openai / gpt-image-1.5 / editImage

Every edit node in this pack has a prompt and an image input. CivitaiImageGenOpenaiGptImage15EditImage has those plus one control nobody else offers: input_fidelity. It's a low/high switch that tells OpenAI's current image model how closely the edit should stick to your source image. That's the exact tension that defines image editing - how much do I keep vs. how much do I change - exposed as a single widget, and it makes this node the most controllable edit in the OpenAI lineup.

What it is. Engine openai, model gpt-image-1.5, operation editImage, from the official civitai/civitai-comfy-nodes pack. Required inputs: prompt and images (wire the source from a Load Image node or anywhere in your graph). The rest is optional: input_fidelity (low/high, default high), size (1024x1024, 1536x1024, 1024x1536 - default 1024x1024), quality (low/medium/high, default high), background (auto/transparent/opaque), quantity, output_format, and image_metadata.

How it works. Standard pack machinery under the hood: serialize your image into the workflow payload, POST to orchestration.civitai.com, long-poll, and get back images (the edit), errors, workflow_id, and raw_json. What matters is how you use the fidelity dial. high (default) anchors the output to your source - good for restyling or light changes where you want the original preserved. low loosens the leash - good for dramatic rewrites where the source is more of a suggestion. GPT Image 1.5's improved fidelity is the reason this control exists, and it's the lever you'll actually pull while iterating.

Inputs that matter. prompt should describe the change in full sentences ("turn this into a watercolor, keep the composition"). input_fidelity is the one to move between takes. size - the tooltip notes "auto" as an option conceptually, but the dropdown is 1024x1024 / 1536x1024 / 1024x1536; match it reasonably close to your source. quantity gives you several candidate edits in one run, cheap insurance for fiddly changes.

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

Or ComfyUI Manager → Custom Nodes Manager → search Civitai Comfy Nodes → Install, restart, find it under Civitai/Image/openai.

Auth and gotchas. Credentials from a wired Civitai Auth node, then CIVITAI_API_TOKEN, then stored OAuth. The usual error hints: 402 = insufficient Buzz, 401 = token expired, 429 = rate limit; 30-minute job cap and an early-preview pack that can shift shapes between releases. Standard cloud caveats: your source image and prompt leave the machine, OpenAI's content policies apply, Civitai's moderation on top.

The take. If your cloud editing work is OpenAI-based, this is the node - the fidelity dial alone beats every other OpenAI edit option in the pack, which either lacks the control (GPT Image 1) or is generationally older (DALL-E 2). For edits that must stay close to source, set high and be specific in the prompt; for bold rewrites, drop to low and let the model run. The one switch covers the whole spectrum, and that's more than the rest of the menu can say.