GPT Image

Here's the shortest useful article in this pack: RightCodesGPTImage - displayed in the menu as "GPT Image" - is not a different node. In the pack source it's literally one line: RightCodesGPTImage = GPTImage. Same class, same inputs, same outputs, same code path. The only difference is the registration name and display title. It's a legacy alias, kept around so older workflows that reference the "GPT Image" name keep loading instead of breaking.

So if you've read the GPT Image Generate article, you already know everything about this one. I'll keep this short and point at the parts that trip people up.

What it is

A streaming /v1/chat/completions client for the GPT Image / Nano Banana family of closed image models - gpt-image-2-vip, gpt-image-2, nano-banana, nano-banana-2, nano-banana-pro - reached through whatever relay you point base_url at. It streams the response so long generations don't die to Cloudflare-style timeouts, converts the returned image (direct URL, b64_json, or markdown reference) into an IMAGE tensor, and passes a concrete size when both resolution and aspect ratio are explicit:

4K + 16:9 -> 3840x2160
2K + 1:1  -> 2048x2048
1K + 9:16 -> 720x1280

The inputs and outputs, same as GPTImage

api_key, base_url, model (dropdown with the five models), prompt, resolution (default 1K), aspect_ratio (default 1:1), timeout_seconds (default 600). Optional image_1 / image_2 become reference images. One image output.

The one thing to remember is the per-model resolution cap, because it's the most common error people hit on this node: gpt-image-2 and nano-banana only allow 1K. Ask for 2K or 4K and it throws on purpose - it's protecting you from a request that would fail or waste credits.

Install

Same pack, same steps:

cd ComfyUI/custom_nodes
git clone https://github.com/Marigold1122/Chat-Image-Bridge-Node.git ComfyUI-ChatImageBridge
pip install -r ComfyUI-ChatImageBridge/requirements.txt

Restart ComfyUI, then find it under api -> Chat Image Bridge -> GPT Image (ComfyUI Manager: search "Chat-Image-Bridge-Node"). Just requests for a dependency.

When to use which name

If your menu shows both "GPT Image" and "GPT Image Generate", don't second-guess it - they're the same node under two names. The advice from the pack's own docs is to use the newer GPTImage ("GPT Image Generate") going forward. If you load someone else's workflow and it errors on a missing "GPT Image" node, this alias is exactly why it resolves: the pack registers both names, so the old one still maps. Pick one, be consistent, and move on.

Nothing here is a model you download or a GPU you burn - it's an API client. Your API key goes in the node, the request goes to your relay, and the generated image comes back as a standard ComfyUI tensor you can wire straight into Save Image.