Illustrious vs Pony: Picking (and Prompting) the Right Style on Yamete

Pick a style, type your usual prompt, and sometimes it just... doesn't land the way it did last time. That's not you doing anything wrong — it's that "style" on Yamete isn't a filter slapped on top of one model. It's a full checkpoint swap, and each one has its own dialect.

Here's what actually changes when you switch styles, and how to prompt for the two anime-leaning options specifically: Illustrious and Pony.

Style is a different model, not a different filter

When you open the style picker in the generator, you're choosing which underlying checkpoint renders your image — not applying a post-processing look. Among the options you'll find an Illustrious-based anime style, a Pony-based anime style, and a realistic-render option, each with its own preview assets and its own trained behavior.

This matters for your Extensions too: Extensions (the modifier chips you slot into a generation) are style-compatible, not universal. Switch styles and Yamete swaps each slotted Extension for its equivalent under the new style where one exists — and drops it if it doesn't. If a favorite Extension vanishes after a style switch, that's why: it simply has no counterpart trained for that checkpoint.

Illustrious: closer to describing the image

Illustrious-based models were trained to hold up well with prompts that read more like natural description than a tag dump. You can still lean on Danbooru-style tags — it understands them fine — but it doesn't need the same density of tag-stacking to stay coherent. Line work and color tend to stay clean even with a lighter prompt.

Practical takeaway: if your prompt currently reads like a wall of comma-separated tags, try trimming it down to the handful that actually matter, plus a short natural-language description of the scene. Illustrious tends to reward restraint over volume.

Pony: built on tag stacking and quality prefixes

Pony-based models come from a different lineage — trained hard on Danbooru-tag conventions, and they respond best when you prompt the way the training data looked: comma-separated tags, most specific concepts first.

The other thing worth knowing: the wider Pony community leans on quality-score prefix tags — score_9, score_8_up, score_7_up... on the positive side, score_6, score_5, score_4... on the negative side — to steer toward the higher-quality end of what the model learned. You don't need to type these yourself on Yamete — the Pony-based style already attaches its own quality prefix to both your positive and negative prompt automatically, so adding your own copy just wastes prompt budget.

Practical takeaway: spend your prompt on the tags that actually describe your image — subject, pose, outfit, composition. The quality scaffolding is already handled for you; if a Pony-based render still feels off, check tag order and specificity first, not the quality tags.

So which one should you actually use?

Neither is strictly "better" — they're different tools:

• Reach for Illustrious when you have a specific scene in mind and want to describe it more than tag it.
• Reach for Pony when you're composing from a precise list of Danbooru concepts and want tight control over each one.

If a generation isn't landing, switching style is a legitimate troubleshooting step — not just a re-roll. A prompt tuned for one checkpoint's dialect can genuinely underperform on the other, independent of anything else about the prompt.

Quick recap

• Style = checkpoint swap, not a filter — each style has its own prompting dialect.
• Switching style can silently swap or drop slotted Extensions that don't have an equivalent.
• Illustrious rewards lighter, more natural-language prompts.
• Pony rewards dense Danbooru tag-stacking — skip the score_9/score_6-style quality tags yourself, Yamete already adds those for you on both the positive and negative side.
• If output quality dips, check whether your prompt actually matches the style you picked before assuming the prompt itself is the problem.