openai / gpt-image-2 / createImage

Here's a small tell about how fast the OpenAI menu in this pack moves: every other OpenAI create node makes you pick a size from a fixed list, but GPT Image 2 hands you width and height as plain integers. That's the newest model in the lineup, and it's also the most flexible - you're not locked to OpenAI's three preset canvases, you just type the pixels you want. CivitaiImageGenOpenaiGptImage2CreateImage is the current top of the OpenAI image stack in the civitai/civitai-comfy-nodes pack, and it shows.

What it is. One of the ~160 generated nodes wrapping Civitai's Orchestration API (engine openai, model gpt-image-2, operation createImage). Prompt in, IMAGE out, cloud compute, Buzz-metered - no local model, no OpenAI account, no API key of your own. As of this pack's current generation it's create-only; there's no GPT Image 2 edit variant in the menu yet, so pair it with the GPT Image 1.5 edit node if your workflow needs editing.

How it works. Standard pack machinery: build a workflow payload, POST to orchestration.civitai.com, long-poll until the job finishes, and get back images, errors, workflow_id, and raw_json (the full response, useful for cost checks). Required input is just prompt.

Inputs that matter. width and height (both default 1024) are the headline - you get arbitrary canvas sizes instead of a preset list, which matters for non-standard compositions and for matching a specific aspect you're compositing into. quality (low/medium/high, default high) is the cost lever; drop to medium for drafts. quantity generates multiple takes in one run. The optionals are otherwise slim: output_format and image_metadata (external metadata stored with the image). Notably absent: no background/transparency option and no mask - those live on the older nodes.

Getting good results. GPT Image models read full-sentence prompts and render in-image text well; the 2 generation builds on that. Describe scene, style, and any text you want legible, and lean on the free-form size when you need something the presets won't give you.

Install. With the pack, once:

cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt   # requests

Or ComfyUI Manager → Custom Nodes Manager → search Civitai Comfy Nodes (publisher civitai) → Install, restart, then look under Civitai/Image/openai.

Auth and gotchas. Credentials resolve from a wired Civitai Auth node, then CIVITAI_API_TOKEN, then stored OAuth. Error hints: 402 = insufficient Buzz, 401 = token expired, 429 = rate limit; 30-minute cap. Early-preview pack - and GPT Image 2 is the newest node in the OpenAI set, so expect it to be the one most likely to shift as Civitai syncs its spec. Your prompt leaves the machine, OpenAI's content policies apply, Civitai's moderation stacks on top.

The take. If you're starting fresh on OpenAI image generation in ComfyUI, this is the one: newest model, arbitrary resolution, quality control, and no legacy baggage. The main caveat is what's not here yet - editing, transparency, masking - so keep the GPT Image 1.5 nodes handy for those. For straight text-to-image, GPT Image 2 is the current best Open the pack has to offer.