>LTXPARAMTXTFP workflow>LTXIMAGETOV Int-🔬 (In group node 'workflow>skyreelslogic') workflow>skyreelslogic MathExpression|pysssss (In group node 'workflow>skyreelslogic') Compare-🔬 (In group node 'workflow>skyreelslogic') If ANY return A else B-🔬 (In group node 'workflow>skyreelslogic') workflowLTXPARAMTEXTOVIDGGUF workflowLTXPARAMTXTFP workflowLTXIMAGETOV workflowskyreelsparam workflowskyreelslogic

Zoran Kanti-Paul

What LARA's work with WAN?

Jack Weems

Was waiting for this! Let's see if my 4080 + 32gb ddr4 RAM can deliver this baby.

Raphael Paz

how much vram minimum?

Kilobe

I'm not sure why it does not start, I cloned actual version from git I assume there is venv somwhere and its missing some stuff based on the errors at start like: ModuleNotFoundError: No module named 'dill' ModuleNotFoundError: No module named 'pydantic' ModuleNotFoundError: No module named 'git' ModuleNotFoundError: No module named 'pykalman' I do not see the whole node manager on side, where I should click and update tne nodes, the whole panel does not load.

VladoPortos

Absolutely, I'll update it asap

Aitrepreneur

Any chance to fix V3-ULTIMATE_FLUX-COMFYUI-MANAGER ? That one installs but there is no option to update missing nodes because the node manager does not start :(

VladoPortos


Related Creators