GPT-Image-2

The README for this pack doesn't mince words about its own GPT Image 2 node: "效果非常不错，但是价格也是真的不便宜" - the results are genuinely great, but the price is really not cheap. That's the honest summary. GPT Image is OpenAI's closed image model - strong text rendering, strong instruction following, real reasoning about what you asked - and you can't run it locally, so if you want it in a ComfyUI graph, this node is how. Just go in with your eyes open on the meter.

How it works

Standard wrapper: your prompt and optional input_images (plus context) get POSTed to the pack gateway under the model id GPT-image-2. Because GPT Image is a multimodal model, it can do more than generate from text - feed it reference images and it'll edit and combine them, and like the Gemini nodes it returns both an image and a text output (the model can describe or reason about what it made), plus a context port for multi-turn work. The model dropdown has exactly one entry today (GPT-image-2), so that's not really a choice - the real controls are quality and resolution.

The inputs that matter

• prompt (STRING) - optional but obviously central. GPT Image follows detailed, natural-language instructions well; the model's "reasoning" is exactly where it shines for complex compositions.
• quality (auto/low/medium/high) - the direct cost lever. High quality on a hard prompt is where the "not cheap" kicks in.
• resolution (1K/2K/4K) - another cost lever, plus a hard quality cap. The difference between 1K and 4K here isn't just pixels; it's price per image.
• size (eleven aspect ratios, from 1:1 through 1:3/3:1 to 9:16/16:9) - output framing.
• seed (INT) - reproducibility.

Optional: prompt, input_images (IMAGE batch), context (ANY). Outputs: image (IMAGE), text (STRING), context (ANY).

Installing it

Part of comfyui-MJAPI-party. ComfyUI Manager → "mojieapi_party" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/MoJIeAIGC/comfyui-MJAPI-party.git

Register at mojieaigc.com, put your key in config.ini (leave BASE_URL = https://www.mojieaigc.com/v1/completions as shipped), and restart:

[API]
KEY = your_api_key
BASE_URL = https://www.mojieaigc.com/v1/completions

No models to download; standard pack dependencies only.

Common issues

• The bill, again. Cost is the single most-reported friction for GPT Image via third-party resellers - the pack's author says it outright. Draft at 1K/auto quality, and only spend on 4K/high for the final asset.
• Moderation is OpenAI's, not yours. GPT Image refuses what it refuses; there's no node setting that changes that, and reseller "looser filtering" claims should be treated as marketing until proven. If the job needs something a hosted filter won't do, the local path is the only path.
• Text rendering is good but not infallible. It's the best-in-class for typography among the big image models, but complex multilingual text can still garble - spot-check any text you actually need to read.
• Error image with text - the pack's standard failure display (key, balance, timeout).