Only tested on Native wan workflow !!
[3/3 [01:03<00:00, 21.17s/it]
120 frames with 3 steps took me 60 sec using sageattn.
-
-
Steps: 3.
Cfg: 1.
Shift: 8 ( 1 to 10 is great, have not tested any higher)
Sampler_name: Uni_pc or Uni_pc_bh2. ( all of them work but Uni_pc is the best lol)
Scheduler: Normal, ( simple, beta or others work too)
For vace lip sync, do not use too many lora's, it will screw up the preprocessor. I prefer not to use any lora.
-
-
Wanvacetovideo Strength: 1.00.
Width: 416.
Height: 720.
-
Set wanvacetovideo strength to: 0.70 and use this prompt with max 81 frames. You can use any image of a woman, close-up or full body image, with or without a dog, it does not matter.
0, 24 "A woman is walking peacefully through a sunny park, holding a leash as her playful dog trots beside her."
24, 49 "Suddenly, the dog breaks free and dashes off— the woman gasps in surprise and reaches out, calling after it."
49, 61 "As she starts to chase the dog, she trips over a root and tumbles to the ground, startled and flustered."
61, 81 "Moments later, the dog turns around and runs back toward her, tail wagging, as if trying to help or comfort her."
-
You can use higher resolution such as 480x832 or 576x1024 but 416x720 has great quality too.
Omegawpn_vace_autoprompt_4-7 steps_workflow can be found here see link below.
https://civarchive.com/models/1806036/omegawpnvaceautoprompt4-7-stepsworkflow
Credit goes to the original creators of the checkpoints. My only contribution was merging them.
Description
FAQ
Comments (17)
32.3GB is quite big! I assume this is a merge of both wan 2.2 checkpoints and the Wan2.1 VACE module? Maybe some speed up lora as well?
The only wan2.1 model that is merged is the Lightx2v V2 lora.
OmegaWPN So how is this named Wan2.2 Vace?
LovelaceA Lightx2v V2 rank128 Wan2.2 High&Low_noise_Vace_FP16
I'm confused too. There is no Wan 2.2 Vace yet, as far as I know. Only Wan 2.1 Vace.
rrgadabvhjxzqxnscz345 wan2.1 vace cannot be used stand-alone, it needs the wanvideo wrapper which you need to use a normal wan2.1 t2v model and wan vace t2v separately. Aka wan2.1 cannot be used on native workflows unless its wan2.1 fusionx vace or a gguf version
So for vace loader we put in this checkpoint and for the WAN loader we put in the high and low model? Or am I missing something?
is there a super basic workflow that this can work with? The one linked is huge. One just to create videos. not audio/florence, etc.
I will make a super easy and simple workflow when i get home from work.
also for i2v please :)
Yannick90 I just made Wan vace to video : https://civitai.com/models/1841883?modelVersionId=2084402
I will make a Image to video workflow soon.
Just for my curiosity, how many Gb of GpuRAM do you need to store this checkpoint, plus the 120 frame animation?
Im sorry, i can't really say. Im using Rtx 5090 (32gb vram with 128gb ram), but a rtx 4060 TI with 16gb vram could do 81 frames without a problem.
OmegaWPN How is possible, since the file is 32Gb itself?
settima_ai It swaps from vram to system ram. That's why you need 32GB of ram and higher for video gen.
Any plan for gguf version?
I never tried merging or converting to gguf models so i don't know, sorry.