Image Advanced Parameters

If you're a ComfyUI person, this node's fields look like home: width, height, guidance scale, steps, seed, sampler, negative prompt. These are the knobs you'd expect on a local KSampler. So here's the honest headline before you wire it in: OpenAI's image endpoint ignores all of them. This node exists for the OpenAI-compatible providers that accept diffusion-style parameters - Venice's flux-dev is the README's own example - and if you point it at DALL-E and wonder why nothing changed, now you know.

How it works

It's a pure parameter builder, like its text sibling. Every field is compared to its default and only non-default values get emitted into the params_json string, which you feed into Generate Image's advanced_params_json input. One thing worth noting: the node doesn't just send your values, it also emits common alias names alongside them - guidance_scale comes with cfg_scale, steps with num_inference_steps, sampler with scheduler. That's the author hedging, because compatible providers don't agree on what the parameter is called. Slightly ugly, but genuinely helpful for hitting an endpoint that wants the other name.

The inputs

• width / height (256–2048, step 64, default 1024) - custom dimensions. Only sent if you change them from 1024.
• negative_prompt - what you don't want in the image. Meaningless to DALL-E, meaningful to diffusion-style providers.
• guidance_scale (1–20, default 7.5) - prompt adherence vs. creativity. The classic CFG tradeoff.
• steps (1–150, default 50) - sampling iterations. Fewer is faster and rougher.
• seed - set ≥ 0 for reproducibility, −1 for random.
• sampler - euler, euler_a, ddim, ddpm, dpm++, or none (which sends nothing).

Wiring it up

Connect params_json to Generate Image's advanced_params_json input. That's the whole integration - the generation node JSON-parses it and merges it into the request. Disconnect it and Generate Image runs on its own defaults, none the wiser.

Installing it

Part of the ComfyUI-AI-CustomURL pack:

cd ComfyUI/custom_nodes
git clone https://github.com/bowtiedbluefin/ComfyUI-AI-CustomURL
cd ComfyUI-AI-CustomURL
pip install -r requirements.txt

Or ComfyUI Manager → search "AI CustomURL", restart. API-only pack, no model downloads.

The honest take

There are two ways to use this node and they're very different. On a provider that accepts diffusion-style params, it's genuinely useful - it's the only place in the pack you get negative prompts and CFG for hosted image models. On plain OpenAI, it's a no-op dressed up in familiar clothes. Read your provider's docs before you build a workflow around it, and if you're strictly on DALL-E, you can skip it entirely.

Like the rest of the pack it's a small single-author project with little community presence - which also means don't expect the parameter names to match every provider's docs out of the box. When one rejects an unknown field, response_json on the generation node will tell you exactly which.