This lora makes a gigantic gaping anus, which you can look into, as if it's a deep dark tunnel.
Trained Triggers:
Tunnel gape: A large round gigantic asshole, with a deep dark depth.
Anal gaping: Refers to the anus itself, without this it sometimes gapes the pussy on females.
Pov hands: A pair of first person hands, representing the hands of the viewer.
Gape pull: The main character's hands, as they spread themself open.
Testing:
To get a deeper detailed hole, use: The hole is endless long, like a deep dark cave.
To get a dark black hole, use: The hole is a solid deep dark tone, devoid of light, a empty void and black chasm. (Or something alike, as this lora hasn't been explicitly trained on this type of language.)
I have tested the lora to a peak strength of +1.2, and haven't had the need to go above that to reach the desired results.
For more details it helps to describe the gape in more accurate, by adding elements such as the color, wrinkles, ridges, and details that refer to the meaty inside part, or outer rim.
the lora works best when the triggers are used in a array, combined with natural language. By incorporating the triggers in a normal sentence has the highest accuracy. Otherwise limbs or hands might get rotated.
For hands on both sides, it's best to use the sentence: Her hands are mirrored on her bottom.
For a single hand, her left hand is on her right buttcheek with the thumb down.
For a single hand, her left hand is on her right buttcheek with the thumb up. (can exagerrate thumb).
It's not necessary to define the position of the hands, but leaving it out decreases the anatomical accuracy of the arms, and can sometimes flip the hand around vertically or horizontally.
Notes:
It was not trained on images of males, but it can create males when prompting, if you add sac or testicles to the prompt, or a flaccid penis hangs in between the thighs and legs. Other solutions might also work here, but I myself tested 99% female characters.
This Lora is trained on only 2D images from a variety of artists, no AI art, 3D art, photorealism, or photographs were used in the dataset. It's purely anime and cartoon based, but it also wasn't trained on those triggers and tokens. So it still works pretty flexible with other lora's. For realism I strongly suggest to combine this model with 1 or more realistic and photo style gape and prolapse models.
To create 2D anime style images, use the artistname, character, and description of the character. Details such as the name of the anime/manga/game helps alot too with the original style consistency.
It is possible to achieve photo-realism with this model, but with my initial testing I have found this very hard to achieve without any additional loras. But I noticed that you need to avoid any words that relate to: games, 3d-rendering engines, game engines, 2D and manga such as lineart or dithering, or realistic/photorealistic/semi-real, etc.etc.
For examples you can download my images, which have a basic workflow and the prompts attached for Comfyui. And I also manually add the prompts to the images, so they can be extracted on the platform itself. But this seems more convenient for different users.
Version 1:
For now I only have v1, which seems to exagerrate the size of the ass, and slims down the waist exponentially. To get around this, either use another lora, or define the bodyshape with detail. To avoid any unwanted proportions.
I only uploaded to civitai, if you see it on another platform let me know.
Don't post any images that violate the law, or the TOS, and keep the platform pleasant and safe for everybody.
For the examples I used Krea2 Basemodel: fineporn_INT8_v4 by stableyogi, CLIP: qwen-3-vl-4b-heretic_int8, VAE: qwen_image_vae_krea2, inference steps: 8, sampler: euler, scheduler: simple, resolutions: 1080x1920, 1080x1080.
Hit the like button on the model page, and on my images if you enjoy what you see! I'm all out of buzz, and every bit helps so I can train more models! And if you download the lora, posting a image to the page grants me 50buzz per user, and also saving my images to your gallery helps me out bigtime! So I would appreciate it alot if you could spare a few seconds! Donations are welcome too, but don't feel obliged if you cannot afford it like me! xD
You can sell your images, and do whatever, but if you make big profit, sending a tip would be a cool gesture. ^^
Description
v1: only trained on digital 2d-anime images, using booru style tags, without natural language.
FAQ
Comments (11)
The model is also capable of a harness, that spreads the gape open. But this is a hit or miss, and haven't completely figured out how to trigger it properly. So will add this in the future, when I figure out how.
So it can do it, but it's just harder. Experiment if you can. ^^
Model is awesome,but i can only get to work with rear poses.Front poses i get gaping vagina instead
Hey thanks for the feedback. I'm gonna test some later to figure it out. So I'll see what I can do. There are some images in the dataset with a woman lying on her back. But for now i just assume there aren't enough. So either i figure it out, or need a v2 of the model already. So I need to save up some more buzz first. Apologies for the inconvenience. =0
But as temporary solution, you can try img2img by converting image to latent in comfyui. But I'll try to fix it ASAP.
Hey alabba2002394, I added some images to the gallery of front poses. Prompts attached to images and seperately on-site. :)
I noticed that this lora works much best with the model you mentioned, FinePorn v4 - INT8 | NVFP4 | BF16 | FP8 - v4 bf16 | Krea 2 Fine-tune | Civitai
@Silver_bullet666 Hey Silver_Bullet! Thanks so much for letting me know. I've been trying a bunch of models and haven't found one that doesn't create extra holes and such. I do think I have pinpointed the issue that causes it, but solving it requires me to re-caption the entire dataset. >.<
But I think the problem derives from the tagging style. Where some finetuned models have been trained on NL, and other more on single tokens. And these have developped the annoying ability to interpret the type of images that come with those styles of tags, so booru for example, or older generated images from the previous generation of models (sdxl-illu-pony-flux-etc.etc.).
The only way to fix it is by using excessive natural language, to force the booru type images to be interpreted correctly using NL. Does this make sense? :P Because if it does it might help you find other basemodels that can do what fineporn does. ^^
And if you do find any others, plz let me know! <3
Solved it, you need to describe the pussy seperately, so that the model understands it is actually the anus gaping. Problem becomes more apparent when used with other ass-focus loras. And also, you have to move the concept triggers up to the start of the prompt, before adding character descriptions. Krea2 accepts a long context-frame, but the further from start you get the lower the accuracy becomes.
Both of these also seem to work really well for other concept lora btw. ^^
I cant load the workflows from the images in the post... is that just a me problem or ?
Well, lucky for both opf us there isn't a actual complicated workflow. :P I just use basic settings to show what the base-lora can do without pulling everything out of the closet. However the workflows should've worked in Comfyui. So I do not use multi-sampling, upscaling, or anything like that. I just put in the prompt, and hit the play button.
However, I do also copy my generation settings and prompts to each individual image manually, by copying the data from the image in civitai using the metadata in comfyui, so it should've worked, as I avoid uploading images without metadata.
For the settings I used: basemodel: Fineporn_int8_v4, Text-encoder: qwen3-vl-4b-heretic_int8, VAE: wan 2.1 / qwen_image_vae. (VAE's depend on amount of saturation when generating).
inference steps: 8 with Turbo, CFG:1, sampler: euler_ancestral, solver: sgm_uniform, neg-prompt: Zero-out, and use image-saver extension to convert latent to jpg instead of png, to save about 1MB space on my SSD with each image, but can also save workflow.
I hope this helps! But if not, and it's still unclear, let me know alright!? I can post a seperate workflow as .json file if needed. Cause I really want to see what you'll create! :D xD
Basic Workflow For Tunnel Gape Lora | Civitai
Figured it out btw, was checking my nodes and it's the simple-image-saver that caused it. Been using it for so long forgot the convenience isn't comfy native, but it's in the manager. :P
However, still uploaded because I said I would, did a quick clean up to make it look a bit better. The workflow also works without the custom node btw. ^^
Minimalistic workflow to reproduce my images:
C666D Tunnel Gape - Anal Gaping Anus Krea2 v1 - C666D Tunnel Gape Krea2-1 | Krea 2 LoRA | Civitai