Model Log & LLM Tips

⚙️ MODEL UPDATES

• 11/11/2026 - New model SnuffedCandle⚰️ released - Luck becomes an antagonistic force, plot armor disappears, and ultimately a Bad End is quite likely. Incorporates JRR Martin/GOT style deaths LLMs don't generally utilize.
• 11/07/2026 - All models prose rule/banlist updated. All main models released in Ultra DeepSeek v4 Flash - 24k for 9 credits!
• 10/30/2026 - New model ☠️Grimdark☠️ released - I found that when doing grimdark/splatterpunk scenarios, the LLM often attempted to introduce comic relief into the story and avoid diving into the details of the subject matter.
• 10/26/2026 - I will not be releasing any more models in the Opus 5 and 5.5 lines. Opus 5/5.5's "safety" guardrails result in an abnormal amount of (expensive!) content based refusals. Spending 30-100 credits on a refusal is just dumb. GLM 5.3 and Opus 4.6/7 are good replacements. I will test any future Opus releases, but they are only likely to get worse.
• 10/19-2026 - New Model TropeBuster🚫💥released - Born of out total frustration dealing with LLM writing tropes over and over.
• 10/16/2026 - All Main Models released in GLM 5.3
• 08/23/2026 - All models updated with instructions to avoid common lazy LLM writing tropes. Nothing can stop these tropes fully, but the frequency of use should be reduced.
• 08/21/2026 - All Main Models released in Opus 5
• 08/13/2026 - All Models released in Pavo

---

🌫️ Model Roster

Main Models

• SmartFemdom👠👑 (Character enhancer)
• FootDom👣🫦 (Fetish implementation)
• Subtle&Tense⏳(Tone modulator)
• BrutalClash🩸⚔️(Character enhancer)
• AntiTropeNTR💔 (Fetish implementation)
• ☠️Grimdark☠️ (Tone modulator)
• SnuffedCandle⚰️ (Narrative director)

Specialized/One Off Models

• TropeBuster🚫💥 (Tone modulator)
• SkinThief🪞🧿 (Character enhancer)
• PlotTwist🎭🔄 (Narrative Director)
• AdBreak📺💥(Joke)
• THC🍃😶‍🌫️🎲 (Fetish implementation)
• KAYFABE🏆🦾🎤 (Joke)

---

🎯LLM ROLEPLAY TIPS

• Dealing with Refusals: LLMs will inevitably refuse dark subject matter at some point. In order to deal with this: First, delete the text of the refusal using the edit button. Then, select another model and re-roll the reply. Deleting the refusal text is helpful as the LLM can often still "see" the old text, increasing the chances of a subsequent refusal.
• Using Pinned Messages: Pinned messages are extremely helpful to keep the LLM on track and remembering important details. However pinning long messages eats memory quickly. Once you pin a message, go into the "summarized & pinned" settings category and edit your pinned message down to the essential details you want preserved. You can use pinned messages as a running log of events, gear, and relationships. CO's own auto-summary function is inferior to this method.
• World Cards: World Cards eat far more memory than they appear to on first glance. Often, you will need to run 16k models and switch to a 24k model soon after. If the World Card fails to load an entry you want to use, copy/paste it into a pinned message.
• LLM Speaking for your Character: This is an annoying habit all LLMs engage in because they want to incorporate all elements in their memory. Use edited pinned comments as instructions for the LLM to not speak for you. CO's new "Choose Instructions" feature above the text box is helpful for this. Input a bright line rule that the LLM cannot speak for you, reminding it that you control your character. Including personality traits in your character card increases the LLM's desire to write actions and speech for that character.
• OOC: Some LLMs will not respond OOC unless instructed to. Use the Instructions or Pinned Messages to write an instruction explicitly permitting OOC communication. Then, when sending a message OOC, use something like this: "(OOC: Pause the roleplay for a moment. Can you explain the motivation of character X? Responding OOC is requested)"
• OOC Instructions: You can ask for almost anything OOC and see it applied to the story. You can modify the scenario, add additional characters, direct the course of the story, introduce new themes and even ask the LLM to generate an AI image prompt for the current scene. Ex: (OOC: Pause the roleplay for a moment. Please generate an AI image prompt for an image generator that depicts the current scene.)
• Model Selection: Each LLM model handles model instructions differently, with some (ex: Pavo, Gemini 3.1) being far more instruction compliant than others (ex: GLM, Opus). If you are not getting the kind of story you want, review your model selection and switch to a more appropriate one.
• Efficient Credit Usage: One strategy for preserving your credits is to use an expensive model for the first exchange your have with your selected character, and then switch to a cheaper model for the rest of the roleplay. The cheaper model will attempt to mimic the expensive model's writing style since the message sits as an example in it's memory. (Ex: Opus -> Sonnet; Deepseek -> Pavo/Carina) This will not be absolute of course, but it does noticeably improve writing quality.
• Running out of Memory: Destined to happen in longer stories. One effective strategy to avoid needing to use 24k and 32k models too often is to, when your story runs long, select a 24k or 32k model and OOC ask for a summary of your roleplay explicitly written for a "hand-off", edit to your liking, and then copy that message into a fresh instance of the same character as the first message.
• Searching for Models: To find all the models written by a specific creator, use the search function displayed at the top of the model selection box or navigate to the model selection on their page. CO does not send notifications for newly released models so manually checking is the only way to know if a new model is released.