SimulateDevDesignDiscussions ♾️Mixlab Podcast

The display name in ComfyUI's node menu tells you exactly what this is for: "SimulateDevDesignDiscussions - Mixlab Podcast." Give it a topic, and instead of one model answering once, it simulates several AI personas discussing it back and forth, the way a small panel podcast would - the pack ships a companion "swarm-made podcast node" workflow specifically built around this node. Under the hood it's built on OpenAI's swarm, an experimental multi-agent orchestration library OpenAI released to demonstrate handing off a conversation between multiple lightweight agents rather than relying on one big prompt. This node is Mixlab's application of that idea to "have a simulated discussion about X."

That dependency is the thing to know going in: the README is explicit that this node needs swarm installed separately, and - if you want the discussion actually spoken aloud as a podcast rather than left as text - the author's own companion pack, Comfyui-ChatTTS, on top of that. Neither ships bundled with comfyui-mixlab-nodes itself. If you install the pack and this node is missing or errors immediately, that's very likely why.

The only required creative input is subject, a multiline string - the topic you want the simulated panel to discuss. model picks which LLM drives the conversation, and the list is broad: the GPT-4 family (gpt-4o is the default), older GPT-4 snapshots, and a run of non-OpenAI names - qwen-turbo, qwen-plus, qwen-long, qwen-max, glm-4, glm-3-turbo, and more beyond that. api_url is a preset picker for which endpoint to hit: openai (the default), api2d, Kimi, DeepSeek-V2, or SiliconCloud - the last few being cheaper or free Chinese-hosted alternatives, the same pattern you'll see on the pack's other GPT nodes. Optional fields round it out: api_key for whichever provider you picked, plus custom_model_name and custom_api_url if you want to point this at an endpoint that isn't in the preset list - a local LLM server, for instance. The single output is text: the simulated discussion, as a block of generated dialogue, which you'd typically hand to a TTS node (the pack's own SpeechSynthesis, or the companion ChatTTS pack the README pairs it with) to actually turn it into a spoken podcast.

Install is the pack-wide standard, plus the extra dependency this specific node needs:

cd ComfyUI/custom_nodes
git clone https://github.com/shadowcz007/comfyui-mixlab-nodes.git
cd comfyui-mixlab-nodes
install.bat
pip install git+https://github.com/openai/swarm.git

then restart. If you want the podcast-audio side of the workflow rather than just text, separately install Comfyui-ChatTTS (shadowcz007's own companion pack) the same way you installed this one.

Given the extra moving parts, expect the first hurdle to be a missing dependency rather than a bad result - a swarm import error, most likely, since it isn't a normal requirements.txt entry the pack's own installer pulls in for you. After that it behaves like the pack's other GPT nodes: wrong or expired api_key fails the call outright, and a custom_model_name that doesn't match what your chosen api_url actually serves will error the same way. Beyond that, treat the output honestly - it's LLM-generated dialogue simulating a discussion, not a transcript of real experts, so fact-check anything the "panel" states as if you'd generated it with a single prompt, because that's essentially what's happening underneath the multi-agent framing.