openai / gpt-image-1 / createImage

This is the model behind the April 2025 Ghibli-pocalypse - the OpenAI image generation that flooded timelines with studio-Ghibli-style portraits, triggered by the relaxed guardrails and the 4o integration in ChatGPT. CivitaiImageGenOpenaiGptImage1CreateImage puts that same GPT Image 1 architecture inside your ComfyUI graph, hosted on Civitai's cloud. If you ever wanted that viral image model without an OpenAI account, this is the door.

What it is. One of the ~160 generated nodes in the official civitai/civitai-comfy-nodes pack (engine openai, model gpt-image-1, operation createImage). Prompt in, IMAGE out, nothing local - jobs run on Civitai's fleet, billed in Buzz. GPT Image 1's claim to fame beyond the viral moment: natively multimodal and excellent at rendering legible text inside images, which remains one of its strongest differentiators against diffusion-based generators.

How it works. Standard pack machinery: build a workflow payload, submit to orchestration.civitai.com, long-poll until terminal, and return images, errors, workflow_id, and raw_json. Required input is just prompt.

Inputs that matter. quality (auto/high/medium/low, default auto) is the cost-quality dial - high for the good stuff, medium/low to save Buzz on drafts. size (1024x1024, 1536x1024, 1024x1536, default 1024x1024) sets the canvas, with the two 1536 options for landscape and portrait. background defaults to opaque here (auto/transparent/opaque) - switch to transparent for alpha-channel assets. quantity and image_metadata (external metadata stored with the image) round out the optional set.

Getting good results. This model responds to natural language, and it's genuinely good at rendering text - "a neon sign that says 'MIDNIGHT DINER'" is a fair ask. Describe the style explicitly; the Ghibli wave happened because it absorbs style directions well.

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 → 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 limited; 30-minute cap, early-preview pack. Cloud caveats apply: your prompt leaves the machine, and OpenAI's content policies (the "baby mode content filters" the community kept complaining about) plus Civitai's own moderation gate what gets generated.

The take. The fair question is why pick GPT Image 1 when GPT Image 1.5 - the same family with better fidelity - is right next to it in the menu. Honestly, for most new projects you don't; 1.5 is the better default. But GPT Image 1 is worth knowing as the compatibility/fallback option: it's the node that matches the older API semantics, it can be cheaper on auto quality, and if you're reproducing an existing workflow built around the original model, this is the one that matches it. Also the history's kind of fun - you're generating on the model that melted the internet for a week.