v3.0 Is Finally Here!
Entirely new layout with grouped step-by-step sections.
New detailed Notes section with links to essential model downloads.
Added Res4Lyf custom node pack for Sigma previews and access to more sampler/sheduler combinations.
Thanks to BillyWicked for their collaborative efforts in testing and refining this version!
The workflow is still fully capable of NSFW content if you wish to use it for those purposes. Feel free to share any of the videos you are able to generate using this workflow!
I HIGHLY RECOMMEND READING THE FULL INFORMATIONAL ARTICLE REGARDING THIS MAJOR UPDATE - https://civarchive.com/articles/21684/torstens-wan22-i2v-gguf-workflow-version-30
v2.0 - still simplified and very efficient
added more quality-enhancing features without impacting rendering speed!
v1.0 - basic workflow, extremely fast, not as many features
TL;DR
It's a great workflow, some have said it's the best they've ever seen. I didn't say that, but other people have. I think it's the best workflow anyone has ever uploaded. You know what we need on this platform? We need to Make Workflows Great Again!
Description
Equally efficient and simplified, more optional features added for better generation results with little to no impact on speed! Please read the full V2.0 Release article for important information.
FAQ
Comments (38)
hi! where can i find cuberpunk a vfx explosion 30 loras?
"Cuberpunk" (filename is Cuberpunk, the Lora is called Cyberpunk 2077 Style) - https://civitai.com/models/1292050/cyberpunk-2077-style-wan-hunyuan-video-lora
Explosion - https://civitai.com/models/1295945/huge-explosion-vfx-hunyuan-video-lora
finally a working workflow for my 12GB card at last. Made some changes and tweaks for convenience.
A question tho, are you sure you are using the color match node right? because I'm sure the image target should be the output after decoding, and the final output would be after it.
Correct me if I were wrong.
I realized I had a couple of nodes out of place and I am working on a Version 3.0 to release very soon, with some bug fixing and adding more optional groups of nodes without affecting performance.
Hi can i please request an auto prompt or/and a loop version of your workflows. You make the best ones.
I appreciate that! I will consider working on auto-prompt / loops in the future
I second that request! Also, any chance you could do a wan 2.2 t2i workflow?
I will continue working on these in the coming week or two. I'm a hobbyist AI creator, doing this in my limited spare time when I'm not working my demanding career field :)
TorstenTheNord Much appreciated!
I've seen experiments where we just need the LOW NOISE (we can skip high noise). I hope we can reach that stage (in WAN 2.2). Thats because in LOW VRAM machines (6gb, 8gb), using high and low 'exchanges' (swap) those checkpoints and sometimes crashes comfyui.
Only workflow to work for me so far and its amazing. Other workflows would take my gpu 1-2 hrs and only turn out to be garbage but here it only takes like 6 minutes for a 65 frame video which is super clean, which I never though wouldve been possible. Would highly appreciate if you could make a seamless loop workflow as those are just so satisfying to watch compared to the 3 second clips. Also an upscaler in the same workflow would also be really nice, the fps booster is already doing its wonders but the quality is still low. Am really thankful non the less for this amazing workflow, keep the good work up
Is there anyway to fix the slow motion problem?
Hey mate, your workflow is a life saver! I was getting OOV constantly using Wan2GP (alternative to comfy, focusing on videos, simpler and supposedly better for GPU poor) using I2V, and even T2V, without even upscaling. I was afraid to touch Comfy until I found this. Now my best videos are using this amazing workflow. Can't thank you enough!
Thanks for the nice workflow, working really well, was using a different wf with Q4, now running Q6 and went from about 500 sec. to 370 sec. for 97 frames but much higher quality output.
One thing though, this model seems to take huge liberties in terms of changing the model in my reference picture. Any thoughts on where i should start looking into it?
Ah.. user error as always :D I was using a vertical aspect reference image and horizontal output. So the face of the model was never part of the reference due to being cropped out :)
It runs fine but im getting distorted and blurry results. not using any loras
try to change the scheduler and sampler to euler beta 57 on both ksamplers this works well for me
also since you are not using any lora it is probably due to not having enough steps, with lighting lora 4 steps total 2 each model. But without lightning lora you need to set the steps higher, like 20 total 10 high 10 low or 8 high 12 low
seconding what @yoavlevit17323 said, but also I found another scheduler/sampler combo that works GREAT it just requires another custom node set to get the samplers & schedulers. Res_2M or 2S samplers are great, especially with Beta57 and a shift value of about 5-6.
I WILL BE CREATING VERSION 3.0 IN THE NEXT WEEK!
Sorry I've been absent from responding to comments, I'm just a hobbyist with AI in my limited free time outside my demanding career.
are all of your loras here on civitai?
@yoavlevit17323 what the 57? euler beta gives me similar results. Increasing steps gives me only static as output
@Asaghon to get Beta57 you'll need the Res4LYF custom node pack, and increase Model Shift value to around 8 on both ends. Version 3.0 is coming out today or tomorrow which uses Res4Lyf custom nodes as well.
@bob717 I believe I had to go to Huggingface for a couple of them because of CivitAI's weird censorship issues.
@TorstenTheNord Yea had already found it but hadn't increased to 8. Amazed at how good video's are looking tbh
I really love your workflow. I originally tried to get zardozai's workflow working well, because the motions are superb. But it always came out looking a bit distorted for my setup.
When I tried yours, the results look great, but the motion wasn't great. So I thought, why not use your workflow with his lora and sampler settings. Boom, instant gold.
Lightx2v high at 5.6 strength
Lightx2v low at 2 strength.
Samplers on Euler
Schedulers at Simple, Normal or even Beta 57 if you want. Doesn't seem to matter.
I didn't notice much difference between the different schedulers, but I'm not really an expert and don't know what to look for.
Anyway, with these settings, your workflow is now my go-to for everything. It looks spectacular and the motions are god-tier :)
It says I'm missing the wan models for the Unet loaders, how do I fix that?
By downloading them and putting them in the unet folder?
Any attempt at further reducing time to generation with sage attention and fp16 accum only seems to increase generation time. Has anyone gotten this to work, if so you do have a workflow?
I know where to get the models...but I am not sure where to get the loras?
Thanx, new to video and got this working (Q4 on 12Gb vram). The animation looks pretty good but faces get pretty blurry. Not sure I I just need higher resilutions or if I'm doing something else wrong (832x480 and all only lighting lora enabled).
use 848x480 for 16:9 upscaling compatibility. I'm almost finished testing a few new theories to release version 3.0 this week. I've been very busy with my full-time career so testing and development has been slow.
@TorstenTheNord Ah no worries, I'm already amazed that everything worked fairly smoothly. Wasn't expecting that at all. I'll try that res
12GB Vram? So a RTX4070? If yes u can use the Q8 checkpoints through multigpu gguf node. Thats what I did until last week before I upgraded my gpu. Q8 will work without hassle I guaranteeee!!
@sikasolutionsworldwide709 Yea running rtx 4070, really regret not waiting for a super
@TorstenTheNord Do you have any recommended workflows or guides for upscaling?
@kitblasters339 I love using this upscale workflow. It includes easy-to-use face swap/restore and interpolation. https://www.nextdiffusion.ai/tutorials/how-to-upscale-videos-in-comfyui scroll down to Requirements and Downloads and you'll find the workflow json file.
Greatly appreciate you, @TorstenTheNord! Cheers!
@kitblasters339 SKAL!