UPDATE:
I've been getting better results with the lightx2v lora distill from kijai (https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Wan21_T2V_14B_lightx2v_cfg_step_distill_lora_rank32.safetensors), which makes the 2 sampler setup obsolete because this works great at CFG 1, so this is just a quick update if anyone wants to use lightx2v with the same workflow layout and settings that work for me.
ORIGINAL:
I've seen people talking about using 2 samplers to increase the movement/quality of CausVid, but no one's posted a workflow yet so I thought I'd share, because I've been impressed with the results.
The basic idea is using the KSampler Advanced which lets you split the generation into 2 or more samplers with a start and end step, you generate the first 3-6 steps using a lower CausVid strength and high CFG, then send the latent with leftover noise to a 2nd sampler to do the rest of the steps at a higher CausVid strength and low CFG.
I've kept the workflow simple, it's adapted from the workflow I typically use for my Wan videos. Should work fine with any of the 720p/480p I2V models, adjust the sliders as necessary.
I'm not sure what the best combination of sampler/CFG/steps/CausVid strength is yet but this one seems to work well, let me know if you find any improvements.
Description
Using the lightx2v lora from kijai gives better results without the 2 samplers setup, so this is just a quick update to replace the causvid 2 sampler section with lightx2v and just the 1 sampler. Otherwise same workflow.
FAQ
Comments (22)
Just added a v2.0 workflow using the lightx2v lora which gives just as good or better results without the need for 2 samplers. Keep the cfg at 1.0, 6-8 steps and increase the shift if you need more movement.
Hi there! So Causvid is lesser / not worth using b/c x2v is better? That's basically what I've found. Although the quality of 480pQ6modelQ3clip loader is clearly worse than running without the x2v model..... What are you seeing that works well? Just super basic simple motions? Anything needing a lora or more expressive motions makes the x2v not worth using.
Quality wise v2 single sampler blows the v1 dual-sampler away. I was super surprised by the difference.
I saw other notes where people were talking about motion being "stiffer", however, and what I'm seeing is that's from the differences in training data and lora merges that were in the dual sampler setup. Switching to base wan2.1 + single sampler and lightx2v produced better results visually but subject performance went dead without the extra training motions, my guess.
I swapped out base model to get some better behaviors than the generic wan2.1 base but it changed the look of the output too drastically, despite having more interesting motion. I got back to feeling more like the results I was getting in the dual sampler setup by mixing in Acc, Causvid and FusionX loras. Much more lively now and "alive" feeling.
Just tried V2.0
Result ended up being a 45 second long video which is just 7 seconds played on repeat for no reason and I'm not seeing a way to change that. Also, not a fan of the "webp" saver, V1.0 was better because it was a 1 click simple download which only saved the video, but this wants to save the entire viewable page.
What do you mean by webp saver? It saves the video the exact same way as v1.0. I only swapped out the sampler and causvid for lightx2v, otherwise the workflow is the same. I just noticed the load image wasn't connected to the resizer, so I fixed that. I have no idea how you got a 45 second video, it should be the length of the Frames slider.
@definitelynotadog somehow the issue i mentioned was for a different workflow, because i went back through my file names it flipped in my downloads and this one got renamed to another. All is well with this one lol. Sorry about that.
where do i get the sageattention files?
I installed sage from here
https://civitai.com/articles/12848
Thank you for the workflow! I've a few questions.
How to adjust the worklfow,If I want to increase the steps? Do I increase total steps by two and "End at step" is always half of "Total Steps"?.
And what about the strength of the causvid loras. Why the different strengths at the ksamplers?
For some reason I can't add the new version of workflow to Comfy, and I was so excited to see this update :(
For both V2.0 and V1.0 I used the wan 2.1 i2v 14b 480p fp16 base model with sage attention and I used a concept lora.
V1.0 generated videos in record time, but most of the videos were unusable.
With V2.0, my animation turn static and most of the time nothing moves at all, not even at step 20.
Is the download broken? I tried importing but it is just a blank workflow, no nodes.
Scroll out and look to the right?
Same here, nothing Happens after import, v1 works
@srfuentes99 Thanks it was there
Having the same issue, weird.
@logenninefingers888 zoom out very far
@f6825800990 for me nothing happens the current Workflow stays open. Your Workflow does not Open, no new tab. But still no Error Message
I'm noob - how do i get this to work? its producing 1 second videos for me. I don't see where to change the video duration or the aspect ratio. Thanks!
hey anyone share v2 link because files are empty
1 - Open the file in a text/code editor and remove the area that adds the UNET loader.
2 - The workflow will load then all nodes on top of each other.
3 - Connect the kijai diffusion loader to the Sage Attention node and change settings accordingly.
FIX UNTIL CREATOR UPLOADS NEW FILE!
1 - open json file in code/text editor
2 - look for a section that has UNET
3 -delete that section and save file
4 - load file into comfy
5 - all nodes will be on top of each other, move them around to see them properly
6 - Connect the kijai diffusion loader to the sage attention
7 - change settings accordingly