Thanks for the clear screenshots! You’ve set up everything correctly for CosmosRP-2.5 on JanitorAI. If you’re only receiving short replies, here’s what you should check and try:
1. Remove ANY Custom Prompts
- Custom instructions, jailbreaks, or detailed prompt templates can restrict model output. Make sure the “Custom Prompt” field is empty or set to a single space.
- If you’ve used custom instructions or prompts in a chat previously, start a NEW chat after clearing them. Old prompts can persist and affect output.
2. Model Limitation (CosmosRP-2.5)
- CosmosRP-2.5 is optimized for quality roleplay but, like most models, longer outputs heavily depend on the prompt's context and the actual back-and-forth.
- Quality improves if you feed the model richer, more descriptive context in each turn.
3. Max Tokens & Temperature Settings
- Your settings are fine (Max Tokens: 1000, Temperature: 1.2). If you want even longer outputs, try pushing Max Tokens up to 2000 (if possible in JanitorAI).
- Make sure your actual message to the bot invites long, story-like responses (e.g., detailed narration, open-ended prompts).
4. Generation Settings
- Always use OpenAI Default for everything except Temperature (leave as recommended, or 0.85–1.2 for lively RP).
5. No Jailbreak Needed
- Pawan.Krd’s models are completely unfiltered. Avoid using jailbreak prompts—they make output shorter and degrade quality.
6. Platform Limitation