<div style="max-width:700px;margin:40px auto;backgroundcolor:rgba(13,11,18,0.97);padding:20px;border-radius:10px;box-shadow:2px 2px 8px rgba(0,0,0,0.3);color:#e0dcea;font-family:Georgia,serif;border:1px solid #3a2a5a;"><h2 style="text-align:center;font-size:1.2em;margin:6px 0 10px;letter-spacing:0.6px;color:#fff;text-shadow:0 0 6px #4a7a9b,0 0 14px #2a5a7b;">Also a guide how to not be an asshole | Update: 21st Feb 2026</h2> <hr style="border:0;border-top:1px dashed #3a2a5a;margin:20px 0;"> <div style="margin-bottom:14px;color:#b3a8d4;"> <strong>How to appropriately leave comments regarding bot error</strong> </div> <div style="margin-bottom:14px;font-size:0.92em;line-height:1.75;color:#d0ccde;"> Let me be blunt first. <br><br> There's a specific type of user who drops <em>"this bot sucks, replies cut off, don't play and I don't recommend"</em> and goes completely silent the moment I ask <em>"which model were you using?"</em> No response. Gone. Review stays up. <br><br> That's not a frustrated user. That's someone who wanted to leave a mark and walk away. And it matters — because <strong style="color:#b3a8d4;">I track which models cause problems and how often.</strong> When you ghost that question, you're not just being unhelpful. You're removing information that could help the next person. <br><br> For everyone who genuinely didn't know what was going on? <em>That one's on CrushOn.</em> They never explained how any of this works — no model breakdowns, no memory warnings, no transparency. So let's fix that. <br><br> <span style="color:#d29ad6;font-style:italic;">Tap the sections below to understand why this happens. 👇</span> </div> <p style="text-align:center;color:#3a2a5a;margin:10px 0;font-size:0.85em;">✦ ── ── ── ── ✦ ── ── ── ── ✦</p> <div style="font-family:Georgia,serif;"> <details style="margin-bottom:7px;border-radius:8px;overflow:hidden;border:1px solid #3a2a5a;"> <summary style="background-color:rgba(16,13,22,0.96);color:#c0b8d8;padding:12px 14px;cursor:pointer;list-style:none;display:flex;justify-content:space-between;align-items:center;"> <span>🤖 The Bot and the Model Are Two Different Things</span><span style="color:#6a5a8a;">▼</span> </summary> <div style="background-color:rgba(10,8,16,0.85);padding:14px;color:#d0ccde;font-size:0.88em;line-height:1.8;border-top:1px solid #2a1a4a;"> A bot is a <strong style="color:#b3a8d4;">character</strong> — a personality, backstory, and set of instructions a creator builds. But the bot doesn't think on its own. It hands those instructions to a <strong style="color:#b3a8d4;">language model (LLM)</strong> — the actual engine doing the generating. <br><br> Think of it like a <em>screenplay and an actor.</em> The creator writes the script. The model performs it. A brilliant script handed to the wrong actor still falls flat — and a talented actor can only do so much with bad stage conditions. Blaming the screenwriter because the production was underfunded isn't a review. It's misdirected frustration. <br><br> CrushOn offers multiple models and they are <strong style="color:#b3a8d4;">not equal — not even close.</strong> But CrushOn doesn't explain this. There's no breakdown of what each model can do or what kind of bot it's suited for. You're expected to figure it out. Most users never do. </div> </details> <details style="margin-bottom:7px;border-radius:8px;overflow:hidden;border:1px solid #3a2a5a;"> <summary style="background-color:rgba(16,13,22,0.96);color:#c0b8d8;padding:12px 14px;cursor:pointer;list-style:none;display:flex;justify-content:space-between;align-items:center;"> <span>💰 Free vs Pro vs Ultra — They Are NOT the Same</span><span style="color:#6a5a8a;">▼</span> </summary> <div style="background-color:rgba(10,8,16,0.85);padding:14px;color:#d0ccde;font-size:0.88em;line-height:1.8;border-top:1px solid #2a1a4a;"> CrushOn splits its models into three tiers: <br><br> <strong style="color:#b3a8d4;">🆓 Free Models</strong> — No coins needed. These are the weakest available: lower intelligence, shorter outputs, shakier instruction-following. They exist for accessibility, <em>not quality.</em> Think of it like a convenience store microwave vs. an actual kitchen — technically both heat food, but you're not getting the same result. Most complex bots were <strong style="color:#b3a8d4;">never tested on free models.</strong> <br><br> <strong style="color:#b3a8d4;">🪙 Pro Models</strong> — Cost coins per message. Some cost more (e.g. Deepseek runs at 2x rate). These are meaningfully more capable. Better coherency, better outputs, better instruction-following. <em>If a bot felt flat on Free and suddenly performs on Pro, that's not magic — that's just using the right tool.</em> <br><br> <strong style="color:#b3a8d4;">👑 Ultra Models</strong> — Fully paywalled. Higher-end models like Claude Opus live here. Theoretically the best experience on the platform. In practice? <strong style="color:#b3a8d4;">Ultra has a track record of inconsistent performance — and that's on CrushOn.</strong> Outages, degraded responses, unexpected behavior, and poor communication when things go wrong. If your Ultra experience suddenly tanks for no reason, there's a real chance that's a <em>platform management problem,</em> not the bot. </div> </details> <details style="margin-bottom:7px;border-radius:8px;overflow:hidden;border:1px solid #3a2a5a;"> <summary style="background-color:rgba(16,13,22,0.96);color:#c0b8d8;padding:12px 14px;cursor:pointer;list-style:none;display:flex;justify-content:space-between;align-items:center;"> <span>🧠 "Memory" Has a Hard Ceiling — You're Never Told Where It Is</span><span style="color:#6a5a8a;">▼</span> </summary> <div style="background-color:rgba(10,8,16,0.85);padding:14px;color:#d0ccde;font-size:0.88em;line-height:1.8;border-top:1px solid #2a1a4a;"> Every model has a <strong style="color:#b3a8d4;">context window</strong> — a hard cap on how much text it can hold and process at once. This includes the bot's definition, your full chat history, your persona settings, and every reply the AI has generated. <em>All of it counts.</em> <br><br> Imagine a whiteboard with a fixed size. Every message gets written on it. When it fills up, <strong style="color:#b3a8d4;">the oldest content gets erased permanently.</strong> The model never sees it again. So when a bot seems to forget something from earlier, it's not being careless. <em>That information was wiped before the message even reached it.</em> <br><br> Here's the part most people miss: <strong style="color:#b3a8d4;">a bigger context window doesn't mean the model reads all of it well.</strong> Research shows the more text a model has to process, the worse it performs. Information buried in the middle of a long prompt is the most likely to get ignored. Even top-tier premium models start making errors once conversations get long enough. Smaller free models hit that wall much sooner. <br><br> <strong style="color:#b3a8d4;">Solution:</strong> make edit and put important information inside your profile card. Keep reminds the AI. </div> </details> <details style="margin-bottom:7px;border-radius:8px;overflow:hidden;border:1px solid #3a2a5a;"> <summary style="background-color:rgba(16,13,22,0.96);color:#c0b8d8;padding:12px 14px;cursor:pointer;list-style:none;display:flex;justify-content:space-between;align-items:center;"> <span>✂️ Why Replies Cut Off — and Why It's Not the Creator's Fault</span><span style="color:#6a5a8a;">▼</span> </summary> <div style="background-color:rgba(10,8,16,0.85);padding:14px;color:#d0ccde;font-size:0.88em;line-height:1.8;border-top:1px solid #2a1a4a;"> The model has a <strong style="color:#b3a8d4;">generation limit</strong> — a hard cap on how many tokens it produces per reply. It stops when it hits that ceiling, finished thought or not. Complex bots with detailed definitions also eat into the space available for a full response. And weaker models often lose track of long formatting instructions partway through and drop out early. <br><br> <em>It's like asking someone to read you a story but they're only allowed to speak for sixty seconds — they'll stop mid-chapter whether the scene is resolved or not. That's a timer problem, not a story problem.</em> <br><br> The same bot on a stronger model may give you complete, well-structured replies every time. On a weak free model, it falls apart. <strong style="color:#b3a8d4;">CrushOn doesn't tell you this.</strong> </div> </details> <details style="margin-bottom:7px;border-radius:8px;overflow:hidden;border:1px solid #3a2a5a;"> <summary style="background-color:rgba(16,13,22,0.96);color:#c0b8d8;padding:12px 14px;cursor:pointer;list-style:none;display:flex;justify-content:space-between;align-items:center;"> <span>🛠️ Custom Models vs Core Models — Know the Difference</span><span style="color:#6a5a8a;">▼</span> </summary> <div style="background-color:rgba(10,8,16,0.85);padding:14px;color:#d0ccde;font-size:0.88em;line-height:1.8;border-top:1px solid #2a1a4a;"> Some of my bots run on a <strong style="color:#b3a8d4;">custom model</strong> — meaning its built on top of a base model (like Deepseek) with its own system prompt, formatting, and instructions layered over it from the creators. <em>That custom layer is what shapes the character's voice and behavior.</em> <br><br> But <strong style="color:#b3a8d4;">the base model underneath doesn't belong to creators.</strong> <br><br> Think of it like a chef working in a restaurant kitchen. The chef controls the recipe, the technique, the plating. <em>But if the supplier sends bad ingredients or the kitchen equipment breaks down, the dish suffers — and that's not the chef's fault.</em> If, for example, Deepseek's core is having server issues, every bot running on Deepseek is affected. No amount of careful prompting on my end can fix a problem that lives in the foundation. <br><br> So if something feels off on a custom model bot, <strong style="color:#b3a8d4;">check whether the base model itself is having a bad day</strong> before assuming my build is broken. Core model errors are platform and provider-side problems. <strong style="color:#b3a8d4;">They are not something I can patch or prevent.</strong> </div> </details> <details style="margin-bottom:7px;border-radius:8px;overflow:hidden;border:1px solid #3a2a5a;"> <summary style="background-color:rgba(16,13,22,0.96);color:#c0b8d8;padding:12px 14px;cursor:pointer;list-style:none;display:flex;justify-content:space-between;align-items:center;"> <span>📚 Not Every Bot Is Built the Same — and That's Fine</span><span style="color:#6a5a8a;">▼</span> </summary> <div style="background-color:rgba(10,8,16,0.85);padding:14px;color:#d0ccde;font-size:0.88em;line-height:1.8;border-top:1px solid #2a1a4a;"> NSFW AI platforms exist for a wide range of purposes — and there's nothing wrong with that. <br><br> Some bots are built for <strong style="color:#b3a8d4;">pure, immediate adult content</strong> — straightforward, accessible, does exactly what it says. These don't need the model to track complex lore or maintain a cast of characters. They serve their purpose and have their own audience. <br><br> Other bots are <strong style="color:#b3a8d4;">high-investment narrative builds</strong> — layered worldbuilding, detailed character psychology, structured story continuity. These ask significantly more from both the model and the user. They're far more sensitive to model quality because there's so much more that needs to be maintained consistently. <br><br> <em>The difference isn't a ranking. It's just scope.</em> Recognizing that a creator put serious craft into a build is the same as knowing the difference between a fast food order and a tasting menu. Both are valid. But you wouldn't review a tasting menu based on eating it reheated in a break room microwave. </div> </details> </div> <p style="text-align:center;color:#3a2a5a;margin:14px 0 10px;font-size:0.85em;">✦ ── ── ── ── ✦ ── ── ── ── ✦</p> <div style="background-color:rgba(10,8,16,0.85);border:1px solid #3a2a5a;border-radius:8px;padding:13px 14px;color:#d0ccde;font-size:0.88em;line-height:1.9;"> <strong style="color:#b3a8d4;">📊 Vala's Recommended Models (lists may update more)</strong> <br><br> ❤️ <strong style="color:#c0b8d8;">Story Narrative:</strong> Claude Sonnet / Opus(Norisor, Raon, Mei-chan), Mochi V1 or V2(Norisor, Raon), GeminiFlash(Norisor, Lilie, Sofia), Mira(Lulupri)etc <br> 🔥 <strong style="color:#c0b8d8;">Smut:</strong> Claude Sonnet Opus(same like above), Deepseek(Raon, Norisor), Stheno &amp; Leo(Mei-chan), etc <br> 🗓️ <strong style="color:#c0b8d8;">Bots with Tracker:</strong>Stheno(Mei-chan, BQ, Trxshpxndx), may add more due to experimental <br><br> ⚠️ <strong style="color:#c0b8d8;">About Free Models</strong>: Use Free Models ONLY with awareness of the possibility of broken or unfinished replies. Proceed with this knowledge. I recommend using the custom model under the same creator's name as already mentioned above. ALWAYS pay attention to the model description. <br><br> <span style="color:#d29ad6;font-style:italic;">Model availability and performance may change — CrushOn manages this on their end, not mine.</span> <div style="width:200px;margin:12px auto;"> <div style="margin-top:12px;font-style:italic;font-size:1em;text-align:center;color:#a89ec0;"> ▼ This is Tracker ▼ </div> <img src="https://i.ibb.co/bjmjnMqt/Tracker-Sample2.jpg" style="width:100%;border:2px solid #7a6a9b;border-radius:6px;box-shadow:2px 2px 4px rgba(0,0,0,0.3);" alt="Tracker Sample"> </div> </div> <div style="margin-top:10px;background-color:rgba(10,8,16,0.85);border:1px solid #3a2a5a;border-radius:8px;padding:13px 14px;color:#d0ccde;font-size:0.88em;line-height:1.9;"> <div style="margin-bottom:10px;color:#b3a8d4;"><strong>📌 Bot refers to the wrong pronouns!</strong></div> <div style="margin-bottom:14px;">Add your pronouns in your profile card for clarity.</div> <div style="margin-bottom:10px;color:#b3a8d4;"><strong>📌 The bot keeps speaking for me...</strong></div> <div>Some models do that — especially with 2nd person intros. Use **OOC: do not talk for {{user}}** at the end of your reply before sending. You can edit the replies and delete the part where it controls you, sometimes if the custom models weren't prompted to not speak for user may also did that. If the problem persists, try switching models(refer back to Free vs Pro vs Ultra section).</div> <div style="margin-top:10px;color:#b3a8d4;"><strong>📌 The bot keeps waiting for me to tell it what to do...</strong></div> <div style="margin-top:6px;">Most LLMs are trained as assistants — conditioned to pause and wait for user input. That instinct bleeds into roleplay even when the character is supposed to be driving. Dominant characters are hit hardest since their whole personality requires taking initiative, but the model keeps handing control back to you regardless. <br><br> <strong style="color:#b3a8d4;">Solution:</strong> Make an edit and put important information inside your profile card. Keep reminding the AI — phrases like <em>"[OOC: Drive the scene. Do not wait for {{user}} — act, continue, move forward. Never end replies waiting for {{user}}'s input.]</em> in your reply also another option.</div> </div> <div style="margin-top:10px;background-color:rgba(16,13,22,0.96);border-left:3px solid #7a6a9b;padding:13px 14px;border-radius:8px;color:#d0ccde;font-size:0.88em;line-height:1.8;"> <strong style="color:#b3a8d4;">📌 What Actually Useful Feedback Looks Like</strong> <br><br> <em>"Replies were cutting off mid-sentence on [model name]. Might be a context or output limit issue."</em> <br><br> That one detail turns a complaint into something actionable. Without the model name, <strong style="color:#b3a8d4;">no one can help you</strong> — not me, not the community, not CrushOn. </div> <div style="text-align:center;font-size:1.1em;color:#ac79b0;font-style:italic;"> 📖 Knowledge referenced from <strong>Small Guide to LLMs</strong> by @whalez (Saucepan.AI) </div> </div>