Pony Diffusion XL is the base under more adult image checkpoints than any other model, and most people prompt it wrong. This guide covers the score-tag system, the difference between anime Pony and realism merges, the LoRA ecosystem, and the prompt structure that gets consistent explicit results.
Short answer: Pony XL is an SDXL retrain that learned explicit content natively, which is why it understands acts, positions and body configurations that other models need a LoRA for. It is prompted with tags, not sentences, and the first six tags are a quality dial called score tags. Realism merges make it photographic; the base is stylised. On MyBabes six of the image checkpoints are Pony-based, which is the largest share of any family, because for explicit multi-subject scenes it is still the most reliable base.
Where Pony came from and why it matters
Pony Diffusion XL retrained SDXL on a very large dataset of tagged art that, unlike SDXL’s own training set, included explicit material with detailed tags. The result is a model that knows what it is drawing. Prompt an act and it produces the act, in a plausible configuration, without a LoRA teaching it what the words mean. That single property is the reason Pony became the base for the 2024–2025 wave of adult checkpoints and remains one of the three families that matter, alongside SDXL realism and Illustrious. The family overview is in our Stable Diffusion comparison; this article is the Pony deep dive.
The score-tag system
Pony’s training data was rated for quality, and the rating is exposed as tags. The convention that works:
`score_9, score_8_up, score_7_up, [source tag], [rating tag], [your tags]`
score_9 … score_7_up pull toward the highest-rated training images. Use all three; dropping to `score_9` alone gives less variety, not more quality.
source tag steers the style: `source_anime`, `source_cartoon`, `source_furry`, `source_pony`. Realism merges mostly ignore it; for the base model it is essential.
rating tag sets explicitness: `rating_safe`, `rating_questionable`, `rating_explicit`. Set it deliberately; leaving it out gives inconsistent results.
Beginners either omit score tags entirely or paste a long string of them; both degrade output.
Base Pony versus realism merges
Base Pony XL / anime merges (Prefect Pony XL, Wildcard XL Pony) | Realism merges (CyberRealistic Pony, Pony Realism, Uber Realistic Porn Merge, Babes by StableYogi) | |
|---|---|---|
Look | Stylised, anime and semi-real | Photographic |
Act vocabulary | Full | Full, inherited from the base |
Skin | Painted | Close to SDXL realism checkpoints |
Backgrounds | Weak | Moderate |
Faces | Consistent, stylised | Slight rendered quality on some merges |
Negative prompt dependence | High | High |
Best for | Anime and stylised explicit art | Photoreal explicit scenes, multi-subject |
Realism merges are what most adult creators mean by “Pony” today. They keep the act knowledge and trade the stylised look for skin that is close to a dedicated SDXL realism checkpoint.
Prompt structure that works
Score block. `score_9, score_8_up, score_7_up`.
Rating and source. `rating_explicit, source_anime` (or omit source on realism merges).
Subject count. `1girl` / `2girls` / `1girl, 1boy`. Pony relies on these; get them right.
Act and position tags. Danbooru-style, specific: the position, the viewpoint, contact tags.
Subject description. Hair, body, clothing state, expression.
Scene. Setting, lighting, camera angle in tag form.
Quality tail on realism merges: `photo, realistic, detailed skin`.
Negative prompt: the family depends on it. A working baseline is `score_6, score_5, score_4, low quality, worst quality, bad anatomy, extra fingers, watermark, text`, plus `3d, cartoon, anime` on realism merges.
Keep tags comma-separated and avoid sentences; a sentence in a Pony prompt is read as a bag of words and loses the position information that tags carry.
LoRAs: where the family shines
The Pony LoRA ecosystem is the largest for adult content by a wide margin: positions, acts, camera styles (POV variants, specific angles), body types, clothing states and lighting. Practical rules from running a hundred of them as selectable mods on MyBabes:
One act LoRA at a time. Two act adapters produce merged anatomy.
Style and lighting LoRAs stack fine on top of one act LoRA at reduced weight (0.4–0.6).
Realism-merge compatibility varies; a LoRA trained on base Pony can shift a realism merge back toward stylised. Test at 0.5 before committing.
Trigger words matter; a LoRA without its trigger does little.
Common failures and fixes
Failure | Cause | Fix |
|---|---|---|
Doll-like, plastic skin | Base model or anime merge used for realism | Switch to a realism merge; add `photo, realistic` |
Wrong number of people | Missing or contradicting count tags | Put the count tag first after the score block |
Extra limbs in contact scenes | Two act LoRAs, or contradictory position tags | One LoRA; one position |
Text and watermarks | Training data artefacts | Add `text, watermark, signature` to negative |
Same face every time | Merge with narrow face data | Add specific face tags; switch merge |
Pony versus Illustrious
Illustrious has replaced Pony for anime because of tighter tag adherence and better hands. For explicit realism, Pony merges still lead on act vocabulary, and the two families are usually used together: Illustrious for composition-heavy or anime work, Pony realism for explicit photoreal scenes. Both run on MyBabes, so switching is a dropdown rather than a download.
Related in the Lab
Key takeaways
Pony XL learned explicit content natively; it has the widest act vocabulary of any image base.
Score tags are a quality dial: `score_9, score_8_up, score_7_up`, then rating and source.
Use realism merges for photoreal output; the base and anime merges are stylised.
Tags, not sentences; one act LoRA at a time; negative prompt always.
FAQ
What is Pony XL?
A retrain of SDXL on a large tagged art dataset including explicit content. It is the base for many adult image checkpoints.
Is Pony XL uncensored?
Yes. The weights have no filter and explicit content is part of its training. Hosted services may add their own moderation.
Which Pony model is best for realistic NSFW?
Realism merges such as CyberRealistic Pony, Pony Realism and Uber Realistic Porn Merge. Several run on MyBabes.
Why do my Pony images look bad?
Almost always: missing score tags, a sentence prompt instead of tags, or no negative prompt.
Last tested: August 2026.
Models covered
Skip the filter
Run these models with adult content enabled
MyBabes runs Wan 2.7, Seedance 2.0 and Krea inside its own generation stack, so the prompts refused on official apps generate as written. No GPU, no setup.
Open MyBabes
Builds MyBabes, runs the Lab
Jiri builds MyBabes and runs the Lab’s model testing: the same prompt set on every image, video and chat model, on the vendor’s official surface and on MyBabes, re-run after every release. He writes up what the models actually do, not what the launch posts say.
Follow on X
More from the Lab
Latest articles
All articles
Analysis
AI Porn and the Law in 2026: What's Actually Illegal, What's Regulated, and Where the Lines Sit
2025 and 2026 turned AI adult content from a legal grey zone into a regulated space with real enforcement: a US federal crime for non-consensual imagery, deepfake laws in 48 states, UK fines for missing age checks, EU labeling duties. Here is the map — and the one principle that organises all of it.
·
5 min read
How-To
Why AI Porn Looks Off: Every Artifact Explained, What Causes It, and How to Fix Each One
Everyone can tell when an AI image is off, few can say why. Each artifact has a specific technical cause and a specific fix — hands, limbs, faces, skin, teeth, text, seams, smear, flicker. This is the field guide, for images and video.
·
5 min read
Analysis
ChatGPT and NSFW in 2026: The Adult Mode That Never Shipped, What's Actually Allowed, and Where People Went Instead
"Treating adults like adults" was the promise in October 2025. Ten months later the adult mode is paused with no timeline, ChatGPT still refuses explicit content, and the search interest hasn't gone anywhere. Here is the full story, the current policy line in practice, and what the alternatives actually are.
·
4 min read