My ComfyUI workflows for using Wan 2.2

No, only when something must be fixed or optimized :)

That does not mean you have to use it, if the other works for you XD

@darksidewalker yeah but u said every new version get better and got some fixes, i like fixes and when stuff getting better

@Moonshi As I use the versions myself, it is always done to fit my needs and standards :)

Btw version 1.7 AiO , I saw in its gguf nodes a model named Tasty Sin, is this something new coming soon to us :D ? or it is a private secret ingredient model exclusive to you ;) ?

@hazzoom82659 You spotted it! I just released it with an announcement :)

@Moonshi Uhm.. we're gonna restrict creators sharing their amazing and free work now on how many things they may upload in which interval or what? ;) Just don't download it then, if you don't feel like it? I mean, it's not like you're forced by auto-update or something. So let's just relax on stuff, aight? :)

@Lyara Why do ya so toxic? its a joke, i'm not aim to insult dasiwa or something

@Moonshi Sounded more like you were in the beginning; it's not that obvious to everyone how you meant it, you know. ;) Aand I'm being calm as a purring cat here, so all's fine then. There are people claiming these thingies seriously, so you need to excuse oneself, if you seemed like it, even if not meaning it. ;)

@Lyara look at my avatar its my face at this moment.

@Moonshi Haha, alrighty, I'm not sure what you wanna hear now, so I'll stick with "sorry to hear that"? 😇However, good that's clarified now. ~

Just be friends ❤️ No harm was done, I got it all right and did notice it was a joke 🤣

@darksidewalker well im glad we are fine bro, wanna ask u something, how i understand your models it is Wan base model with some additional stuff added by ya, so this additional stuff is working with different languages? or just only english, can't find answer by myself from my tests

@Moonshi I did not alter the language understanding, so I assume it understands all what the base would. I think English and Chinese are the main languages. But I also saw Russian prompts. You can test with basic WAN 2.2 or the other models.

@darksidewalker i tied a lot and didn't find much difference to be honest, so there is no much difference i guess.

@Moonshi Language comes from the CLIP text-encoder not inherently from the checkpoint itself. So there cannot be much difference. Someone had to train a other clip for other languages. But I do not think I'll do that, it is not worth the time if 95% just use English anyway and most other AI tools only understand English.

@darksidewalker yeah thats true, but what about lora's for wan? do they only work on one language like if lora creator main is english lora will only work with english promt?

@Moonshi language is converted into tensors from the text-encoder, so like:
If "kissing" is like 1.00002 it will be in ENG or RU or any other language if the text-encoder can convert "kissing" by understanding the language. So it does not matter what you write as long as the encoder can understand it and use the correct tensor for it. If you train a lora with prompts the prompts are converted from the text encoder and if "kissing" remains 1.00002 it will regardless of the language.
But if the encoder does not understand the word like "kawaii" and you prompted it on training it will just use it literally like a "trigger-word" because it does not know. So it would not understand regardless what language you use because it only knows it literally.

@darksidewalker thanks for provided information, very interesting, so if im gonna use enchanced encoder with nsfw stuff that understands what bj mean it would works fine with your bj lora on my native language, cuz your bj lora doesn't have any new trigger words.

This is a old point from older workfows, just select the feature you want. I'll delete it :)

It does not save both at this point.

To have the original, you can just use the same seed with no interpolation and get it. I felt that having 2 or more types from one generation with no real purpose just accumulates files to manage. So I ditched that.

@darksidewalker I see. Okay, thanks. I'll keep that in mind about the seeds.

I have no clue. I can not imagine what should happen to repeat 28 times, there is no loop logic inside, so this can basically not happen inside the workflow.
I used AiO 1.7 for days now, with no issues.
Anyone have an idea?

Im not have any problems like u said, may be its custom node problem, some or few nodes may be overwritten and cause instability, do not forget to check loop count.

@darksidewalker I updated all custom nodes, restarted comfy UI, checked your raw workflow for missing custom nodes again (just in case but it wasn't missing any).

But when I went to close the workflow and it asked me to save changes, like comfy had changed something to the workflow just from viewing it. I said screw it and saved it, I then started a new workflow, dragged your 1.7 out there, made all the tweaks, tested it... and it worked fine. I imagine this is all do to ComfyUI's weirdness and my own user error. I'm by no means a master with Comfy, I just know enough to be dangerous (obviously).

Anyway, I think you can disregard this, I imagine it was 100% on my end and Comfy seems to have resolved whatever the issue was (for now).

Dunno. Never had this problem anywhere.

Your RAM is just not enough. Maybe your configuration can not swap out enough with 40gb. A Q8 can work with 16gb VRAM, if you have about 32+ GB RAM, it will also swap, 64GB RAM are recommended. My guess is you need a Q6 or Q5 model, but I did not measure it

@darksidewalker thank you very much <3

You have to read up on the custom_node pages how to properly install them or maybe use comfyui-manager. I can not help you there, if you install it correctly it will work.

could anything going wrong, but I had similar issues - all custom nodes not installing via the comfyui node manager. It seemed to be a python venv issue for me, comfyui was trying to install it under system python and not my comfyui venv.

So I had to install each custom node / dependency manually, cloning via git and then installing each with its requirements in the correct venv. All working after that.

@hansomething i'm not really sure what i did to make it work to be perfectly honest but I just asked chatgpt how to fix it. it made me run a bunch of git, pip, upgrade, etc in CMD and it seemed to be downloading stuff. slowly but surely, it fixed every node problem. I think I did the same thing you did

i think i'm beginning to realize that maybe i should've used a node manager, and i wouldn't have had to run like 150 different lines of code in CMD that downloaded random stuff a robot told me to download. but oh well

rgthree will not work with Nodes 2.0, read his warning

@hansomething sometimes that happen to certain custom nodes packages whose dependencies are not installed by the comfyui manager and you have to install those nodes manually following their instructions on their github page, as you already did. This is more an issue with those node packages than with a specific workflow