v1.1 update:
Small update, adding Prompt Stash Saver and Prompt Stash Manager. Also fixed default resolution. Other minor fixes/additions, see About This Version for the rest.
I've also incuded the .json data for my prompt stash saver. This is a node that saves all of your prompts, which is an upgrade from copy/pasting them from the Notes in the workflow.
If you want to try and use my saved prompts, the file is uploaded here and goes in ComfyUI\custom_nodes\confyui-node-prompt-stash-saver
I don't even know if you can just put this file there or not, give it a try and let me know.
In the workflow I've been using it to store the prompts and I retrieve them and copy/paste but you can hook the node up directly and use it as input.
Edit:
Some of the LoRas that are present in this workflow (but aren't used) are no longer available on the site. Just an FYI for folks who are wondering https://civarchive.com/articles/13749
Edit 2:
I found out the default resolutions are a bit different than I originally thought, so the workflow is currently set for unoptimal resolutions (720x720 & 480x480)
14B Model I2V 720P - 1280x720 720x1280 768x76814B Model I2V 480P - 480x832 832x480 512x512The next version of this workflow will add some requested features as well as fix the resolution. For now you can change the dims in the WANImageToVideo node.
This is the workflow I use, I spent some time cleaning it up - keeping in mind other people would be using it. There are lots of notes with LoRa trigger words and prompts, and the current generation setup is a nice one ;)
Current prompt and loras setup to create a "dancing bear" type scenario where 2 gals are in the photo and along comes some guy that skeets on them. Then they kiss. Crazy how well it works. This is obviously extremely goonish so beware...
Many LoRas are referenced in the notes, but the two that are setup in the workflow are:
https://civarchive.com/models/1333275/passionate-kissing
https://civarchive.com/models/1350447/wan-cumshot?modelVersionId=1602715
Just grabbed a random AI generated pic of 2 chicks from Google and ran the prompt that's embedded in the workflow. This is the first generation that came out. You can expect to see a lot better.
720x720 in 153 secs on 4090.
The WAN Cumshot v1.1 LoRa is really good, it's set to .7 and that tends to keep char consistency well enough. If you give it a good pic you will get good results. This image is not a good pic but showcases the concept.
FROM WAN CUMSHOT:
To use with an image where no guy is present, you can either use
cum shoots from off screenor
A man enters the frame from the side. If her hand(s) off screen you can try
She is stroking his penis with {her hand | both hands}You can try:
"A woman / two women are sitting in front of a man / next to a man even when he's not pictured to cause him to show up.
Description
New version of workflow from previous post. Cleanup and add some QoL, also preconfigured to generate facial + kiss
FAQ
Comments (31)
I'm testing out the new Amorous Lesbian Kissing LoRa, will update if the results are better than Passionate Kissing, which is pretty damn good
it is just as good in general, seems to give better tongue interaction
great workflow.
i noticed when selecting a T2V and using a white blank image u can use t2v
oh cool, good to know
Thats interesting. That would imply that you could leave part of an image black and inpaint over it with a t2v model. Could it be that easy?
You should checkout Prompt Stash Manager. This will let you save all of the prompts into a drop-down list per custom category.
Thank you, I was hoping someone would comment a suggestion like this. I played around with some other options and didn't like them. I'll check that out
@Kiefstorm It'll definitely help!
I've been using it a lot. I doubt the prompts in the manager get shared with the workflow, but the nodes will be in the next version for sure
Great workflow. And it's good to be able to actually get a workflow that's not outright broken and requires me to spent between 10 hours and 3 days trying to get it to work because comfy ui has missing components that are either outdated or just broken. I am wondering through how possible it would be in a future update for you to be able to allow videos to be combined by allowing a video-to-video functionality using the last frame? Because right now the length of the videos is one of the major limiting factors. Thanks again.
Thanks for the feedback. To answer your question, yes thats the plan. I have a work in progress portion of the workflow that will be accompilshing that. Right now there are 2 nodes that allow saving the last frame, and then that frame can be input as the starting image. I just haven't gotten around to testing it, but I left it in the workflow since it should work.
But currently I use another workflow for video extension, search for video extension workflow you should find it, if not I can link it for you
@Kiefstorm Thanks for the reply. and yeah, i know there are other workflows that allow this. However so many of them lead to dead ends because of some thing i don't have installed that may or may not work.
at least your workflow allows me to turn off sage attention until i have long enough to be able to successfully install it (8 hours, 2 attempts so far), other workflows often involve components that comfy ui has already integrated (making parts redundant or incompatible) or certain plugins from manager require older or newer versions of certain files which takes an incredibly long time to track down. The fact that the rest of your workflow works for me, means that having everything centralized really helps with the sanity levels :D
really appreciate the quality of your output as well, the other workflow i had would upscale a 480 output, but these outputs seem quite a bit better even without the upscaling/rife component.although i do miss that extra second of footage, since wan manages to remain coherent up to around 3s, and right now yours is about 2s. i realize that there is frame increase options, but they go over the 3s mark. still figuring out how to get it to be custom
@whitespider9999 I know what you mean, ran into the same stuff myself on other workflows. As for increasing the length, if I understand what you want correctly - you can change the node that has the height/width/length value for length to be input instead of widget.
So right click the node, and choose change input to widget > select length. That will allow you to manually enter any value you want. Lmk if you need more help with that.
So right now in the workflow you can choose 33/65/97/129 or 2/4/6/8 secs. if you switch to manual input you can input 17 if you want 1 sec. I haven't ever tried to do anything other than even divisibles of 16, so idk if something like 41 works.
But I should mention I do not have issues with coherence or otherwise even w/ 8 sec videos, although it does cause my PC to trip the circuit breaker sometimes... :(
@Kiefstorm thanks for the advice. I'll look into it. And yes i know what you mean about the extra seconds becoming extremely taxing. I have a 4090 and as soon as i start increasing the length i get a lot of slowdown in the app, i presume a lot of data is being moved from vram into ram and it's spilling over the 24gb.
As I understand it, some of the models used in workflow are no longer available on the site. Can you tell me where you can find them?
Which ones specifically are you interested in? There are a bunch of LoRas shown because this is my personal workflow and they've accumulated over time. The only two you need to create something like the sample vid I've linked in the description.
@Kiefstorm All loras with undressing, for example rip-shura
P.S. Many thanks for the workflow and comments in it.
@tltphone63587 wow, you're right. I'm seeing the top ones missing too...I'm not sure what that's about. Seems unlikley that the authors would all pull them at the same time today (I saw some as of yesterday available). Take a look here, I've uploaded a few. The t***drop is really good.
@tltphone63587 wow, just saw this post by the Author explaining what happened. disappointing https://civitai.com/articles/13749
@Kiefstorm Respect to you!
huggingface
thanks for amazing workflow. it will be amazing if
You add frame interpolation and upscaling ..
how to verify that sage attention is working?
I mean you can set it to disabled and see if anything changes
I wish this workflow worked.
Not sure what the issue is since you didn't post the problem but my guess is you don't have SAGE or Triton installed. If you don't then install or disable it in the workflow
@Kiefstorm I ended up tinkering for several hours and got it working after some research and fighting with Python. Cheers.
this works great and I've learned a lot from it.
you working on an update? looking forward to it!
Yes, will be publishing soon
https://civitai.com/models/1480535 small update pushed, Prompt Stash Manager is great. Let me know if there's any other stuff you'd like to see in next version
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.