PonyDiffusionXL: https://civarchive.com/models/257749?modelVersionId=290640
IF DOWNLOADS FOR ANY OF THESE MODELS ARE DISABLED, PLEASE LET ME KNOW! I WANT EVERYONE WHO WANTS THESE TO BE ABLE TO USE THEM!
Version Guide:
r5-POPPING: An earlier Epoch than r4. Should have less artifacting.
r4-POPPING: Slightly over-trained and can look a bit artifacty unless used at a reduced strength (0.8). Handles popping and various balloon shapes a lot better. Can potentially handle psuedo-realistic images.
r3-POPPING: IF YOU DON'T KNOW WHICH TO USE, PICK THIS ONE. A complete retrain. Should be more flexible in allowing balloon-popping content. Much larger file size. (Has a 128 Dim Unet and uses Prodigy optimizer to handle the learn rate, instead of AdamW with CosineR.)
r2: More flexible, better tagging
r1: The first version. Less consistent results, and less stylistic change to the existing PonyDiffusion model, but should be able to work well for img2img.
--
A variant of the previous Sitting on a Balloon LoRA. https://civarchive.com/models/217624/sitting-on-a-balloon-sdxl?modelVersionId=287009
A WIP/experiment. The original version of the LoRA worked well enough with Stable Diffusion, but not with PonyDiffusion. I'm not used to Pony Diffusion, so my dataset might not be properly aligned with the PonyDiffusion standards.
For people who really want to see characters sitting on a balloon. Geared more towards semi-poppers and non-poppers, I can't promise that balloon-popping images will never be generated. Happy squeaking!
Description
Last version I'll be having for some months. This one has significantly less artifacts than r4, but should otherwise be effectively the same.
One of the backup earlier epochs I saved before shutting down the VM. I saw a decent local minimum that was roughly half-way between the slightly burned r4 and the less-reliable r3.
FAQ
Comments (2)
The prompting is a bit difficult to understand. As far as I understand it till now (using A1111). You need this "score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up" stuff to generate decent images. Also, either "rating_questionable" or "rating_explicit" must be included, otherwise only noise is generated.
Not having the rating tags causes noise? I didn't know that! I had been using a lot of rating_questionable out of preference. There's also a rating_safe tag but I hadn't used it much.
PonyDiffusion6-specific stuff can be found here: https://civitai.com/models/257749?modelVersionId=290640
"
Make sure you load this model with clip skip 2 (or -2 in some software), otherwise you will be getting low quality blobs.
This model supports a wide array of styles and aesthetics but provides an opinionated default prompt template that allows generation of high quality samples with no negative prompt and otherwise default settings
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, just describe what you want, tag1, tag2(previous Pony Diffusion models used a simpler score_9 quality modifier, the longer version of V6 XL version is a training issue that was too late to correct during training, you can still use score_9 but it has a much weaker effect compared to full string.
[...]Other special data selection tags include, 'source_pony', 'source_furry', 'source_cartoon' and 'source_anime' and ratings of 'rating_safe', 'rating_questionable' and 'rating_explicit'."
--
Some of the tags I tried to induce via the dataset (but did not test very well) are explained here: https://civitai.com/models/226777/sitting-on-a-balloon-sd15-lrev7
"
Some useful tags:
NonPop (For images with no hint of popping balloons. Not reliable, but it seemed like a good idea to have.)
For Controlling the Balloon Style:
Realistic Balloon (Skews towards photography)
Unrealistic Balloon (Unintuitively, this skews towards balloons that are physically impossible. Primarily intended for the Negative Prompt.)
Detailed Balloon (Attempt to invoke more details on balloon necks and ties)
Simplified Balloon (More Cartoony)
Bubble Balloon (For balloons that are effectively indistringuishable from a bubble in that they tend to be perfectly round and have no opening.)
Balloon Clip (Attempts to add a clip instead of a tied knot)
Tied Balloon
Necked Balloon
Overinflated Balloon
Soft Balloon (When the balloon has low resistance against the force pressing on it.)
Squishy Balloon (Attempts to show more stretching and distending on the balloon surface.)
Underinflated Balloon (There were few of these in the dataset, probably won't work well.)
One Balloon (Will reduce the balloon count, but won't always get it down to one.)
Two Balloons (Probably won't work well, there are few samples for this.)
Many Balloons
Popped Balloons
Balloon Pieces (synonym)
Imminent Balloon Popping (It looks like a balloon will pop soon, intentional or not)
Trying to Pop a Balloon (When a character looks like they're trying to pop a balloon.)
Pop Tease (When a character looks like they're going to pop a balloon, but there is no evidence that they have popped any yet.)
For Controlling Viewing Angles:
From Above
From Below
From Front
From Side
From Back
For Controlling Pose:
(Honestly, I highly recommend using ControlNet instead to do custom poses if you want something specific.)
Bouncing (Which should involve more visibly energetic riding)
Grinding (Might be more suggestive.)
Rubbing (Similar, but a bit less abrasive)
Humping (Blatantly suggestive)
Crotch (Often visible with sitting and bounces poses, might be useful)
Spread Legs (Same Idea)
Implied Masturbation (When the hints are mostly in body language.)
Masturbation (When the hints are more explicit than body language.)
Trying to pop a Balloon
Sitting on a Balloon
Sitting
Balloon Fetish
Looner
"
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.










