My ComfyUI workflows for using Wan 2.2

No, just i2v. Almost all videos are with metadata to copy if needed.

I'll have a look, thanks for the hint :)

I read a bit about this, it is only for NV users, so nobody witth other gpu's can use this.

@darksidewalker Confirm. I disable upscaled setting, I test add RTX nodes in between 'Checkpoint settings' & 'video combine'. More sharp than before.

@mpayne1985106 It is already planned for my next release as 3rd option

v6.3 is online, adding RTX SR upscale (2x)

@darksidewalker Hell yeah!

You can load any audio clip and use it with cozyvoice or s2v.

I assume this is a problem with wrong or missing dependencies. I can only advice to use my installer or repair the comfyui installation.

Never mind, turned out to be an issue with ComfyUI's new Nodes 2.0

Did you use my comfyui installer or made sure every dependency is installed correctly?
For me the setting is working, besides 10s is not a good thing for wan 2.2.

@darksidewalker thank you, maybe its beacuse i use the github version installer. My purpose is to generate longer and stable video, someone told me i need to use SVI. I thought it can be achieved with your wf.

@lzq13060022351140 SVI needs a SVI workflow. Like the other I provide.

I guess some dependencies or custom nodes are broken. Nothing I can do, you have to install comfyui and nodes correctly.
If you are not sure how:
https://civitai.com/models/2364056/tool-dasiwa-comfyui-installer

@darksidewalker I'm encountering the same issue and the funny thing is I was able to generate one video earlier today but the moment I wanted to generate another one, I get job#<x>. This specific node looks like its a default note by comfyui, so I was wondering what version you're on and if you are on desktop or portable. I am on the latest version of desktop, and I am some what confident its due to the latest version of comfyui causing the issue. Also, I want to use your installer as a last resort mainly because I don't want to re-set up everything.

@revolti If it helps, I'm on:
ComfyUI 0.17.2
ComfyUI_frontend v1.41.19
EasyUse v1.3.7
rgthree-comfy v1.0.2512112053
ComfyUI-Manager V3.39.2
Python Version 3.12.10
Pytorch Version 2.10.0+cu130
... All custom nodes latest (as today) and some re-deployed fixed pip packages (for dependencies like hunyuan-foley).

@darksidewalker hmmm, that was helpful and I think the issue boils down to the version difference for ComfyUI_frontend v1.41.19 vs v1.41.20. Too bad I have no idea how to rollback the update and I was hoping your installer would install 1.41.19 one but it didn't (Also getting the same error on the custom install one as well). So I cannot confirm if my theory is true.

Maybe once v1.42.5 rolls out, your workflow will start working again.

@darksidewalker I second @revolti concerns. I too seem to have the same issues while running the flow, I manually added the load image node at the GetResolution subgraph it passed, but the error now happens later down the node at, I believe the VFX make video node maybe? I don't remember that exact node but images do not seem to work will with the new comfyui update. It very will may not be your issue, but I am just giving you a heads up.

@good_bye2005 I can confirm it is not my installer or workflow:
I did a fresh install with my dasiwa-comfyui installer on W11 and Linux and run a generation just fine, no errors.

@darksidewalker it is not a problem with your workflow or the installer, it is the latest update that has messed up how the subgraphs are taking in inputs. I edited quite a bit, by unpacking and repacking the problematic nodes, its seems to work for me, but it most definitely not your issue. As always comfyui updates breaks workflows.

For anyone facing issues, you would have to "unpack" and then "repack" the subgraphs and then connect the corresponding inputs.

@good_bye2005 damn bro, thank you very much!

I guess some dependencies or custom nodes are broken. Nothing I can do, you have to install comfyui and nodes correctly.
If you are not sure how:
https://civitai.com/models/2364056/tool-dasiwa-comfyui-installer

Hi, for Stability Matrix, here's the trick :

When you start Stability :

ComfyUI > 3 dots on the right > Change Version > Change to 0.12.1

ComfyUI > 3 dots on the right > Python Packages > frontend > Change to 1.37.11

Then, on the workflow, uninstall rgthree-comfy

Apparently, it automatically installs the right version because now, it's working perfectly for me... and faster than before.

Don't ask me why xD.

✅ Solution, use my installer to use the functional version and frontend or manually downgrade to:

ComfyUI 0.16.4

ComfyUI_frontend v1.39.19

@darksidewalker I used to get that comfy version and it started to work. Also installed new version of workflow... And had to do a clean install on GPU drivers too. A lot of work but I can enjoy it again

@Biiva_kun v6.7 works with the newer comfyui versions too

@darksidewalker I note that. Sadly other workflows I use not xD

@Biiva_kun I believe that, was really tough to build almost the complete workflow anew with changed nodes to make it work ...

Same issue. Tried reinstalling all the nodes, tried 6.3 and 6.1 workflows and neither works (where they previously did).

Fresh install , same issue , i`m dead without your WF bro :)

I finally got a video generated by bypassing a lot of the groups and just getting the VAE decoded images to the Video Combine node and running that alone. But I've repeatedly had one image link break intermittently, I think triggered by saving the workflow (the link going into the Checkpoint Settings - Resolution Logic subgraph). Recreating that link then allows generation again. I guess something is broken - but this was a clean ComfyUI install just with using ComfyUI Manager to install all the missing custom nodes.

@a17575 I would assume some custom nodes and pip dependencies broke by updating. My made installer works. I'm on latest comfyui and nodes.
Fresh install does not always make right dependencies, they must be checked.
My workflow has no missing wires, as I'm aware. Did like 20+ gens today on v6.3

yeah i got the same problem. not sure what to do

@XDghost425 As I can not tell what is wrong with personal installations of comfy, I can tell mine works if all dependencies are correctly installed. So an fix could be to use my installer.

@darksidewalker I do agree it's something with my install rather than the workflow, because I've just had the same wire disconnect again from doing Ctrl+Z and save. But I'd like to just fix my install rather than using your installer if possible. Any steps you'd suggest? Does your installer lock particular custom node versions? I just did standard install of all missing nodes in your workflow through ComfyUI Manager.

@a17575 My installer resolves all dependencies automatically and makes sure some pip packages are re-locked to versions the nodes need.
It also makes sure that pytorch matches the driver and the correct python major version is used.

This can all be done manually, but many ppl will mix things up.

The pb is for me for example is that i run on runpod , not sure if the instaler can be used on it

@darksidewalker 

Prompt executed in 0.01 seconds

got prompt

Failed to validate prompt for output 28:

* (prompt):

- Required input is missing: images

* VHS_VideoCombine 28:

- Required input is missing: images

Output will be ignored

Failed to validate prompt for output 1512:1588:316:

* GetImageSize 1512:1588:328:

- Required input is missing: image

Output will be ignored

Failed to validate prompt for output 1512:1588:318:

Output will be ignored

Failed to validate prompt for output 9:

Output will be ignored

Prompt executed in 0.01 seconds

Tested on ComfyUI latest version and fresh installs of 0.17.2 portable and 0.16.4 portable. All three gave same error. Using the AIO with I2V.

EDIT: I wrapped that output in a code block and the preview shows it formatted but for some reason it isn't formatted after I add the comment.

latest comfy ui update completely borked the workflow

@sjwjetty463 @agentgerbil well, there is not a thing I can do about that. It boils down to a correct install of comfy with matching dependencies for all custom nodes.

@darksidewalker I mean... I can't really mess up a portable install. With comfyui manager I should just have to load your workflow and then install all missing nodes with their defaults, correct? Or is there something more that needs to be done in order to verify the correct versions are used?

@darksidewalker using AiO workflow v61 worked without making a single change from the environment in which v63 was throwing errors.

@darksidewalker I installed another instance of ComfyUI through your installer, and started getting the same errors.

I think I figured out what triggers it: If I just import the downloaded workflow it works perfectly well AS LONG AS I DON'T SAVE IT. As soon as I save it, I start getting the errors. No idea why.

Seems to be related to the rgthree custom node not liking the latest release of comfyui

@RandomBozo Than we can only wait for an update, I personally don't have the problems, but I can not fix the custom nodes.

I can confirm it is not my installer or workflow:
I did a fresh install with my dasiwa-comfyui installer on W11 and run a generation just fine, no errors.

@darksidewalker the installer- can we use it for runpod ? i`m tehnically inept :D , no clue if we can use it as a template or inside the Runpod ,thankssssss

@cristianlyon67811 I don't use runpod and so I do not know how this works.

@plsknotthisagain So if you import the workflow without saving then you have all necessary wires and it runs? Did you drag the .json into your UI to load it the first time?
@RandomBozo I was suspecting comfyui-essentials because it seems that custom node is no longer maintained and might be rotting.
@darksidewalker any chance you could post your working pip freeze to pastebin or something? PS: The main reason I don't want to use your installer is because I don't want to use uv and have parallel Python caches eating my disk space. Your workflow and checkpoint are awesome tho, got really good motion and prompt adherence, it's good enough that I'll eventually cave after I've tried a few things 😂

It seems rgthree broke on Windows. You can only wait for an update from the custom nodes or downgrade everything.

I can confirm it is not my installer or workflow:
I did a fresh install with my dasiwa-comfyui installer on W11 and run a generation just fine, no errors.

@@darksidewalker it was running smoothly before update

Downgrade the frontend or wait for rgthree/comfy to fix it.
To downgrade add startoption to the launcher:
--front-end-version Comfy-Org/[email protected]

✅ Solution, use my installer to use the functional version and frontend or manually downgrade to:

ComfyUI 0.16.4

ComfyUI_frontend v1.39.19

It is not, works absolutely normal on my end, you are doing something wrong or comfy is broken the workflow works since weeks.

Maybe you are on Windows there are some broken custom nodes with latest update 0.17.2

I can confirm it is not my installer or workflow:
I did a fresh install with my dasiwa-comfyui installer on W11 and run a generation just fine, no errors.

@darksidewalker It seems to be a problem with ComfyUI; I encountered the same issue after updating.

@darksidewalker So, do you suggest to completely do a fresh install from zero to make your workflow works ? I have the same error since I did "Update all".

@zerbukii have tried multiple completely fresh installa, using darksidewalker's install, other 3rd party installs and the original comfy install. None make a snot of difference. Regardless of where the issue lies, it all adds up to make comfy an unstable and unreliable option. Not that I know an alternative, but as soon as there is...

@PD63 yeah, same problem, really frustrating... i was using it for days and then without any change (just comfyui update) it just stops working...
just editing to add that I tried to use a generated video to load the worflow and then used this loaded workflow to generate new videos and it is working

Downgrade the frontend or wait for rgthree/comfy to fix it.
To downgrade add startoption to the launcher:
--front-end-version Comfy-Org/[email protected]

✅ Solution, use my installer to use the functional version and frontend or manually downgrade to:

ComfyUI 0.16.4

ComfyUI_frontend v1.39.19

Same here... however, if I change the prompt, it fails again.

Manage to solve the problem by uninstalling "comfy-rgthree"

@zerbukii Is it not necessary to reinstall it? From what I remember, this node is a fairly important node.

Downgrade the frontend or wait for rgthree/comfy to fix it.
To downgrade add startoption to the launcher:
--front-end-version Comfy-Org/[email protected]

@darksidewalker Initializing frontend: Comfy-Org/[email protected], requesting version details from GitHub...
Failed to initialize frontend: 403 Client Error: rate limit exceeded
Falling back to the default frontend.
ComfyUI frontend version: 1.41.20

So sad...

@darksidewalker I replaced the files in D:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\comfyui_frontend_package\static with the v1.39.19 version, and successfully rolled the frontend back to 1.39.19 at startup, but the issue still could not be resolved.

ComfyUI version: 0.17.0

comfy-aimdo version: 0.2.12

comfy-kitchen version: 0.2.8

Initializing frontend: Comfy-Org/[email protected], requesting version details from GitHub...

[Prompt Server] web root: D:\ComfyUI_windows_portable\ComfyUI\web_custom_versions\Comfy-Org_ComfyUI_frontend\1.39.19

Asset seeder disabled

Is that what you had in your console ?

got prompt

Failed to validate prompt for output 1512:1588:316:

* GetImageSize 1512:1588:328:

- Required input is missing: image

Output will be ignored

Failed to validate prompt for output 1512:1588:318:

Output will be ignored

Failed to validate prompt for output 9:

* VHS_VideoCombine 28:

- Required input is missing: images

Output will be ignored

Failed to validate prompt for output 28:

* (prompt):

- Required input is missing: images

Output will be ignored

Prompt executed in 0.03 seconds


Since my last "Update All", that workflow no longer works.

@zerbukii yep, just gotta downgrade the frontend version to resolve,
Comfyui version is 0.17.0 btw, with all custom nodes at the latest version.

✅ Solution, use my installer to use the functional version and frontend or manually downgrade to:

ComfyUI 0.16.4

ComfyUI_frontend v1.39.19

✅ Solution, use my installer to use the functional version and frontend or manually downgrade to:

ComfyUI 0.16.4

ComfyUI_frontend v1.39.19

@darksidewalker Ok.

You need to upload your own audio file as reference.

Comfyui issue or setting if some nodes are demanding too much system rights.

Also, I got some cases when installing through Comfyui manager does not work, It is missing some nodes. The solution for me is to delete the custom nodes folder and trigger an install using git (you can do it inside comfyui manager). That worked for me on several ocasions.

✅ Solution, use my installer to use the functional version and frontend or manually downgrade to:

ComfyUI 0.16.4

ComfyUI_frontend v1.39.19

@darksidewalker Yesterday I downgraded the frontend and now it works.

Thank you elaboration

drop a solution for windows I cant fix it I have been trying since yesterday

✅ Solution, use my installer to use the functional version and frontend or manually downgrade to:

ComfyUI 0.16.4

ComfyUI_frontend v1.39.19

✅ Solution, use my installer to use the functional version and frontend or manually downgrade to:

ComfyUI 0.16.4

ComfyUI_frontend v1.39.19

Happen to me also, for some reason it works fine on the first gen and after few tweaking on resolution and post processing features I got the same error as yours, so I'v unpacked the subgraph from Resolution Logic and you can reconnect into all the node that need this input,
all the image links that goes to color match and interpolation was broken too

Yes I got the same issue with the last workflow

Downgrade the frontend or wait for rgthree/comfy to fix it.
To downgrade add startoption to the launcher:
--front-end-version Comfy-Org/[email protected]

@darksidewalker I just did this but the error stays, thank you

@HmNike can you upload the fixed wf json somewhere ?

@Elmer588 I can not upload a fixed wf (also mine works), it is not a workflow error.... Comfy/rgthree needs to be fixed. Nothing I can do.

@darksidewalker I think the get image subgraph into global subgraph is causing it with the windows version, rgthree seems to not like it for the image link, can be also session cache on my end, I'm using the comfy desktop version and I'm used to this sort of weird behaviour from comfyui, I did try the frontend fix without success

@Elmer588 I will upload a video you can check if it work on your end 

@HmNike did you able to get it work?

@XtreamRish yup, you can drag the last video I'v uploaded on your comfyui see if it works for you

@HmNike It did not work. I am using VastAI instances. I deleted everything and started from scratch. It only generated one video and then same error started to appear.

@SappAI Seems different fixes work for different setups. I tried all the options suggested along with complete comfy reinstalls, none worked. I went back to version 6.1 of the AiO workflow and that works consistently now for me. Worth a pop.

6.1 doesnt work as well

✅ Solution, use my installer to use the functional version and frontend or manually downgrade to:

ComfyUI 0.16.4

ComfyUI_frontend v1.39.19

@cristianlyon67811 for me neither now

Restart comfyui or make a bug report on comfyui, this is not a workflow issue.

Never mind, it only worked once...

It works until you save the workflow, and then it breaks.

same

✅ Solution, use my installer to use the functional version and frontend or manually downgrade to:

ComfyUI 0.16.4

ComfyUI_frontend v1.39.19

Make sure to activate flf2v with the switch

@darksidewalker For me - FLF2V was not visible in front end. I enabled it all in back end, and then it was visible - and working :) Thanks!

i reported the issues 2 days ago. you need a previous frontend version on top of that (go check DaSiWa answer there)

@Le_Fourbe doesn't work for me. I'm using linux and have already downgrade the frontend to 1.39.19.
~/comfy/ComfyUI$ pip list | grep -i comfy

comfyui_frontend_package 1.39.19

@kyonn92 if you installed a new, make also sure to redownload the workflow, the newer comfy break it and deletes wires.

@darksidewalker I already redownloaded but still got the same issue

@Chibako I did 3 new installs yesterday and all worked, if I deleted the old comfyui and workflow first.

@darksidewalker thanks. I reimport the workflow and it works

@darksidewalker Thank you... was having an issue with "image" error and didn't seen that wires needs to be reconnected... It's a shame how ComfyUI can break older workflows like that.

Okay ... than don't use it ...

Rude

Using darkie's WF and checkpoints for a couple months already because it is literally the best. It is hard for you probably because you're one of these "UUUM I WANA MEKE SUM PORNIE" dummies. WF is awesome for creators.

open your eyes dude there's a solution literally in the description lol

@eugskb532 Not everyone's eyes are as sharp as yours. In this ComfyUI workflow, there is a specific node with a connection that tends to break.

It works the first time you execute it, but as soon as you save the file or refresh, that connection automatically disconnects. This specific line links to a sub-group (and potentially further into other sub-groups).

Trying to find that one broken connection among hundreds of nodes is like looking for a needle in a haystack. If you know exactly where you've connected it, you should take a look.

@kent1026 The problem is the newer comfyui's they delete wires and doubles input values on json file level, destroying the file itself.

if u are using svi workflow.did u try to save the workflow and reopen it? and did u duplicate the workflow and renderd more then 1 video on the duplicated tab?

Will elaborate :)

@darksidewalker if it works do you have a updated svi workflow with rtx node?is there a way to test out the new nvfp4 version?

Does not work, breaks after saving. Will wait for next release to test.

@hartweizen As I do not understand the trail feature and how this would work for EA (because of the absence of any detailed explanation on civitai), there is no way to just test it.

@darksidewalker i wanted to try the render times rtx5090 with resolution faktor 1.3. safetensore vs nvp4 vs gguf 8.
for me gguf has way better prompt adherence.gguf sd resolution already does what i need and safetensor needs 1.15 for similiar results(i2v) i speak about overall motion not details.

@hartweizen Well, as I have no way to make this happen before EA ends you would have to get them or wait. On my tests fp8 is slightly better than q8, but maybe it depends on other factors too. Normally safetensors is superior to gguf at least for speed and memory management.

I feel that frustration, I also have to redo and repair so many things, it is eating time I could spend for fun things. But I really want anyone to enjoy the tools I make, so I care to make them work and better, but sometimes it is hard to keep up. 🫤

Thank you so much for everything you've done and for all the effort you've put into it. I'm truly sorry that the updates constantly require you to redo and fix so much. It must be exhausting, and I hope you know how much we appreciate the care you take in making these tools accessible to everyone.

PS : Sorry for bad english <.<

@Khiran That's nice to hear! Thank you!

@darksidewalker I want to thank you too!! From an idiot like me your tools are the only ones i can use. Thanks for all you do!!

@iTzTeEZzY1 I hope you can use them to make fantastic things!

I can assure you that it is working with latest comfy (installed with my tool) and workflow, installed it like 10 times this way. If not there must be another problem.

@darksidewalker 

"DaSiWa WAN 2.2 i2v FastFidelity C-AiO-66.json" is it working properly with latest Comfyui version?

Or continue stick with ComfyUI 0.16.4 & ComfyUI_frontend v1.39.19?

@mpayne1985106 updated both workflow to be compatible with the latest comfyui by rebuilding them almost from scratch with new nodes.

@darksidewalker Thanks so much for the fix. Any issue, I will let you know.

@darksidewalker Test 4 different prompt & image, so far no error or failed generate video. Good to go.

Comfyui: v0.18.1

Comfui_frontend: v1.42.8

@mpayne1985106 ComfyUI: v0.18.1, ComfyUI_frontend: v1.42 or v1.43 — it sometimes works with the v6.6 workflow, but after I close it or import another workflow and then switch back, it stops working. I haven’t made any changes; whether the same workflow works or not is completely random, and most of the time it doesn’t work.

@CAT_NYA maybe you are using incompatible custom nodes? Would be possible.

It may because the bug of comfyui version, the connection line of input slots on subgraphs is lost. Especially these nodes : context switch [rgthree] , their input and output connection lines are all lost.

I reconnect these slots of these nodes, and now everything is running normally.

@zhendao v6.6 shouldn't have this problem

@CAT_NYA So far I don't have issue even with 1.43.2

UIpdate: Also don't have issue with 1.43.3.

@darksidewalker I suggest add version number inside workflow top title in the next version. (Ex: Dasiwa - Wan 2.2 FastFidelity - C-AIO v6.6). This will help when load .json or drag video into workflow, easy to reduce issue which workflow version use.

You're welcome

also comfy wont install rtx thing

@user78ar406 So make a bug report on the rtx custom node repository?

@user78ar406 Install the missing custom notes in Comfyui Manager. Once done install, kill/closed comfyui app, then navigate to Comfyui folder (Example: C:\ComfyUI\ComfyUI-Easy-Install\ComfyUI\). On 'File Explorer' address bar, clear 'C:\ComfyUI\ComfyUI-Easy-Install\ComfyUI\' address then type 'cmd' & then enter. the command prompt will appear. Then paste this 'python_embeded\python.exe -m pip install ComfyUI\custom_nodes\comfyui_nvidia_rtx_nodes\requirements.txt' then press enter. Let it finish install, closed & run again comfyui.

@mpayne1985106 thanks !

This bug should not happen on comfyui 0.18.x what version you are using?

@darksidewalker Maybe since C-AiO-4.2, since you use the node <Context Switch>.

After several versions of Comfyui, I have encountered the issue of lost connections.

even comfyui 0.17 & C-AiO-6.4，the problem persists...

Due to some subgraph bugs, I did not upgrade to the latest 0.18.

I have roll back to 0.16.4.

@darksidewalker I replace the subgraph <P09# Extract Last Frame> with this node <Get First & Last Frame (CRT)>, it can directly get the last frame from the output image_FINAL of the subgraph <Backend>, and more quickly. And the advantage is that there is no image quality loss caused by video encoding compression!

@zhendao v6.4 is not compatible with newer comfy 0.17.+ you need v6.6

Check if the files are created, maybe just browser cache bug. Clear the browser cache.

@darksidewalker The files are created fine. I also tried the gguf q8 v9 models for the workflow. But unfortunately, I am only getting black videos.

@adarshrao black videos always are a problem with comfyui. Maybe your comfy is broken.

@darksidewalker  I reinstalled ComfyUI like normal and installed portable version instead of desktop version, and it worked perfectly fine on launch. I don't know what went wrong before. Thank you.

Not with my workflow atm

@darksidewalker How does the Video Combine function work then and what is it for?

@kontraktnik85706 This is describes in detail inside the workflow notes.

Doing this fixed getting the RTX node for me:

O:\ComfyUI-Easy-Install\ComfyUI-Easy-Install\python_embeded\python.exe -m pip install --upgrade pip setuptools wheel build

O:\ComfyUI-Easy-Install\ComfyUI-Easy-Install\python_embeded\python.exe -m pip install wheel-stub

O:\ComfyUI-Easy-Install\ComfyUI-Easy-Install\python_embeded\python.exe -m pip install -U --no-build-isolation nvidia-vfx --index-url https://pypi.nvidia.com/

But I'm also not able to use the video combine feature, can't find any nodes to put the videos in.

Update the workflow, comfyui latest updates break the json of old workflows

ty sir

C-AiO v6.7 and C-SVI v2.3 fixed this

Make sure to read the notes, you need like 24 fps for mmaudio

@darksidewalker thank you for your reply.

Found out it was the interpolation causing this issue.

Maybe its a idea to put the interpolation differently in the workflow. Strangely I didn't have this issue when I edited one of your older workflows where I added sound nodes in (it worked for me but I'm not 100% sure why it worked back then, this one also worked on 16 fps).

Anyway 24 fps is good enough for me, thanks for this amazing workflow!

It is a copy paste type. No seconds setting in SVI, it is a static calculation because of the requirements.

Best would be to follow the git link from the description/notes and make a correct install. If in doubt how to do it you could use my installer.

@darksidewalker  after my comment it seems to went away

Ok for those who are wondering and beginning with these models, I solve the problem by using the "TastySin" Model instead of the "SynthSeduction", it seems it's is solved in the new "Boundbite" model too tough :)

@Mortyh yeah should be. SynthSeduction has somehow camera drift with SVI

Manual download everything like stated on their githup repo, would be the only other solution.
Or you repair the issue that you can not connect.

You have 10 resolutions to choose from, from fast to ultra quality. "Preview" renders in seconds.
The pixel-aspect is automatically calculated, maybe you did miss the resolution switch.

@darksidewalker I don't know why, but in your process, even in the lowest selection, generation takes twice as long.

@darksidewalker If I set it to 368x480, it's 180 seconds. Your minimum is 350 seconds.

@Renessance If I set it on preview the video renders in like 30-40s. I do not know what you are setting there.

@darksidewalker https://skrinshoter.ru/sakRoP2fkhx

5060ti 16gb + 32gb

@Renessance I see, but I have no clue why this happens for you. Maybe 32GB are not enough for zero offload. Did you try q6 or q5?

It is always on

me too ... did you figure out a fix?

@uncannyvfx95 No, I just went back to 3.1 lol

Patience and setting adjustments or manual video editor :)

If the video duration is set to 5 seconds but the final output becomes 10 seconds, I suspect the frame_rate in Video Combine is being forcibly changed to 8, which causes the previously set 16 fps to be overridden.

Previously, the frame_rate field in Video Combine was grayed out, which meant it had no effect. Now I suspect it has been forcibly enabled.

In 'Video Combine' node setting, try use 'video/h264-mp4' before run the workflow.

@mpayne1985106 Thank you so much for your answer

@mpayne1985106 I changed both Video Combine nodes to 'video/h264-mp4', but still the same

@TohnoAkiha did you manage to fix the issue by just changing video format?

@SameImage8905168 I switched back to version 31.

@TohnoAkiha version 32 works for me, thanks!

I use 0.18.2 atm

@darksidewalker Thank you so match! I'll try same one.

I feel comfy is breaking things from day to day atm

@darksidewalker That's it. I have same feeling.. and version number increase is so rapid. It's awful.

BTW, you have extracted sub-subgraph to handle comfy's f*cking bugs. Thank you so match again!

Some settings of the node may create intermediate data and this node prunes them.

I did like 20+ videos today and got not one broken result ... so what exactly is weird?

@darksidewalker What I mean by “weird” is that v7.0 seems to ignore or underweight parts of the prompt. In several of my tests, key details that were consistently respected in v6.9 (same prompt, settings, and model) were either missing or only partially reflected in the output. It’s not that the results are broken—they just feel less aligned with the full prompt compared to v6.9.

@mpayne1985106 That change is not that big that it could be influence prompt adherence in any way I can imagine. I have no clue and all my prompts got followed fine 🤷But I'll test anyway.

I hope you also updated your comfy properly. A model can not make ooom, it's always the app.

@darksidewalker Yes, before this, I was using an old version with an old frontend so that 6.4 would work, then you updated the workflow so it would work on new versions. So I updated Comfy to 0.18.2 for 7.0

@darksidewalker I'll try the GGUF version of the model, if it doesn't work I'll roll back to Comfy version.

Well... No outofmemory with gguf q8

@MrDxrk gguf does stronger offload to RAM and SSD, maybe that's why. Newer comfy also pins memory, what will give you I stand oom if insufficient. Maybe you have just not enough memory for fp8 in general.

@darksidewalker Well, yes, I don't have the most optimal characteristics, but overall it worked, and safetensor did too. I think rolling back the Comfy will fix this. But is it worth it for Safetensor? There seems to be almost no difference between the GGUF Q8 and Safetensor versions of the models.

@MrDxrk safetensors is more memory efficient even more when loras are used.
But the pure checkpoint fp8 and q8 are really close.
https://civitai.com/articles/27174/comparison-safetensors-fp8-vs-gguf-q8-q2-vs-nvfp4

Yeah I use dummy files or I would run out of space. I just make a text file and make the ending gguf or whatever.

I can not change the nodes behavior, there is no option for "none"...

The workflow works with every file.

@darksidewalker I think I discovered a bug that is related to ComfyUI and some node update etc. When looking at the main Backend / Checkpoint setting where the 3 "output/video/" fields are they are invisible. They show nothing there but if you click in that area then it shows the folder location for the merge videos and you can type in the location. I know it is not v7.0 workflow because now it is happening with the older 6.0/6.1 workflows. So I need to investigate further when this problem began. Need to try the older ComfyUI when I saw it last working. It might be from updating Comfy. Look at this picture where the green arrows show. It shows nothing! Have you seen this problem? (invisible fields)

https://ibb.co/pjBqh8B0

@essence25 I know this problem, it is a comfyui frontend bug. Comfyui has to fix this.

@darksidewalker Great! Has this been reported to Comfy or VideoHelperSuite creators?
Are they aware of this or should I create a bug report? (ComfyUI version 0.16.3 is not affected)

Somwhere in between the updates this issue has been introduced.

I tested the other nodes from VideoHelperSuite and only the "Load Video FFmpeg (path)" node is affected. Can we swap that node with something else in the mean time? if yes which one would be compatible or we are strictly requiring the (path) node?

@darksidewalker I also discovered a different bug that is not related to Comfy or your workflow. If using Chrome/Edge/UngoogledChrome, and going to full screen there can be stuttering (audio cuts out when preview nodes or listening to music, usb audio etc)
That is related to Windows 11 MPO / Multiplane Overlay. It is a system wide high DPC interrupt problem that affects AMD. Not sure if Intel is affected since I don't have Intel platform to test.
I just wanted to give you heads up in case you ever encounter this. MPO is also responsible for many people when they put PC to sleep then wake it up only to find that the GPU PCIE links at lower gen speeds eg. 1.1, 2.0, 3.0 instead of their max rated GEN speed like 5.0 for 5090 or 4.0 for 40XX series. (this can happen at boot also randomly, GPU-Z can show the PCIE gen link speeds)

To fix this download MPOGPUFIX v6.6 tool and flip the MPO Fix toggle switch to enable it. Reboot and if I'm correct, you will never encounter stuttering when in full screen or when moving through tabs etc. This took a very long time for me to discover and pinpoint and I have posted in all forums the cause and solution.

IF you want to test for this issue:
Using Edge/Chrome/UngoogledChrome, open yandex, and do a random search then hit F11 for full screen, then move/hover the mouse pointer over all the results in circular motion touching the top of the screen on every rotation, keep doing it over and over. Have Latency Monitor tool open and check it regularly. You will notice super high DPC latency, see the attached pic. What happens the whole system stutters. (audio will stutter if you have something playing, usb sound devices will also stutter since this is system wide freeze)

Are you missing a node pack? Or using ComfyUI (which is nearly broken right now)?

@rkyha639 I'm using ComfyUI, but this error appeared right away-that is, everything was fine in the first generation, but as soon as I started the second, it gave me this error specifically with the model created by the author of this workflow, but when I use other models, this error doesn't occur. I read online that it's caused by "kitchen," but reinstalling it didn't help. In the end, I reinstalled Triton, which was somehow linked to this issue, but that didn't help either.

@rol4ik007119 Loading a model cannot change a workflow. Just saying. So there cannot be a coincidence with whatever model is used.

@darksidewalker I fixed it by installing a different bootloader, and I managed to fix Kitchen somehow, but everything's fine with v70 now, thx for your works

I just forgot to set to 5s before export, just do what fits for you :)

The resolutions are optimized or WAN 2.2 and the VAE latent processing, ensuring optimal pixel distribution and count with as less as possible distortion and latent tiling.
The latent should be Div32 and lean towards supported resolution estimations fitting inside the VAE grid. 👍

C-AiO is not for SVI, there is a separate workflow. C-SVI.

@darksidewalker sory, i meant to say v33 for c-svi

Thank you!
No secret! - I'll tell you!
It is all of them. I tinker with the workflows and nodes till I get results, I read up on many nodes and techniques and look into others workflows, there is every time a thing to learn. Other people other ideas and solving. Sometimes I got help from community members and refined their ideas. And with time there comes experience. My first workflow was not like that 😁

its buggy in comfiu maybe to safe vram?! u can right click and pause and then right click resume then it should play it.last few updates it was enough to click sync videos but it doesent work anymore.

Maybe you used interpolation, there was a bug, fixed in v3.4

@darksidewalker yeah I used it. Got this error "Unknown pack (1) DaSiWa_ResolutionScaleCalculator" on v3.4

@SameImage8905168 Yeah well, you have to install the custom node. Link is in the notes.

So switching back and forth results in 1-frame videos?

Fixed in v3.4

Standard is overall better. The other can introduce errors and unwanted motions. Or just do nothing, like placebo.

try https://huggingface.co/zootkitty/nsfw_wan_umt5-xxl_bf16_fixed/tree/main

Do you have Disccord? Can I text you there and send screenshots of what I see?

@RussianMan_BRUTAL download gguf q8 high and low and put them both in the unet folder.restart comfiui or press R to reload the page and then klick unet name then pick high and low models.u have to that with gguf high/low or safetensor high/low, Svi pro if u are using that workflow high/low,Clip(umt5xxl),vae(wan2.1vae) klick everything and replace them.

@hartweizen Do you have a discord? It would be easier for me if I sent you a screenshot and you told me what to click, I don't understand

@hartweizen My discord bunnyr1ven, I would be very grateful.

@RussianMan_BRUTAL he has a discord channel join there and ask there.its easier if u download all files on the left side.umt5 put that in the text encoders folder,model(download the q8 high and low version 15gb each) and put them in the unet folder,vae into vae folder,svi 2 pro(if u are using svi workflow) put them into the lora folder.u have to replace all files thats written on the worklow by pressing R to refresh then u will see each downloaded file u can choose.

Most likely your comyui deleted some internal wires between nodes.
Redownload the workflow, delete the old one, delete browser cache, make sure your comfyui is up2date, reinsert the downloaded wf.

@darksidewalker I cleared the browser cache and even switched browsers, but it still won't run, with the same error. My ComfyUI version is 0.18.3, and all plugins are updated to the latest versions.The test workflow is functioning properly.

@darksidewalker The workflow still has this issue in version 7.1. I don't know what's causing it, and ComfyUI has also been updated to the latest version.

@TTeslaThunder I cannot tell, for me it works, I did like 4 videos today

@darksidewalker Alright, thanks. I'll try other workflows.

Thank you!
I'll look into the errors, but the reason why there is a custom seed for every node is, that this is to control only the segment that one would resample instead of all. With adding math like yours, it is not possible to change just one segment.

The next version fixed the preview issue and 1-frame after interpolate.

@darksidewalker Amazing! I'm going to check it out!

I did not notice that. But I'll have a look ☺️
Can you specify the node in question?

My negative (text widget) needs 0.6s. Only the very first run after start needed some seconds, maybe a caching thing.

@darksidewalker I'm talking about a native "CLIP Text Encode (Prompt)" node, that serves as a negative prompt and is connected to "WanFirstLastFrameToVideo" 's negative input (or similar node). Whenever you change loras (or lora's strength), or input images, it takes 5-10 seconds for umt5_xxL_fp8_e4m3fn_scaled.safetensors CLIP to work with CLIP Text Encode (Prompt) node (negative prompt).

So when I constantly tinkering with loras, it always takes that additional 5-10 seconds, but it's useless because I have cfg=1

@rozenkristall I see, but unfortunately there is nothing I can change on the behavior of the comfyui nodes.
If this is a problem or you think it is a bug, you may open an issue on comfyui github.
I cannot change the workflow to always zero the negative encode since it would not be possible to use non distilled checkpoints with higher cfg or NAG.

@darksidewalker Its not a bug, it works as intended, it's just not optimal since CLIP encoder and nodes work independently. And I almost told the solution I see:

in your workflow we can use the switch where

if cfg=1 then have negative prompt muted and use a positive prompt node connected to "ConditioningZeroOut" (important) and then connected to negative input of some nodes (like WanFirstLastFrameToVideo)

if cfg>1 then have negative prompt unmuted and have it connected to negative input of some nodes (like WanFirstLastFrameToVideo) directly

I hope you got what I meant 🙏🏻

Link is inside the workflow notes, as always

here: https://github.com/darksidewalker/ComfyUI-DaSiWa-Nodes

I totally missed that! Thanks!

I'm stuck in this step too. Fairly new with ComfyUI and not sure I understand what "Activate your venv inside your comfyui folder" means in practice, let alone steps 2 and 3. Any chance for someone to provide more hints on how to get this done? Thanks.

If you have ComfyUI Manager search for DaSiWa-Nodes and install @linsuxx333 

@ebinloop835 That was easy. Thanks, got it working.

@ebinloop835 yes works since yesterday, published it into the manager repo😊

Thank you! Glad it helped :)

Well, put your prompt from 4 to 3 - deactivate 4 (and further) and you are done.

@vicaut  I know but it's not just the prompt, it's also like the 5~7 loras I used and their respective weights.... And if I'm aiming always the same "ending" then it's nice to keep the last just there and play with the inbetweens

I'm having this issue as well, and not just with this workflow. I use runpod, and a few other workflows want outputs in different folders, and they just don't go.

The path maybe not valid on your system. Just change the path.

@darksidewalker Exception: video is not a valid path: /workspace/ComfyUI/output/

Every time I try to combine. Every video is in that folder. What am I missing?

The fastest and most efficient at least

It is for I2V, S2V will produce it's own voices or references

ahh i see now. Thanks again for all of the hard work; you're stuff really is top notch.

I too am getting this! Strange flash at the end of the FFLF

Same. Did you ever discover a solution?

No idea

Ps: Actually added that myself and it's crazy good

how to use it bro?

Install the kijai custom node pack

@darksidewalker  thx for help. When i try using color Match now i get errors running out of vram (5070 12gb) any work around or is this option just for high vram users?

@christophbueltmann908 color match need much vram, depending on the amount of frames.

Flf2v is activated properly?

@usasan The FLF2V switch (type switch)

I apologize for asking so many times.

Is it correct that there is a toggle switch for the FLF2V?

The [type switch] item was set to [Enable T01# I2V] [Enable T01# I2V].

Even if I switch one of them to [yes/no], it seems to affect both [Enable T01# I2V] and [Enable T02# FLF2V].

@usasan Only one can be active

I looked at the HowTo text guide, but it seems the display name has also changed.

It's probably something wrong with my settings.

Thank you for letting me know.

Make sure high and low are correctly setup

Hi , thanks for the reply, I was wondering whether it might be the VAE but I am using the exact one setup in the workflow, I am also using your High and Low checkpoints,if I use a single image the blurring is very bad with the I2V setting and if I use a single image for first and last fame setting the video isn't too bad, I have added a lightning lora to the high and low channels which improves the final video, but motion and character movement is still not good.

@Jezz You are using ether the wrong checkpoints or settings for the checkpoint you are using

@darksidewalker hi thanks I will check the workflow settings and the checkpoints again

Hi, I have it working now thanks I had the wrong checkpoints , thanks, I am also using your LTX2.2 workflow it's my favourite workflow at the moment, keep up the good work, thank you 💯.

Whiteout knowing your settings and version of the workflow there is no way to know if this is a bug or problem on your side.

@darksidewalker Thanks for replying so fast! I'm on the following:

workflow:

DaSiWa WAN 2.2 I2V FastFidelity C-AiO v72

Platform: Windows

** Python version: 3.10.6
Total VRAM 8188 MB, total RAM 32681 MB

pytorch version: 2.11.0+cu130

Set vram state to: LOW_VRAM

Device: cuda:0 NVIDIA GeForce RTX 4060 : cudaMallocAsync

ComfyUI version: 0.18.5

comfy-aimdo version: 0.2.12

comfy-kitchen version: 0.2.8

ComfyUI frontend version: 1.41.21

I found a conflict with another node called ComfyUI-GGUF-FantasyTalking. Going to try fixing that first and see if the error persists

@darksidewalker it was my bad! Apparently I installed this other mode from another workflow and didn't see the conflict. It run its own version of wan.model.wanmodel and this version doesn't treat all the variables correct.

Anyways bro, thanks for the beautiful work!

everything worked fine, but for a few days didnt manage to run this workflow because of same issue. couldnt update said node. seems incompatible. last update long ago. how to find and change to some alternative?

You both installed the wrong node, use the one listed inside my workflow notes.

I'm kind of new at this, so I'm not sure what you mean by your workflow nodes, so how do I use it? Also, as far as installing custom nodes go, I've tried both from the extensions menu, as well as the install all option in the right hand sidebar underneath the errors tab, and both seems to lead to the same outcome.

@Ande85 the manager seem to try to install the wrong, I did links to all the nodes inside the workflow notes and on the DL page.

@darksidewalker Looked at the notes section to the Left, downloaded everything, unzipped them, and to my surprise it worked! Got the Rife 47/49 installed, and now I'm dealing with 14 error messages and two missing sample images that used to be there. I'm making progress...just slowly. I'll take a look and see what I can do on my own first before asking for more help...I try to limit things to no more than one question per week to ensure self-reliance. Anyway, thanks for the help, I'll see if redownloading the sample scripts will get the sample images back or not.

Wouldn't recommend that since no WAN model is made for that, but there are 2 ways.

1# set to 30 fps and add interpolation

2# Set the fps to 60 and render directly on 60 fps

I did not include an option to render from 16 or 24 fps to 60 since that introduce quality loss

Comfy I frontend bug

try to change frame rate in the Video Combine node

Reading is key here. 😆 Wrong repo.
Also why is this a problem of the workflow if you have problems installing a repo?

just use MMAudio, delete Foley nodes))

@darksidewalker how is it wrong if it's the error when I load workflow? Trying to install missing repo, which prevents me from using it. Just asked if there is alternative. Okay, thanks for MMaudio. Will try

@quazaque As I said, wrong repo. I have links in the workflow and page description, you tried to install the wrong one.

@darksidewalker sorry for misunderstanding, don't get me wrong. i didnt try to install anything. this workflow is great. it worked like a charm previously. but then something happened and i had to remove desktop comfyui and use portable just to get the same error trying this workflow. incompatible repo node... that i cant update or install or whatever.
thank you for your share. i hope i will understand how to change some nodes (dependencies?)

Comfyui has a bug, the combination fields in the frontend are not showing properly.

If you OOM your settings are too high for your rig.

@darksidewalker 

I figured out how to select a video, but it works very strangely, it doesn’t allow me to generate 3 videos, it says OOM with a huge number of gigabytes. And two videos, it renders strangely, it does not render a single video (frame gap is visible)

@Flwsss the combination of videos is not for creating, just combine