A workflow uses the latest Wan2.2 Plus Insight model. Better In Character Details, Dynamic Effects, Camera Control, Prompt Adherence, and More.
Please check the video for detailed instructions before you start:
If you think this is helpful, please like the video and subscribe for more.
Description
FAQ
Comments (11)
followed the instructions - I get no errors but just black output? 5070ti latest drivers
same, RTX4090
There are custom nodes not available through manager you need from the model creators GitHub
@m0no1 Do you have a link? I cannot find the EnhancedBlockSwap node anywhere
I got black screen when using "new" models in elder wf's. Maybe, it's sampler\sheduler problem.
I have fixed black screen unconnected CFG and change to cfg 2 or more
Any way to make this faster? It's a lot slower than my other wan2.2 workflow. about 2 1/2 minutes per sampler vs 50-60 seconds with the same res and framerate from my other workflow. Running on a 5090 with 32GB a VRAM. I did notice it was only using about 40% of my VRAM vs the other workflow I have using 95%.
It uses less vram because of batch size node (second from upper-left), lower it from 40 to 16 and you will see difference in vram (maybe even OOM, but with 32gb it's less possible than with mine 16). AND I never saw OOM on comfy nodes, but on wvw - here we go, unless you don't use crutches, like "batch stuff" from also wvw.
About why it's slower for you - I can't tell. For me, it's faster than mine wf for about two minutes for 720p 77sec. TWO MINUTES, dude (yeah, now it takes +-5 minutes, not +-7).
Anyways, it's one and only wf on wanvideowrapper nodes that not OOM's for me (difference - torch stuff bypassed, hm... ). But (as all wvw stuff) it has some problems, now with input picture resolution, which I kinda managed with another crutches, but it works now almost without croping input picture (also I deleted most of unnecessary stuff so wf looks now like not so scary).
upd: damn, sorry, I used Palingenesis Version of wf, it's a bit different, but not so big different. Maybe will try insight version, when finally someone tell me where's difference between them. In terms of face consistency - palingenesis works suspiciously good (maybe because I'm drunk).
@forfreelsd368 128 frames with sa_oda_samler ? omg 15 min on 3090 for me . cfg 2 shift 8
@Relven can't say about 128 frames (oom's for me), but 81 frame 704p with uni_pc, 5 steps, cfg 2, batch 40 (vram 87%, 4070tis 16gb) - 5 mins first run, 4:30 second (with full offload of all models). But it's Palingenesis version (with a bit of changes).