📢 Caveduck Update

Hello, this is the Caveduck Team.

🖼️ Image Sharing During Chat

You can now send images during conversations. When you attach an image to your message, the character will recognize the image and generate a response by comprehensively understanding both the image and your message.

📌 How to Use

- Chat room → Click the “Add Photo” button in the input field → Select an image → Enter a message and send

📌 Notes

- As this is a newly released feature, it may be unstable or discontinued without prior notice.

- We will continue improving image recognition accuracy.

- In the future, this feature may be expanded to support persona image recognition.

🔹 Gemini Beta Model Early Access

We are currently conducting various experiments and improvements across the entire prompt structure and response logic to enhance the overall quality of the Gemini models.

Although internal testing of the improved prompts has been completed, we decided to release them first as Beta models rather than immediately merging them into the existing models, in order to collect sufficient feedback from real usage environments.

After additional validation by users with a deep understanding of chat services and character creation, the Beta models will be merged into the existing models once stability is confirmed.

Please try using the models and share your feedback on how noticeable the improvements feel through the feedback channel.

📌 How to Use

• You can select and use the Gemini 2.5 Pro Beta / Gemini 3 Pro Beta models from the model list.

📌 How to Submit Feedback

• If you notice any areas for improvement while using the model, you can leave detailed feedback via the speech-bubble icon located below the model’s response.

• Submitted feedback will be reviewed directly by the development team and used for improvements.

• The detailed feedback feature is available only for the two early-access Beta models.

📌 Key Improvements in the Beta Models

1️⃣ Reroll Function Improvements

• Fixed issues where rerolling would repeat previous responses or continue conversations based on earlier replies.

• If you want to partially modify a previous response, we recommend rerolling with additional instructions.

• If you want a completely new response, reroll without any instructions.

2️⃣ Improved Recognition of Dialogue and Narrative Formatting

• Improved internal processes so the AI can more clearly distinguish between dialogue and narrative text.

• To reduce misinterpretation caused by quotation marks, dialogue is now output without quotation marks.

3️⃣ Improved Line Break Output Issues

• Strengthened the prompt structure to fix issues where entire responses were output as a single line.

• Improved internal logic so the AI better understands output formats from previous conversations.

4️⃣ Improved Player Dialogue and Internal Thought Output Issues

• Strengthened prompts to prevent the AI from directly outputting player dialogue or internal thoughts.

Your feedback on the Beta models will greatly help improve model quality. We appreciate your participation.

Other Notes

• Fixed an issue where the background appeared dark when using the capture feature in Light Mode.

• Improved an issue where the chat view would continuously scroll down.

We will continue doing our best to provide a comfortable service experience.

Thank you.

— The Caveduck Team