HowTo: https://civarchive.com/articles/19108
UPD: New version of Loop + Scenario + Loras
The most powerfull workflow in this series, highly recomend update to this one.
Single precondition text
Single postcondition text
Unlimited scenario steps (5 by default but you can add)
Unlimited lora for all (1 by default)
Unlimited SEPARATE Lora per step (1 by default)
Color correction
UPD: New version of Loop + Scenario + Color is here!
Separate precondition
One-line scenario
Separate postcondition
Color correction is works now : )
Updated version of https://civarchive.com/models/1826257
Now you can use Scenario with different prompt for each step.
This workflow contain For-Loop so you can create as much step as you wish just copy-past Prompt for each step and change what you need.
Do not forget to change precondition (for ex. If you undress wooman and take off sweater do not desribe next step as "weared in sweater" - it confuse model)
Lighting: https://huggingface.co/Kijai/WanVideo_comfy/tree/main/Lightx2v
FastVideo: https://huggingface.co/Kijai/WanVideo_comfy/tree/main/FastWan
Wan2.2: https://huggingface.co/QuantStack/Wan2.2-I2V-A14B-GGUF/tree/main
Description
Change some nodes, added color fixing, change loras
FAQ
Comments (57)
no face consistence
Yeees! Fantom still waiting for me : )))) But it's Wan2.1 and I personally like 2.2 but... in case of nothing will use this one : )
instead of complaining why don't you post a fix/solution? that's how AI works... stop complaining about people putting in time and effort to share with us for free while you do nothing
Any chance to change it so that LORA's are included for every step? Now, it only includes the LORA's on step 1.
Other than that, amazing Workflow.
you can use easy-bypass for that
thesharque588 Isnt that only for if you want to bypass it? I am talking about that the LORA's only appy for the first run, but on run 2,3,4 they are currently being bypassed/skipped.
nevermind, im an idiot! Great workflow<3
YokoBono what did you work out?
jwcoop3r631 It seems as if the LORA's are actually implemented in all runs, but for some reason something feels off... Can't put my finger on it exactly...
Like, for example.. If you use any breast expansion lora's and add a prompt for all the steps that the breasts will expand, they only expand on the first step. then they stop on step 2,3,4,5 But if you only include breast expansion in for example step 3, they will expand in step 3. Meaning that the lora's work on all steps... But again, its like they only take effect for 1 of the 5 steps regardless... Same thing happens with all the other Lora's ive tried for different scenarios.
What Lora did you use for the undressing?
Nothing! just desribe this and it works : ))))
You can download any video - ComfyUI Workflow attached.
Why is this model post always at the top of the list when I sort by Newest? Odd.
IDK, but I just published new video, may be that is the reason.
thesharque588 I have seen this glitch on Civitai before. It didn't have anything to do with the OP, it was some internal issues the site. Maybe some indexing issue or something? No worries.
fyi, you have the wrong model set for this workflow: "Wan Video 14B i2v 480p" there is a separate 2.2 category now.
Old loras works fine : ))) But yes I update them one by one
how is this lora at the top of the list for days?.. how you do that?
IDK... Just make a new movie, set correct resources and publish it...
Something similar happened about a year ago. It's not the OP. It's some glitch with the site.
Someone always has to think themselves 'clever' enough to game the system. Then they cry like a baby when they get banned!
BTW it's not a Lora it is a workflow 🤣
WTF? Published : Aug 1, 2025
@Fleival yeah @theally please do that with my loras too XD
@K3NK Do what?
@thesharque588 keep my early acess loras on top !! xD
@K3NK Try to use your loras as much as possible : ))) it's a really good loras
Thank you so much for the workflow! I would like to be able to set up independent Loras at each stage. Well, an Upscale wouldn't hurt.
Loras is a good idea but already done by another user, let me add this worflow in "Suggested Resources". What about "upscale" - It was here : ))) but on my PC it fall with OOM and I remove it : )
@thesharque588 I looked at the workflow you presented. I found it very cumbersome and difficult to understand. I am asking you to make customized Loras in your workflow. Your workflow is much simpler and performs the same functions as that workflow. Thanks in advance.
@dirtysem You asking for separate loras? Take it : ))
Love this workflow, as ever! I do miss having separate LoRAs for each step, though I suppose WAN didn't always translate that properly anyway in the generation!
I have found that it works far better with WAN 2.2 LoRAs compared to WAN 2.1 - with the latter, facial consistency and video quality take a nosedive, whereas the 2.2 LoRAs seem to largely avoid these issues. Now we just need all of our most beloved 2.1 LoRAs to make their way to 2.2 and we will be golden!
Great work my friend!
Checked the latest 2.2 s+c+c, it's amazing, at 4 step count i can get 16sec vid in 10min at 848x480 on my 4080 with no image degradation. Adherence is good enough, color match is solid between transitions, and the workflow is very easy to see and understand, stellar job. Time to play a bit with loras and prompts, but this is huge to take AI generation out of the 5 sec limit.
You ask for separate loras? Take it : ))
Uh oh, looks like Snakeskin Lora is taking over as the new perma-newest model.
sometimes the real friends were the "hide content from this user" buttons we met along the way :p
Hi, I get the error on the second step, don't you know why?
"File ...\videohelpersuite\nodes.py", line 397, in combine_video
kwargs, passed_kwargs = prompt[unique_id]['inputs'], kwargs
KeyError: '78.0.0.3.0.0.63'"
ok, after removing color match node it works
@darkjudicator551 it looks like you change type. Use mkl color match (other become too error)
Hi, not tested yet. You need to add the Power Lora loader instead of all this lora node. It will be much easier to add as much lora as we want.
You also need to add the new Resolution Master node, it's way better and easier to chose aspect ratio and resolution.
Power Lora - require a clip (and I need model-only)
Resolution Master - yes but it's a +1 more Plugin required for just 2 numbers?
@thesharque588 I just modified your WF. It works perfectly, test it. What i added:
- Sage attention
- Power Loras loader
- Resolution Master
Here is the link:
https://drive.google.com/file/d/1gYRl7W9GPS8X8oPyv2kgsMp1RvY8eYg-/view?usp=sharing
You can add it as WF in your page, just add credit to me in description.
Another issue i noticed after all my tests. The color match ruin all the video. My T2V is excellent quality, but as soon as it starts the I2V part with color match, the quality is awful.
How to remove color match from the WF?
I also tried to add an load image to switch from I2V instead of T2V for first scene, don't work.
@Pat3dx It's a great addon, but... I have Intel i9 : ))) and there no SageAttention : ) But I'm sure community of civitAI thank you for it!
@Pat3dx Post it here as your workflow and I add a link in "suggested resource"
@thesharque588 Can you tell me how to bypass or remove the color match node. All my video are ruined by that :(
@Pat3dx just select this node and click "bypass"
@thesharque588 I've tried this and many things before posting my question, of course :)
Don't work. If i just bypass the color match node, only the T2V first scene is generated.
@thesharque588 Ok, now my T2V infinite scenario work pefectelly i can generate 2mn+ video without any quality loose.
But i just tested your I2V loop with scenario. It don't work good.
Two major issues:
- it don't really follow the prompt 1 line per scene, and mix all Loras, as it do perfectly for the T2V. Example, in the first line he girl just squat and smile.
But when the first scene is generate, she squat and make some action of others scenes, plus the Loras that will be triggered in other scene with trigger words only in the other lines, are triggered in the first scene, when in the first line scene the trigger word are not present.
- The generated video are horrible. pixelated, etc....really bad motion, a lot of flicking, and bad movements.
I don't touched anything, just put my own models and lora. and clicked Run.
After i see the bad result, i changed the sampler, and put my usual one, unipc simple, and also tried euler/simple.
Exactely the same bad motion and video quality.
I don't know what you changed for the I2V WF, but it don't work as good as the T2V WF.
I don't made the same change i made for the T2V for your I2V WF yet.
Resolution master, sage attention and Power Lora loader.
@Pat3dx I have no sageattention and I can't have it (I do not use Nvidia/CUDA) I use split-cross-attention because this is the only way to generate more than 17 frames : )
@thesharque588 I don't changed anything or added anything (i don't added sage attention, power loader and resolution Master yet, i want to test your WF as it is first, as i've done with your T2V WF, before modifying anything) on your I2V Wan 2.2 loop scenario WF.
Just added my own models and Loras. and clicked Run.
So by default, the I2V scenario WF don't work for me, many issues.
When i tested your T2V scenario WF, by default, without changing anything, it was working fine. Witch is not the case with the I2V scenario WF.
The quality is very bad pixelated, motion issues, etc....plus, as i explained, it trigger the Loras at the start in the first scene, when the trigger words are only in others scenes line.
I don't have this issues with the T2V scenario WF. The T2V WF work perfectly. But not the I2V.
@Pat3dx Hmmm... I'ts strange because I've use it almost each day. Did you check the latest version? ... +clip +colors?
@thesharque588 This is the version i'm using.
@Pat3dx Please try last version with separate loras. BTW - I if you want to baypass Color Correction you need to connect "IMAGE" from VAE-decoder and "Images" from "Crop last image", and then you can baypass "Image Color Match". IDK why, but if you just baypass it connect "image_ref" and "images" : )))) So actually connect first image forever.
@Pat3dx thanks for that sage wf!
Could you please make a version for the 5b model? It's so much faster. And the option to start with an image would be nice, too.
Actually I have but quality is terrible! But ok, I will deploy today
@thesharque588 Thank you!
@AndyZocker Here, take it : ) https://civitai.com/models/1911027?modelVersionId=2162986
BTW - If you create something not really ugly like shit, please upload to workflow gallery.