Eden_GPTPromptEnhancer

If you've ever stared at a blank prompt box, Eden_GPTPromptEnhancer is the node that writes your first draft. You give it a bare description - "a beautiful landscape" - and it returns a fleshed-out version with lighting, composition, and style details, ready to feed a CLIP Text Encode. It's the same API call as Eden_gpt4_node but pointed at a prompt-engineering specialist system prompt.

How it works

The node builds a system message that casts GPT as a prompt-engineering expert, then sends it your original prompt plus whatever you put in enhancement_instructions. The default instructions are decent: add specifics about lighting, scene elements, composition and style, be creative, cap at 100 words. The temperature input (optional, default 0.7) controls how adventurous the rewrite gets - crank it toward 1 for variety, drop it to 0.2 when you want the same idea restated reliably.

Inputs that matter

• basic_prompt - your starting description.
• enhancement_instructions - the actual prompt engineering. This is where you steer: ask for a specific art style, a lens, a color grade. Treat it like an instruction block, because that's what it is.
• model - gpt-4o, gpt-4o-mini, gpt-4-turbo, gpt-3.5-turbo. Use gpt-4o-mini for batch runs; it's a fraction of the cost and fine for prompt rewriting.
• max_token (default 500) and seed, which passes through to OpenAI.

Output is a single STRING - the enhanced prompt, with no commentary around it (the system prompt explicitly forbids extra text).

Matching the node to your model family

This is where the KB's prompt-engineering advice pays off. If you're on an LLM-encoded model (Z-Image, Flux 2, Krea-style), rich prose is exactly what the encoder wants, so enhancing is a straight win. If you're on SDXL-lineage booru-tag models (Illustrious, Pony, NoobAI), a paragraph of prose can dilute tag semantics - you may prefer to keep enhancement light or have it output comma-separated tags. The enhancement_instructions box is how you encode that preference.

Setup and gotchas

Same story as every Eden GPT node: it needs OPENAI_API_KEY in a .env file in your ComfyUI root directory, and the pack installed either via ComfyUI Manager (search "Eden.art nodesuite") or:

cd ComfyUI/custom_nodes
git clone https://github.com/edenartlab/eden_comfy_pipelines
cd eden_comfy_pipelines && pip install -r requirements.txt

No key means the node returns an error string instead of crashing - wire that into a sampler and you'll see it in your image. Each run is a paid API call, so keep max_token sane. If the output ever feels repetitive across seeds, that's OpenAI's seed being a nudge, not a guarantee.