
Worldhopper 1.1: now with less generic prose, multiple preset options, improved state tracking, and an improved extension for possession and mind control RP
Hello again, friends!
I know it's only been about a week since 1.0, but I've been testing and tweaking pretty much nonstop since then, and I've been going through every bit of feedback y'all sent. Quick refresher in case you missed the first post:
The presets are for any roleplay and don't need the extension, and the Engine is a niche tool for metaphysics-specific roleplay (mind control, possession, etc).
Prose Updates
- Style Lenses. Six optional narrator voices: Hardboiled, Wry, Literary, Gothic, Pulp and Warm. Each one is a short description of the voice plus a sample to tune to. Switch on one per chat (or none, if you like the novelistic default), and the Smut lens still stacks on top of whichever you pick.
- How People Talk. Characters stop talking like an assistant ("Here's the deal", "Two things.", "Real talk") or a therapist ("that's completely valid", "how are you holding up?"). Personal questions come from what the character wants out of you, and when you say something heavy, people react like people, which sometimes means badly.
- Nobody bends toward you without a reason. A bold line doesn't win instant respect, pleading doesn't undo a refusal, NPCs don't magically share your persona's hobbies, and characters don't slowly drift nicer and more agreeable over a long chat.
- The Director now also picks what the prose pays attention to each turn (hands, doorways, who's looking at whom, that sort of thing) and a first letter for any new character's name. Goodbye, Elara and Kael.
- Intimacy Draws. Random picks for position, flavor, tempo and what's driving the other character, so sex scenes don't all play out the same way. It only kicks in once a scene is already sexual, and it never decides anything for your character.
- The banned-phrase list picks up the AI slopisms that kept showing up in my own chats (and WH Editor helps to enforce this as well).
Preset Updates
The flagship Worldhopper preset has every model's setup in one: Claude, Gemini, DeepSeek and GLM. Blocks meant for specific models start with a colored dot for each (🟠 Claude, 🔵 Gemini, 🟢 DeepSeek, 🟣 GLM), like 🟠 ✚ Teeth [Claude], so you can go down the list once, switch on the ones with your color and switch off the rest. Blocks without a dot are for everyone. It's the one to grab if you hop between models.
If you'd rather not touch anything, each model family also has its own preset: WH Shifter for Claude, WH Planeswalker for Gemini, WH Delver for DeepSeek and WH Nomad for GLM. Each one holds only its own model's blocks, without the other models' stuff or any dots to sort through, and comes with its own samplers set (temperature, top P, top K, penalties), since those really aren't one-size-fits-all. WH Delver also leaves out a few optional blocks, because DeepSeek follows a shorter prompt more closely. They're all built from the flagship.
- One setting to change, whatever model you use: set Prompt Post-Processing (in your connection settings) to Semi-strict (no tools). While testing I found that Claude Sonnet through NanoGPT was ignoring the instructions that sit after the chat: the per-turn rules, the last-second checks and the formatting. With Semi-strict it follows them, and it didn't change anything for the models that were already fine.
- Impersonate works properly now. It used to fight the rules that tell the model never to write for you. It gets its own instruction now, and every model I tried writes your next message in your own voice.
- GLM support is new, as WH Nomad. I tested GLM 5.3 on real scenes next to its thinking version and DeepSeek, and it needed very little: the shared blocks plus the last-second check that stops it from writing your character's part. The plain model follows your lead closely, while the thinking one takes more initiative on its own.
- If you were on WH Planeswalker or WH Delver, they're back as 1.1 with everything above, and if you were on Worldhopper 1.0 with Claude, WH Shifter is your new home (Worldhopper itself is the flagship now). They're separate imports, so your old 1.0 presets stay as they were.
Worldhopper Engine 1.1
Content note: this is adult fiction about control, and nothing switches on until you pick a mode for a character. You can skip this extension entirely if you're not into it.
- The Mind Book is the big addition in 1.1. Hypnosis, mind control and altered perception each keep their own record now, separate from the Body Ledger, and there's a bar above the chat for it (like the Ledger's strip) that you can tap/click to edit.
- Hypnosis is realistic by default. Inductions take time, trances deepen in stages from light to somnambulistic, susceptibility varies from person to person, and every session makes the next one easier. The model only gets the line for the stage someone is actually in, so it knows a light trance won't hold lasting conditioning, and when you say a trigger word, it fires.
- Mind control keeps track of what the controller set in someone's will, with newer settings replacing older ones, and what's still their own. By default they stay fully themselves apart from that, so lines like "somewhere deep inside, she screamed" get cut, unless your card or your story has set them up as drones or as fighting it.
- Altered perception (renamed from Perception Rewrite) tracks who can't perceive whom. When that person is in a scene with the people who can't see them, the model gets told how the scene looks from their side.
- The Body Ledger is leaner. It only runs for modes where someone is actually in another body, so it stays out of the way (and off your model bill) in chats that don't need it, and bodies no longer fall off it just because the scene moved somewhere else or nobody mentioned them for a while.
- There's an on/off switch in the Engine's header now, so you can turn it all off without disabling the extension, and your boxes and badges still show while it's off. Most of the old toggles are gone and just part of how it works.
- Under the Ledger there's a "Last turn" line that shows how long each turn spent on the Engine, so if things feel slow you can see whether it's the Engine or your model.
WH Editor (new, separate extension)
The Editor from 1.0 is its own extension now: WH Editor. It catches stock phrases, "not X, but Y" framing, chopped dialogue and the like, and has your selected model rewrite only those lines, with one tap/click to bring the original back. It works with any preset and any card, no Engine needed, on Chat or Text Completion, and the switch in its header is the only on/off you need. If you had the Editor on in 1.0, install WH Editor and your settings and slop list carry over (the Engine will remind you once).
Links
- Presets, regex and the Card Builder: https://coz-rp.github.io/
- Engine: https://github.com/Coz-rp/SillyTavern-Worldhopper
- WH Editor: https://github.com/Coz-rp/SillyTavern-WH-Editor
The Engine updates from SillyTavern's extension manager like before, WH Editor installs from its link the same way, and the presets are fresh imports from the site.
Credits
I tested a pile of other presets this week and re-tooled some of the best ideas for these presets, so this list got a bit longer:
- The toggle setup in the preset follows Marinara's preset.
- Realistic Frankenstein: the last-second gate checks, Vocation, Vocal Limits, the attention rules in Voice & Prose, and the Impersonate fix.
- Sola V2 Flame and Realistic Frankenstein: How People Talk.
- Pura's Director: the Director's focus and name draws (with Geechan's name randomiser and Deigo's banned-name list), and Intimacy Draws.
- Megumin Suite, Writer's Block Unlimited, Sola V2 Flame and Realistic Frankenstein: the "nobody bends toward you" rule.
- Megumin Suite, Sola V2 Flame, Pura's Director and Writer's Block Unlimited: the voices behind the Style Lenses, and a good number of banned-phrase entries.
- The Director module, and the idea behind the Editor, come from Orb's Director → Writer → Editor setup.
Everything is still public domain, so take whatever's useful. Bug reports and "the model still does X" reports are very welcome, here or as GitHub issues. Same ask as last time: please don't turn the replies into a debate about the Engine's topics. If it's not for you, then it's not for you!
Thanks, homies!!