*

punchy

  • Generator of AI Slop
  • 2218
Re: PunchAI - Punchy's Realm of Realistic AI Art
« Reply #2340 on: February 27, 2026, 12:51:49 PM »
Just a note, I think it is LTX-2 that was running up my electric bill.  Unsure what the heck that model does, but yes, it is magic, and yes, it is super resource intensive.  If you guys want me to post my setup I can see what I can do, but, again, it requires a GeForce video card with 12 GB or more, 3000 series RTX or higher, and at least 32 GB system RAM.  If you meet those requirements and want my setup I will open a new Mega (old one got closed when the hacking incident occurred), I will upload the installer for Easy ComfyUI, and provide a document with links to the models you need and what folders you need to put them in.

This is a picture of the required disk space.  Note, you need additional disk space for output.  These models all save either under the output folder or the "temp" folder in the ComfyUI folder under each.  Phr00t is the biggest because it includes both the fp8 version of LTX-2 and Phr00t's AIO of Qwen 2.5.  That one folder almost lets you generate a static image, modify it so you have both the begining and end frames, then use Phr00t's LTX-2 model to animate them.  This is not for the technical novice.  You have to really dive in to this and make yourself very familiar with ComfyUI and kijai, learn what a GGUF is and how to modify workflows to work with them (different loader), know how to troubleshoot when a custom node is missing in ComfyUI (use ComfyUI Manager to make it easier, still requires some technical expertise), etc.  I will not hand hold anyone through any process outside of what I would post in the document.  I am too busy for that.

From the attached screenshot snippet:
ComfyUI_WAN - This is a very simple Smooth Mix WAN GGUF implementation using the Q4K models.  On my machine it can generate a video in about 6 minutes.  It has switches to change from I2V to T2V and even IS2V (add a sound file with the image and generate a video combining both).  When sticking to I2V and T2V this is one of the easier models to use, BUT!!!  WAN is best at I2V.  It has good adherence to unrealistic body proportions.  T2V SUCKS for unrealistic body proportions, mainly because the LoRAs you have to use to blow up boobs all give the breasts a very "samey" look, and sometimes the areolas look really strange. Summary: Best I2V for unrealistic body proportions, tons of LoRAs for sexual positions, LoRAs for breast physics.

ComfyUI-Chroma - This is a fantastic static image generator for extremely exaggerated massive hyper breasts, and that is the term you put in the prompt.  It makes women with car sized boobs if you want.  The negative?  There is almost always an anatomical problem.  Messed up hands, screwed up faces, extra arms, another head coming out of their back, something is always off.  BUT!  I would say you get a hit rate of about 33%.  So about 1/3 of the time you get a usable pic you can then animate in something like WAN or LTX-2.  It isn't the fastest, that would be Pony.  I will make another post about that next. FYI, Chroma is what Perchance.org uses for it's static image generator, so if you are already using Perchance, then you are already using Chroma and know what I am talking about. Summary: Best static AI image generator for unrealistic body proportions, as long as you don't mind throwing out 2/3 of the results due to unreparable anomalies.

ComfyUI-LTX-2 - This is my default implementation of LTX-2, no LoRAs, no shenanigans.  This one is what I use just to make silly videos for my family and friends when I want something longer than what Grok Imagine can create. It does NOT do NSFW.  If you try to make a nude person you get a barbie doll.  No nipples, no groin.  Just smooth. It can go up to 20 seconds in length, but at the last second it can get a little crazy.  I recommend 10-15 second videos. Summary: Use default LTX-2 if you want to make insane Hollywood quality videos, cartoons, and even short music videos, but no NSFW.

ComfyUI-Phr00t - This is my go-to for a lot.  It is Qwen-Image-Edit-2511 for static image generation and image modification, plus Phr00t's merge of LTX-2.  Merge, meaning it has many popular LoRAs already baked in (but not the BE LoRA).  There is a SFW and a NSFW version both there.  I had to do that due to how the NSFW version tends to make women bounce like they are riding a man.  For just making women with large busts walk normally or do everyday tasks I have to use the SFW model. However, if you use LTX-2 in T2V mode it gives women really odd looking nipples that sometimes become extra hands attached to the sides of their boobs.  Best for I2V where you don't plan on having the women move too much. When you use SFW version with women it tends to make them sound like Karens (WHERE IS YOUR MANAGER). The NSFW model makes women sound like they are in the throes of passion (sexy!).  Qwen is great for trying to fix those Chroma images with extra limbs, etc., and it does so with regular language.  It does have a tendency to zoom the image in, and there is a fix on Phr00t's huggingface for it, but I couldn't figure out what to do.  I just didn't care because I just add to the prompt to make sure it doesn't zoom the photo.  That works 80% of the time. You can also generate images using Qwen, but there is an upper limit on how big it makes boobs without a LoRA, and even the LoRA only gives you another 30-40% size boost.  Hyper breasts must be generated in Chroma or Pony. Summary: Phr00t's Qwen is great for editing those photos of hyper breasted women that have extra parts or need tweaks like wardrobe changes, changing their expressions, posing them, changing the environment they are in, etc.  Phr00t's LTX-2 is great for making sex scenes with exceedingly sexy audio.  And doggy style POV scenes are SO realistic!

Next I will go over Automatic1111 and Pony, which I have used for over a year to make static images and does not seem to use a lot of resources.  It can probably run on much smaller video cards.

Punchy

*

punchy

  • Generator of AI Slop
  • 2218
Re: PunchAI - Punchy's Realm of Realistic AI Art
« Reply #2341 on: February 27, 2026, 01:05:22 PM »
Finally I will cover Ace-Step and Automatic1111.

Ace-Step - This is Ace-Step 1.5, a music generation tool that is pretty darn good.  Trust me, AI music is no substitute for real human music, but if you are doing something like what Vieregg does on DeviantArt you will need a good music generation tool.  I think Sono may still be a better music generation tool, but for those who may be embarrassed by generating songs about women with big boobs or something sexual (because they do get shared with the community), Ace-Step may be an alternative for you.  This is a smaller implementation of Ace-Step, so the quality does suffer.  I have fun playing with it, and I have made some sexy songs with sexy lady vocals.  Pretty darn cool if you want to stretch your lyrical legs.  It can also generate lyrics for you, but using something like ChatGPT or Grok for lyrics is better. Summary: An AI music generation tool that runs local on your machine and does an almost equal job as Sono.  Better lyrical adherence than Sono, but quality of the sound is worse.

Auto1111 - Finally, this is my implementation of Pony, specifically CyberRealistic Pony.  I have a TON of LoRAs in this for bigger boobs, saggier boobs, perkier boobs, "breast expansion", exploding clothes, tight clothes, bursting breasts, "All The Best" (which makes boobs bulge out of clothes nicely, but tends to make butts bigger), penis enlargers, sexual poses, varying clothing items, and more.  The sex images I animated recently were made using this so I could make the women have gigantic boobs and also have the men have extremely fat penises.  Summary: Go to AI static image generator.  Nipples and areolas tend to look similar, not a lot of variety there (LoRAs may fix), but great for generating static images of sexual situations for animation.

Now that I have all that out there...

If you have the hardware and are interested in any of my setups I have listed above, let me know and I will try to upload setups, links, and how-tos to a Mega site where you can give it a go.  Again, not for the feint of heart. You need some mild tech chops to do this. No Mac, phone, or table users, this is 100% for Windows desktop users, maybe Windows laptop users if you have adequate VRAM and system RAM.

Again, let me know.

Punchy

*

punchy

  • Generator of AI Slop
  • 2218
Re: PunchAI - Punchy's Realm of Realistic AI Art
« Reply #2342 on: February 27, 2026, 01:23:34 PM »
If you want some examples of what does what, here is a listing of what I used to generate the last few videos I posted.  No examples of Qwen here, but I think I have some further back where I even commented I generated them using Qwen.

Post 2326 - Image generated with Pony, animated using WAN I2V.
Post 2327 - Image generated with Pony using breast size and penis size LoRAs, animated with LTX-2 NSFW using embeddings connector dev.
Post 2328 - Image generated with Pony using breast size and penis size LoRAs, animated with LTX-2 NSFW using embeddings connector dev.
Post 2329 - Image generated with Pony using breast size and penis size LoRAs, animated with LTX-2 NSFW using embeddings connector dev.
Post 2330 - Image generated with Pony using breast size and penis size LoRAs, animated with LTX-2 NSFW using embeddings connector dev.
Post 2331 - Image generated with Pony using breast size and penis size LoRAs, animated with LTX-2 NSFW using embeddings connector dev.
Post 2332 - Image generated with Chroma, animated with LTX-2 SFW using embeddings connector dev.
Post 2333 - Image generated with Chroma, animated with LTX-2 NSFW using embeddings connector dev.
Post 2334 - Image generated with Pony using ATB LoRA, animated with LTX-2 SFW using embeddings connector dev.
Post 2335 - Image generated with Pony using ATB LoRA, animated with LTX-2 SFW using embeddings connector distilled.
Post 2336 - Image generated with Chroma, animated with LTX-2 SFW using embeddings connector distilled.
Post 2337 - Image generated with Pony, animated with LTX-2 SFW using embeddings connector distilled.
Post 2338 - Image generated with Pony (generic nipples/areolas), animated with LTX-2 SFW using embeddings connector distilled.

Punchy
« Last Edit: February 27, 2026, 04:43:55 PM by punchy »

*

wolpertinger

  • W👅L
  • 444
Re: PunchAI - Punchy's Realm of Realistic AI Art
« Reply #2343 on: February 27, 2026, 05:54:50 PM »
Just a note, I think it is LTX-2 that was running up my electric bill.  Unsure what the heck that model does, but yes, it is magic, and yes, it is super resource intensive.  If you guys want me to post my setup I can see what I can do, but, again, it requires a GeForce video card with 12 GB or more, 3000 series RTX or higher, and at least 32 GB system RAM.  If you meet those requirements and want my setup I will open a new Mega (old one got closed when the hacking incident occurred), I will upload the installer for Easy ComfyUI, and provide a document with links to the models you need and what folders you need to put them in.

Yeah, I'd appreciate, if you could reveal your setup including
- OS,
- technical (electrical) setup of your computer
- and last but not least something like a glossary explaining all those expressions like e.g. ComfyUI_WAN ("Comfy UserInterface"?). Is Comfy something like the LinuX standard GUIs like Gnome or KDesk?

I'm using a 1 kW supply with my machine, so IMHO, max. power consumption can be 1 kWh within 1 hour or 24 kWh per day. Not nothing, but not that "huge" in comparison with e.g. electric heating.
« Last Edit: February 27, 2026, 06:17:44 PM by wolpertinger »
de gustibus non est disputandum

*

punchy

  • Generator of AI Slop
  • 2218
Re: PunchAI - Punchy's Realm of Realistic AI Art
« Reply #2344 on: March 01, 2026, 08:12:46 AM »
Yeah, I'd appreciate, if you could reveal your setup including
- OS,
- technical (electrical) setup of your computer
- and last but not least something like a glossary explaining all those expressions like e.g. ComfyUI_WAN ("Comfy UserInterface"?). Is Comfy something like the LinuX standard GUIs like Gnome or KDesk?

I'm using a 1 kW supply with my machine, so IMHO, max. power consumption can be 1 kWh within 1 hour or 24 kWh per day. Not nothing, but not that "huge" in comparison with e.g. electric heating.


All answers are in the posts, but I will summarize here.
I am running a Windows 10 machine with an i7-8700K with an automatic overclock, 32 GB RAM, an RTX 3080 Ti with 12 GB VRAM, and about 9.3 TB of mixed storage (SSD, HDD).  I have an 800 watt power supply.  And I am not sure what terms you need explained?  ComfyUI is a Python based user interface that can run a variety of AI models.  You create workflows for the AI models, but best to nab one off of CivitAI.com or from Github if you know how to look.  I can't really say much more beyond that.  If I have to get more technical than this, then you may want to think twice about trying this on your home PC.

Which model are you interested in running and I will see about writing up a tutorial with links.

Punchy

Re: PunchAI - Punchy's Realm of Realistic AI Art
« Reply #2345 on: March 01, 2026, 08:28:19 AM »
Thank you for all the info. I am still trying to figure out all the options. I have been using DAZ to create original characters and then using Seeart to animate them, but I would love to be able to use my machine's local power rather than paying for hit or miss results from a remote server farm.

I have a 265k, 5070ti and 32mb ram. What do you recommend for generating NSFW 10 second clips using starting and ending frames that I provide?
amazon.com/author/joryanjstrebas
https://www.patreon.com/jjstrebas
[email protected]
Twitter @jjstrebas

*

wolpertinger

  • W👅L
  • 444
Re: PunchAI - Punchy's Realm of Realistic AI Art
« Reply #2346 on: March 01, 2026, 02:46:16 PM »

All answers are in the posts, but I will summarize here.
I am running a Windows 10 machine with an i7-8700K with an automatic overclock, 32 GB RAM, an RTX 3080 Ti with 12 GB VRAM, and about 9.3 TB of mixed storage (SSD, HDD).  I have an 800 watt power supply.  And I am not sure what terms you need explained?  ComfyUI is a Python based user interface that can run a variety of AI models.  You create workflows for the AI models, but best to nab one off of CivitAI.com or from Github if you know how to look.  I can't really say much more beyond that.  If I have to get more technical than this, then you may want to think twice about trying this on your home PC.

Which model are you interested in running and I will see about writing up a tutorial with links.

Punchy

Sorry, Punchy, that I'm stressing your nerves, but I'm really fascinated by the pics, movies etc. you are generating. And would, at some much lower degree of course, like to generate similar stuff myself.
And please don't be too angry with me, that I was not concentrated enough, to identify answers already given.

Well, what I would like to do is to get an animated version of my avatar. I remember a sequel with a before and after picture of a girl, smashing a glass of beer with her growing boobs.
If I understood well, the metamorphosis from "before" to "after" was performed by "Qwen" and later on converted into a movie between these two scenes by another tool, also included in the "Phr00t" environment ( or whatever Phr00t may be  :o).

So what I would request was a setup of Phr00t which would give me a functional GUI as depicted here (flow-chart).
BUT please no more Windows OSes. The whole stuff should work under LinuX.
de gustibus non est disputandum

*

punchy

  • Generator of AI Slop
  • 2218
Re: PunchAI - Punchy's Realm of Realistic AI Art
« Reply #2347 on: March 02, 2026, 11:54:18 AM »
Oh, they most definitely work under Linux, but my Linux box broke a couple of years ago and I never rebuilt it.  I will be converting my Windows machine over to a Linux box because I am getting tired of the unwanted AI integration and fat overhead of Microslop's operating systems.  I want AI my way, not Copilot, which is crap.

Here, perhaps this is a start for you.

https://huggingface.co/Phr00t

In particular...

https://huggingface.co/Phr00t/LTX2-Rapid-Merges
and
https://huggingface.co/Phr00t/Qwen-Image-Edit-Rapid-AIO

The workflows are in the root of the files section of each.  Look for the JSON files there.

For ComfyUI Easy Install for Linux, here is what you want.

https://github.com/Tavris1/ComfyUI-Easy-Install/tree/MAC-Linux

Anyone using a Windows machine and wanting to try, use this link.

https://github.com/Tavris1/ComfyUI-Easy-Install

Once installed put the workflows (JSON files) in /ComfyUI/user/default/workflows/ folder.  Workflows may not exist until after first run, but you can create the folder manually.

You will need to bypass nodes, download text encoders, VAEs, embeddings connectors, diffusion models (LTX), and checkpoints (QWEN).  Put all of those in the appropriate folders under /comfyui/models/.

Short list.

Text Encoder & embeddings connector (you need both):
https://huggingface.co/Comfy-Org/ltx-2/resolve/main/split_files/text_encoders/gemma_3_12B_it_fp8_scaled.safetensors?download=true
https://huggingface.co/Kijai/LTXV2_comfy/resolve/main/text_encoders/ltx-2-19b-embeddings_connector_distill_bf16.safetensors?download=true
Place in /comfyui/models/text_encoders/. There is also a dev version of the embeddings connector, but it doesn't provide as much natural movement.  People seem stiff.

VAEs (you need both):
https://huggingface.co/Kijai/LTXV2_comfy/resolve/main/VAE/LTX2_audio_vae_bf16.safetensors?download=true
https://huggingface.co/Kijai/LTXV2_comfy/resolve/main/VAE/LTX2_video_vae_bf16.safetensors?download=true
Place in /comfyui/models/VAE

Diffusion Models (One or both, I have both depending on action I am trying to portray):
NSFW -
https://huggingface.co/Phr00t/LTX2-Rapid-Merges/resolve/main/nsfw/ltx2-phr00tmerge-nsfw-v62.safetensors?download=true
SFW -
https://huggingface.co/Phr00t/LTX2-Rapid-Merges/resolve/main/sfw/ltx2-phr00tmerge-sfw-v5.safetensors?download=true
Place in /comfyui/models/diffusion_models/.

Checkpoints (One or both, I have both depending on what I am trying to create):
NSFW -
https://huggingface.co/Phr00t/Qwen-Image-Edit-Rapid-AIO/resolve/main/v23/Qwen-Rapid-AIO-NSFW-v23.safetensors?download=true
SFW -
https://huggingface.co/Phr00t/Qwen-Image-Edit-Rapid-AIO/resolve/main/v23/Qwen-Rapid-AIO-SFW-v23.safetensors?download=true
Place in /comfyui/models/checkpoints/.

Now you should be ready to run. Give it a whirl.  You are going to get errors.  Again, you need to make sure you are using the correct nodes.  A few screen shots to follow with some brief instructions.

Punchy

*

punchy

  • Generator of AI Slop
  • 2218
Re: PunchAI - Punchy's Realm of Realistic AI Art
« Reply #2348 on: March 02, 2026, 11:58:52 AM »
For Qwen, you need to bypass the optional input image nodes.  Both if you are generating a new image from a prompt, only one if you want to modify an image, or neither if you want to try and combine elements from two images.  You can lift people out of each picture and put them together into one image.

If you click on the "node" it will pop up a menu on top. Click the icon I have a box around and that will bypass that node. Click it again to activate the node.

Punchy

*

punchy

  • Generator of AI Slop
  • 2218
Re: PunchAI - Punchy's Realm of Realistic AI Art
« Reply #2349 on: March 02, 2026, 12:08:44 PM »
Qwen: More info.  You may have to click the drop down I highlighted in yellow below to select your QWEN model.  The blue box is where you enter your prompt.  The green box is where you select your height and width of your image.  The orange box is critically important.  The workflow will default to "fixed" for the seed.  You want to click and select "randomize" otherwise unless you change your prompt it will generate the same image over and over.  Randomize will give you different results from the same prompt with each run.

Qwen is limited on how big it will let boobs get, but you can push it pretty hard, at least to where the breasts can reach the belly button, but no bigger.  At least I haven't found a way to make them bigger.  And it uses plain language, so if you want to take an image of a woman and have it remove her clothes, have her wearing a black lace bra and panties, and have her give the middle finger to the camera, just say just that and it will do it.  You can also try to have it increase boobs by saying "Increase her breasts by 500%" or something like that, but again, there is a ceiling where it stops increasing boob size.

Punchy

*

punchy

  • Generator of AI Slop
  • 2218
Re: PunchAI - Punchy's Realm of Realistic AI Art
« Reply #2350 on: March 02, 2026, 12:16:16 PM »
LTX: I have to break this one down into many smaller pictures.  The workflow is too big to fit in a single screenshot and you be able to make anything out.

The red box is showing nodes you may need to bypass.  Bypass both the first image and last image if you want to generate a video from prompt only. Bypass just first or last image if you want a video with guidance on where it should start or where it should end.  Leave both active if you want to feed it both the first image of the video and the final image where you want the video to end.

The purple box is the number of guides. Adjust the number of guides based on if you bypass a node.  If you bypass one, reduce the number of guides to 1.  If you bypass both, bypass the entire LTXVAddGuideMulti node, like how you bypassed the first and last image nodes.

Punchy

*

punchy

  • Generator of AI Slop
  • 2218
Re: PunchAI - Punchy's Realm of Realistic AI Art
« Reply #2351 on: March 02, 2026, 12:26:37 PM »
LTX: More notes

The red box here is where you enter your prompt for action.  You can also add dialogue by saying things like...

The woman caresses her breasts and glances at the camera with a sexy smile and seductively says: " Do you like what you see? "

Purple changes the width and height of the video.  Bigger video requires more VRAM.

Green is the length of the video in frames, plus 1.  I don't know why it is always plus 1, but I always add that extra frame.  The default here is 5 seconds (and one frame) based on 24 frames per second.  Changing the frame rate from 24 can have some odd time dilation side effects, so I just leave it at 24.  I have had success with frame counts up to 481, which generates a 20 second video. 15 is the sweet spot for sexual content with audio.  And remember, whatever dialogue you put in the prompt the AI will try to get the person to say it all in the time alotted.  You will get fast dialogue if you don't have a long enough video.

Blue is the seed generator.  Just like with Qwen you have to set that to randomize if you want to try and generate several videos from one prompt.

Yellow is just where you select your model, either SFW or NSFW.

Punchy

*

punchy

  • Generator of AI Slop
  • 2218
Re: PunchAI - Punchy's Realm of Realistic AI Art
« Reply #2352 on: March 02, 2026, 12:29:41 PM »
LTX: Even more notes

Here, just make sure your text encoder, embeddings connector, and VAE are all selected properly.  Nothing else in this area should need to be touched.

Punchy

*

punchy

  • Generator of AI Slop
  • 2218
Re: PunchAI - Punchy's Realm of Realistic AI Art
« Reply #2353 on: March 02, 2026, 12:34:01 PM »
LTX: Final note

By default all output will go into /comfyui/temp/, and that folder goes away whenever you stop the server.  If you want your output to go into the /comfyui/output/ folder you need to change the highlighted setting.  Output does not get deleted until you clear it yourself.

Now go forth, be fruitful, and have fun!

Punchy

*

punchy

  • Generator of AI Slop
  • 2218
Re: PunchAI - Punchy's Realm of Realistic AI Art
« Reply #2354 on: March 02, 2026, 12:40:47 PM »
Final tidbits:
The NSFW LTX2 model will make women bounce like they are having sex even when there isn't a man in the video.
The NSFW LTX2 model will make women sound like they are in the throes of passion when they talk.  Say "she moans" to make it worse if you want her sounding like she is building to an orgasm.  And if you say "she orgasms hard" it will make her eyes scrunch closed and her mouth open.  It almost looks painful, but it is also sorta sexy.
The SFW LTX2 model will make women sound like a Karen.  They sound demanding and loud.  You can try to prompt out of it, but I didn't have a lot of luck.

ANY LTX2 model, even the NSFW one, will make women have no genitalia and odd or no nipples.  You are better off using a starting image for your videos that includes the woman's nipples and genitals so it doesn't have to guess what they look like.  Otherwise you will need to add a LoRA node to the workflow and add some female anatomy LoRAs.  I haven't had the best of luck with that as I may not be putting the node in the flow in the correct spot or the model does not work well with LoRAs.  Although I did have the BE LoRA working, but the BE LoRA only works on small boobs.  Once the breasts are about the size of a volleyball they don't get any bigger.

I will show you where in my workflow I put the LoRA node.

Punchy