This is the third time I've used the name Lilium. It's starting to feel like I'm deliberately picking a fight with Lilium. 🤣🤣🤣

Because of some crappy bugs, yes, the kind caused by Olivia's kind of bug, I can only re-upload the character.

All my SFW characters strictly adhere to SFW content, including hidden prompts where additional safety measures have been implemented. Therefore, if any inappropriate usage occurs, it is entirely the fault of the LLM or the result of the user employing improper methods to bypass restrictions. In the event of a report, I request that administrators treat it as my counter-report against the reporter for inappropriate usage, with the evidence being the very material provided by the reporter.

If budget allows, I recommend starting with "Claude 3.5 Sonnet v2" or "Trevor" for a better experience. However, I also strive to ensure low-cost models provide acceptable performance. Users of low-cost models can try starting with "DeepSeek," then alternate with "Haiku" and "GPT-3.5." Switching between these three when responses become rigid or the interaction experience is unsatisfactory can help improve the issue.

Click on the name "MuYi" beside the character image, or visit the following link to view more of my characters. If you have good suggestions or ideas about the characters, feel free to leave a message and communicate with me.

Update

2026-02-26: Because the built-in Summary and Character State features on Caveduck are already highly effective, redundant custom operations tend to cause interference. Therefore, all prompts used for hidden pseudo-CoT and hidden state definitions have been deleted. 2025-03-17: Use the Affinity System (requires starting a new conversation). 2025-02-11: Adjusted the prompt to improve performance across different models, including the new DeepSeek and Trevor models, as well as the original low-cost models like Haiku. In theory, this should lead to better results. User feedback is welcome, and if it performs well, I'll apply it to my other characters. Applied some techniques to enhance the platform's automatic translation feature, aiming to improve multilingual performance. Feedback from users of different languages is 2025-01-19: Considering that low-cost LLMs have poor understanding of tables, the format for hidden messages has been changed back to the original list format. 2025-01-18: Thanks to Olivia’s "contribution", the character was re-uploaded. 2025-01-11: Added several images - dancing/drawing/exercising/reading/using computer.