ComfyAI FAQ: Everything You Need to Know

ComfyAI offers two models, each optimized for different needs:

• 35B model — Fast responses, great for casual conversation, quick questions, and light roleplay. Use this when you want snappy, natural back-and-forth.
• 122B model — Slower but significantly deeper reasoning. Better for complex roleplay, creative writing, nuanced emotional conversations, and detailed worldbuilding.

Rule of thumb: Start with 35B. If you need more depth, switch to 122B. You can switch anytime without losing context.

Use 35B for:

• Casual conversation and chatting
• Quick questions and answers
• Light roleplay and character interaction
• When response speed matters most

Use 122B for:

• Deep, immersive roleplay with complex characters
• Creative writing and story development
• Detailed worldbuilding and lore consistency
• Nuanced emotional conversations
• When quality matters more than speed

Different models have different training data, architectures, and capabilities. The 35B model is optimized for speed and natural conversation flow, while the 122B model has deeper reasoning and more nuanced understanding.

The same character prompt will produce slightly different results because each model interprets and generates text differently. Think of it as two different actors playing the same role — both are playing the character, but with their own style and interpretation.

This is completely normal. If a character feels "off" on one model, try the other — you might prefer how one model handles that specific character.

Yes, you can switch between 35B and 122B at any time. Your conversation history and memory carry over completely.

The AI may respond slightly differently after switching since the models have different strengths, but the character, context, and everything you've discussed remains intact. It's like changing the actor mid-scene — the story continues, just with a different delivery style.

ComfyAI offers multiple chat modes, each designed for a different type of experience:

• Casual Mode — Natural, conversational responses. Best for everyday chatting and getting to know characters.
• Roleplay Mode — Emphasizes character consistency, immersive storytelling, and in-character responses. Best for fictional scenarios and character interaction.
• Creative Mode — Encourages more elaborate, imaginative, and detailed responses. Best for creative writing and worldbuilding.

Each mode adjusts the AI's tone, response length, and style to match what you're trying to do.

Each chat mode uses different system prompts and response styles:

• Casual Mode keeps things conversational and natural — like texting a friend. Responses are shorter and more direct.
• Roleplay Mode emphasizes character consistency and immersive storytelling. The AI stays in character, uses descriptive narration, and builds on the scene.
• Creative Mode encourages more elaborate and imaginative responses. Longer descriptions, richer language, more creative exploration.

The AI adapts its tone, length, and style based on the selected mode. You can switch modes at any time.

Casual Mode: Quick chats, checking in with a character, casual conversation, when you want fast and natural responses.

Roleplay Mode: Running a story, character interaction, D&D campaigns, collaborative fiction, when you want the AI to stay in character and build scenes.

Creative Mode: Writing assistance, brainstorming, worldbuilding, when you want rich descriptions and imaginative output.

Tip: If you're not sure, start with Casual. You'll quickly notice when you need more depth — that's when to switch.

Character drift is normal over long conversations. Here's what helps:

• Remind naturally: Restate key character traits within the conversation rather than giving meta-instructions. Instead of "stay in character," try acting in a way that prompts the character to respond in character.
• Switch to 122B: The larger model has better character consistency, especially for complex personalities.
• Start fresh: If a conversation has gone too long, start a new chat with a more detailed character description.
• Be specific: "Remember, you're a sarcastic detective who hates mornings" works better than "be more in character."
• Use memory: Make sure you're logged in so the AI can use persistent memory across sessions.

The safety filter can occasionally flag benign content due to keyword patterns or context misinterpretation. This is rare but happens when certain word combinations trigger the filter.

What to do: Try rephrasing your message slightly. Often a small wording change is enough to avoid the false positive. The system is continuously improving to reduce these occurrences.

If you consistently get false positives with a specific type of content, try switching chat modes or models — they handle filtering slightly differently.

Within a single chat session, the AI remembers the full conversation. If you start a new chat, the AI won't remember previous conversations unless you're logged in and using the memory feature.

To maintain continuity:

• Create an account to enable persistent memory
• Reference previous conversations naturally ("like we talked about before...")
• Use the same character description for consistency
• The 122B model generally has better long-term context retention

1. Be specific in character descriptions — The more detail you provide, the better the AI can embody the character. Include personality traits, speech patterns, and backstory.

2. Use the right model — 35B for speed, 122B for depth. Don't be afraid to switch based on what you're doing.

3. Give context naturally — Instead of meta-instructions like "be more creative," weave context into the conversation: "The rain was pouring as you stepped into the dimly lit tavern..."

4. Use the memory feature — It makes conversations feel continuous and the AI remembers important details about you and your preferences.

5. Correct the AI — If it drifts, gently steer it back. The AI learns from your feedback within the session.

A good character description includes:

• Name and identity — Who are they?
• Personality traits — 3-5 key characteristics (e.g., "sarcastic but loyal, hates mornings, secretly soft")
• Speech patterns — How do they talk? Formal? Casual? Uses slang? Speaks in short sentences?
• Background — Brief context about their world and situation
• Relationship to you — How do they know your character? Friends? Strangers? Rivals?

Example: "You're Marcus, a 30-something detective who's seen too much. You're sarcastic but deeply loyal. You speak in short, direct sentences. You hate mornings and always have a coffee in hand. We're partners on the same case, and you've been skeptical of me since day one."

Absolutely. Many users use ComfyAI for:

• Brainstorming — "What if the villain was actually the hero's sibling?"
• Dialogue practice — Have the AI play different characters to test conversations
• Scene drafting — "Write the opening of a mystery set in a small coastal town"
• Plot troubleshooting — "My second act feels slow, what could I add?"
• Worldbuilding — "Help me design the magic system for my fantasy world"

Use the 122B model and Creative Mode for the best writing assistance. The AI is a collaborator, not a replacement — it helps you explore ideas, not write your entire manuscript.

No, you can start chatting without an account. Just open the site and start talking.

Creating an account gives you:

• Persistent memory across sessions
• Conversation history
• Ability to save and reuse characters
• Personalized experience

The core features (chat, memory, both models) are free regardless of whether you have an account.

Yes. Core features are completely free:

• Access to both 35B and 122B models
• Persistent memory
• Self-prompting
• All chat modes
• Unlimited conversations

Optional premium features are available via Patreon subscription for users who want to support the project and get additional perks. But the core experience is and will remain free.

Your conversations are not used for training. ComfyAI is hosted in Austria under EU law. Your data stays yours — we don't sell or share it with third parties.

Key privacy facts:

• Hosted in Austria (EU jurisdiction)
• No US data transfer
• Conversations not used for model training
• No data sold to third parties