Image Prompt by Seed Idea

The most common way a generation dies is a half-formed idea: "hmm, neon-lit rainy alley at night" is a vibe, not a prompt. Image Prompt by Seed Idea exists to close that gap. You type the vibe into seed_idea, pick a target model, and the node hands the job to a backend that turns it into a ready-to-use image prompt string.

Here's the catch that trips people up: despite the name, this node needs an N8N-style backend, not the LLM Backend. It POSTs {"seed_idea": ..., "model": ...} to the webhook_url on the backend it receives, then reads "prompt" from the JSON that comes back. The LLM Backend produces a dict with api_key but no webhook_url, so feeding this node an LLM backend crashes with a KeyError. Set up N8N Backend with a webhook, and let your n8n workflow do the actual idea-to-prompt transformation - you can point it at whatever image-model API you want.

The inputs that matter

• backend - the AI_BACKEND from N8N Backend. This is the route the request travels.
• model - Nano Banana (default) or GenHQ (Universal). These are just labels the pack passes through to your webhook as "nano-banana" or "genhq-universal" - so they're only meaningful if your n8n workflow understands them. Nano Banana is the community name for Google's Gemini image model; GenHQ is Runway's universal image model. Your webhook decides what they actually map to.
• seed_idea - the multiline description of what you want. Garbage in, garbage out: "moody city" gets you a worse prompt than "a detective waiting in a neon-lit alley, rain on the asphalt, late night."

The output is a single prompt (STRING) - the finished image prompt, ready to wire into a CLIP Text Encode or a sampler.

Installing

ComfyUI Manager → search "StudioDeep", or:

cd ComfyUI/custom_nodes
git clone https://github.com/studiodeep-ai/comfyui-studiodeep StudioDeep
pip install -r StudioDeep/requirements.txt

Restart and you're set. No models to download, no .env - the only real setup is getting your n8n webhook working (test it in n8n's own console first).

Troubleshooting

• KeyError: 'webhook_url' - you connected an LLM Backend. Swap it for the N8N Backend; this node only works over a webhook.
• Empty prompt output - the webhook returned JSON without a "prompt" key. The node does response.json().get("prompt", ""), so a wrong-shaped response silently gives you nothing. Check what your workflow actually returns.
• Timeouts - the request has a 30-second ceiling. If your n8n chain calls several models, that's tight. Keep the workflow lean.

It's the node you reach for when you want the human workflow (spreadsheets, queues, your own prompt logic) to own the creative direction, and ComfyUI to just consume the result.