My ComfyUI workflows for using Wan 2.2

By the way, my settings are: 4 steps, 1-1.5 cfg, 16 fps with interpolation up to 32, resolution 608x1072

@MrDxrk I would assume your comfy is broken somehow. I have overall better times with the recent 1.1 across the board.

What I can do is to look if I somehow uploaded a broken workflow, I'll have a look.

@darksidewalker I don't know about the broken comfyui. After a restart, the artifacts disappeared, but the time is still very long. I might go back to the older version, which had no time issues.

The workflow is working for me. I did change the sampler for testing, but this did not have impact on the time. I generated 4 videos with 528*768, 81 frames (with lora and multiple samplers). All in about 220s to 320s, Euler+Simple was the fastest.
But all inside the estimated time range.

@darksidewalker Do you update all custom nodes to the latest versions through the manager? Perhaps my problems are due to the fact that I don't update node versions.

@MrDxrk Yes my nodes are all up2date.

@darksidewalker In short, artifacts and noise still appear. All nodes are updated, all settings are default. This didn't happen with 1.0.

@MrDxrk I can not reproduce it, sampled like 20 videos for a project now and nothing is noisy.

@darksidewalker Unfortunately, I can't find any pattern to this problem other than the version, since in the previous generation all the generations were normal, but in this one only one turned out normal, and all the others have artifacts and noise, as if I'm watching really bad cable TV.

@darksidewalker By the way, where in your workflow can I choose Sampler/Scheduler? I haven't seen such an option.

@MrDxrk I would help, but I have no idea atm. The sampler settings are inside the sub-nodes.

@darksidewalker The sampler and scheduler were completely different in the i2v subnode. I changed them to Euler/Normal and trying to generate right now. Perhaps that was the problem.

@darksidewalker No, it's not the scheduler or sampler. The video still has noise and artifacts. I'm downgrading to the previous version.

I found a bug, did you disable NAG? This somehow breaks the workflow. I have to elaborate why ...

@darksidewalker Yes, I disabled NAG because I was using CFG 1.5. So that's the problem?

In 1.0 I also turned it off, but there were no such problems.

@darksidewalker If this helps, in the node manager KJNodes showing conflict with WanVideoNAG.

@MrDxrk I do not think this is the problem, I think there some connections wrong. I'll look into that

@darksidewalker Okay. I'll using version 1.0 for now.

@MrDxrk Should be fixed in 1.3 now, the sub-graphs do not work with model paththrough ... well lets say it is a bug of comfy.

The S2V needs the S2V model and the I2V needs the both high and low I2V models

There is no better, they have different purposes. As stated in the page description. EX is experimental. At first I would recommend AiO, I use it myself.

thanks for your patience and disposal

I did falsely deactivate the "save output" in v1.1 for testing. Just activate it.
On v1.2 it will be fixed.

@darksidewalker I did but I still have the problem with the corrupt mp4 file. Only when using this workflow tho, your model works flawlessly.

@tenshiabse the workflow just uses ffmpeg to create the mp4... nothing spacial. Maybe you did not install all dependencies?

@darksidewalker So far... I installed all the links in the "requirements". On the 1.3 version I still get the video glitching at the end, it flickers black, like it wasn't "completely compiled" or something like that (the 5 seconds of the video are ok, this happens at the end). Just to make sure is not that, is ok to use this workflow with Midnight flirt?

@tenshiabse I use the wf daily and it works with midnightflirt no problem. why your video is glitching I can not tell. You could join my discord and we can discuss there with maybe some screenshots?

@darksidewalker I would like, at least to share one of the videos so you can see if they give you some problem too

@tenshiabse please do 2 things:
1# try the backend test workflow and look if the problem still persists.
2# update to version aio 1.5

@darksidewalker Sorry can you explain the "backend" thing?

@tenshiabse Load this https://civitai.com/models/1823089?modelVersionId=2405252
and press generate. If there is a working video, your beckend (comfyui) is ok. If not, your comfyui is broken.

Wow a lot has happened in the last few days, I just can't catch. Guess what? Now it seems to work... I don't really know if I did something or not, but this 1.6 does work as intended.

Oh. I have to look into that, there is also a bug with disabled NAG

The double speed I could not recreate anyhow, but I fixed a bunch of bugs in 1.3

@darksidewalker Hm ok will try 1.3 and maybe look into it further

i'm also getting this with 1.4

@yannberg240398 Need more info or I can not do anything...

@darksidewalker If I have the bug again, want me to save the workflow and send it to you?

@yannberg240398 Please try with AiO 1.5 and if the bug occurs, send me on discord

@darksidewalker In 1.4 it seems somewhat fine now but somehow the order changed. In Detail when Using the Workflow before we had Subgraphs it generated the I2V Video and then used that Video to do the Interpolation and it worked fine. Now with the Subgraphs in Version 1.3 I had the Issue that the Workflow when FLF2V was enabled the Workflow would not function at all. It would start with the Upscaler Step and error out because it hadn't even loaded the image or anything before that yet. Restarting Comfyui reloading the Workflow and anything else I tried had no effect on the Behavior. The I2V Parts worked fine. With FLF2V the Workflow would always start processing somewhere at the end of the Workflow in 1.3 depending on which Options were enabled. Now with 1.4 Interpolation seems to be working again but the odd thing that happens for me is it first creates the interpolated Version of the Video and then the non Interpolated is created. So reversed basically. This results in the I2V Video showing a slow Motion Video and the Interpolated Video running at what appears to be normal speed now. All I did with 1.4 this time was load up the Workflow and enable Interpolation, load up an Image and run it as is. After that I enabled the exact Loop option and loaded up the last Frame image. When I pressed Run it loaded the Clip and ran the Prompts but completely ignored everything else and went straight into the Interpolation Subgraph and errors out because ofcourse nothing else was executed yet besides the ClipTextEncode. It didnt load the Images it didnt load the Models it Went straight Clip -> Interpolation. Hope this helps in some way. To me it appears my Setup gets completely confused by the Introduction of Subgraphs.
got prompt

Failed to validate prompt for output 645:

* RIFE_VFI_Opt 579:413:

- Required input is missing: frames

Output will be ignored

Failed to validate prompt for output 646:

Output will be ignored

Failed to validate prompt for output 334:

Output will be ignored

Failed to validate prompt for output 213:

Output will be ignored

invalid prompt: {'type': 'prompt_outputs_failed_validation', 'message': 'Prompt outputs failed validation', 'details': '', 'extra_info': {}}

@schmongo Oh you did test a lot, but unfortunately I was fast and changed the logic in 1.5. Because I noticed some bloat connections and nodes. Maybe the problems are gone with 1.5.
Sorry for the inconvenience!

@darksidewalker No Inconvenience caused on my End ☺. Just trying to help out by pointing out what didnt work for me so it can get fixed for others who dont speak up.

@schmongo thank you, I couldn't have been as thorough as you.

@yannberg240398 FLF is fixed in AiO v1.6

I'm on 0.3.75, as reference and all is working

@darksidewalker I'm very sorry to bother you. I'm encountering some issues where all my outputs are blurry and have noisy artifacts—any idea what might be causing this? The workflow I'm using is: https://civitai.com/models/1823089, and the model is dasiwaWAN22I2V14B_midnightflirtHigh.safetensors. I've also downloaded the VAE and CLIP as specified in the Dependencies you listed.

@shiikoe you are for sure using aio 1.5? The former had a bug, if you disabled nag it got blurry/artifacts.

@darksidewalker Yes i'm sure using AIO 1.5, i didnt disabled nag, the default settting of your workflow (your hand draw pic )is also blurry and have noisy artifacts.

@shiikoe The hand drawn pic is the standard example pic of comfy, not mine 😅

I do not know what you are doing, but maybe your video resolution is set too low or your initial images has artifacts?

@darksidewalker I also experienced artifacts yesterday. I don't know how it happened. Initially, three videos were fine, but then the artifacts started appearing in the next ones. NAG was disabled from the very beginning. Image resolution 1024x1536. Comfy version is the same as yours, all nodes are updated.

@darksidewalker I'm using aio 1.5's default setting,I tried: change pic, higher vid resolution, re-download models,and reinstall comfyui, but output still blurry and have noisy artifacts. is possible that caused by my server's configruation? I use V100 card with 32G Vram. Now I tried use your models example profile photo and prompts, still not working.

Well, I do not know how to reproduce it, did like 10 vids today and all where fine...

@darksidewalker hi, I sent a pm to you of my server link,could you please take a look? That will be very greatful for us!

Yeah that's a good question and i wanna know answer either, the biggest problem of manually extended video via last frame extraction is that part of details of main image can be lost, like if girl eyes are closed on last frame there is a big chance that girl eyes on next generation will be completely different

There is no automatic extend video feature in my workflows. You have to do it manually or use another workflow.

I can look into that, but did not notice that yet.

For nodes, you can always move them like you want, I like it that way atm.

FLF connection fixed

@darksidewalker thanks