How to Optimize Janitor AI Generation Settings for Better Results in 2026

Open your web browser and search for "Janitor AI." Click through to the official Janitor AI website from the search results. You'll land on the homepage displaying various character options.

Ensure you're logged into your account before proceeding. The initial navigation, covered at 0:16, establishes your starting point for all optimization work.

Browse the available characters and choose one that matches your intended conversation style. Consider whether you want casual banter, creative storytelling, or informative dialogue—your choice determines how optimized settings manifest in actual conversations.

Before starting your chat session, look for the "limitless" option on the character's profile page. Selecting this option unlocks the full range of customization parameters you'll adjust throughout this tutorial. Click on your chosen character to open their chat interface. The character selection process occurs at approximately 0:16 to 1:02 in the video walkthrough.

Look for the hamburger menu icon (three horizontal lines) in your chat interface, typically located in the top corner or side panel. Click on this menu icon to reveal navigation options.

From the dropdown or side menu, locate and select Generation Settings. This control panel is where all AI behavior modifications take place. You'll see a new panel or modal window open with various sliders and input fields.

This step corresponds to 1:02 to 1:12 in the video. Keep this panel open as you work through each parameter in the following sections.

Locate the temperature slider or input field within your generation settings panel. This parameter controls the creativity and randomness of the AI's responses. Set the temperature value between 0.7 and 0.9 for optimal results.

Lower temperature values (closer to 0.7) produce focused, predictable, and conservative responses. Higher values (closer to 0.9) generate imaginative, varied, and creative outputs. A value of 0.8 provides lively responses while maintaining coherence.

Temperature is one of the most impactful settings you'll adjust, covered at 1:12 to 1:34. Start with 0.8 as your baseline and note how responses feel during testing. You can fine-tune this value later based on whether you need more creativity or more consistency.

Open or expand the advanced settings section within your generation settings panel. Locate the Top K parameter, which controls how many word choices the model considers at each generation step. Set this value between 40 and 80 for balanced performance.

A Top K value around 60 is recommended as a starting point. This setting keeps the AI's vocabulary varied and interesting without allowing it to drift off-topic or use inappropriate word choices. Lower values create conservative word selection, while higher values permit more experimental vocabulary.

The Top K setting prevents the model from considering extremely unlikely words while maintaining creative flexibility. This parameter is discussed at 1:34 to 2:07. After setting Top K, you should notice more controlled yet still varied language in responses.

Still within the advanced settings section, find the Top P parameter slider or input field. This setting implements nucleus sampling, which narrows word choices to the most likely candidates that cumulatively reach a specified probability threshold. Set Top P between 0.85 and 0.95 for natural results.

A value of 0.9 is recommended as your starting configuration. This allows the model to take creative turns while maintaining a natural, flowing conversational style. Top P works in conjunction with Top K to refine the selection pool of possible words at each generation step.

Higher Top P values permit more variety and risk-taking in word selection. Lower values create more predictable, safer responses. This parameter is covered during the same 1:34 to 2:07 video segment as Top K and forms the second half of your advanced vocabulary control system.

Locate the Repetition Penalty parameter in your settings panel. This control discourages the model from echoing the same phrases or sentence structures repeatedly. Set this value between 1.05 and 1.2 to prevent monotonous responses.

A value around 1.1 represents an excellent sweet spot for most use cases. This setting is firm enough to break repetitive loops without being so aggressive that it flattens the character's distinctive voice or speaking patterns. The repetition penalty directly addresses one of the most common quality issues in AI-generated conversations.

This setting is particularly important for longer conversations where repetition tends to emerge naturally. The repetition penalty is discussed at 2:07 to 2:38. After implementation, you should notice significantly less phrase recycling across multiple messages.

Find the Frequency Penalty parameter, typically located near the repetition penalty in your settings panel. This setting reduces the overuse of specific terms across the entire conversation history. Set the frequency penalty between 0.2 and 0.6 for balanced results.

Start with a value of 0.4 as your baseline configuration. This strikes a balance between curbing redundancy and preserving important contextual details that need to be referenced multiple times. The frequency penalty operates differently from repetition penalty by tracking word usage across the entire conversation rather than just recent messages.

This parameter helps maintain vocabulary diversity throughout extended chat sessions. It prevents the AI from becoming fixated on certain terms or concepts. The frequency penalty setting is covered during the 2:07 to 2:38 section alongside repetition penalty.

After adjusting all parameters, locate the Save Settings button at the bottom of your generation settings panel. Click this button to apply all your changes to the current chat session. You should see a confirmation message or visual indicator that your settings have been saved successfully.

Close the settings panel and return to your main chat interface. Send three to five test messages to your character to evaluate how the new settings affect response quality. Pay attention to creativity, coherence, repetition, and overall naturalness.

This testing phase is crucial and is covered at 2:38 to the end of the video. The AI may need a couple of conversation turns to fully settle into the new parameters. Observe whether responses feel too safe, too chaotic, or appropriately balanced before making additional adjustments.

Analyze the responses you received during your testing phase. If replies seem too safe, conservative, or repetitive, nudge the temperature value up by 0.1 increments. Alternatively, increase Top K by 10-point increments to expand vocabulary choices.

If responses become too rambling, unfocused, or incoherent, reduce the temperature value by 0.1 increments. You can also bring Top P down by 0.05 increments to tighten word selection. Make only one adjustment at a time to clearly identify which change produces the desired effect.

After each single adjustment, give the model two to three conversation turns to stabilize. This iterative refinement process ensures you land on the perfect configuration for your specific use case. The fine-tuning methodology is discussed in the final segment of the video tutorial.

What does temperature actually control?

Temperature controls the randomness of the AI's word selection at each generation step. At lower values (0.1-0.5), the model nearly always picks the most probable next word, resulting in safe but potentially boring responses. At higher values (0.9-1.5), the model samples from a broader probability distribution, creating more surprising and creative outputs but with increased risk of incoherence.

What is the difference between Top K and Top P?

Top K limits the model to considering only the K most likely next words at each step, creating a hard cutoff. Top P (nucleus sampling) instead considers the smallest set of words whose cumulative probability exceeds the P threshold, creating a dynamic selection pool that adapts to context. Both work together to constrain vocabulary while maintaining quality.

How do repetition and frequency penalties differ?

Repetition penalty examines recent tokens in the immediate context window and penalizes exact repetitions. Frequency penalty tracks how often each token has appeared throughout the entire conversation history and increasingly discourages overused terms. Repetition penalty addresses immediate loops, while frequency penalty addresses long-term vocabulary diversity.

When should I adjust these settings?

Adjust settings whenever you notice quality issues like repetitive phrasing, off-topic drift, overly conservative responses, or incoherent rambling. Different characters and conversation styles benefit from different configurations, so experimentation is encouraged. Save multiple configuration presets for different use cases if the platform supports this feature.

Below is a reference configuration that represents the recommended baseline settings from this tutorial:

TEMPERATURE: 0.8
TOP_K: 60
TOP_P: 0.9
REPETITION_PENALTY: 1.1
FREQUENCY_PENALTY: 0.4

These values provide an excellent starting point for most conversational scenarios. Adjust individual parameters based on your specific quality requirements and character personality. Remember to save your configuration after making changes and test with multiple messages before finalizing.