Openai Gpt Image 2 Text To Image (Official Edition)

The Official Edition of GPT Image 2 text-to-image is what the Channel Edition is not: explicit. Instead of leaving size to the model, it hands you width and height as integer inputs with real constraints, plus a quality picker. Same cloud model, same pack, but this is the version for when you need a specific canvas rather than "whatever the API feels like."

GPT Image is OpenAI's DALL-E successor line, and its claim to fame is text-in-image rendering - clean legible words inside generated images, the thing open models still struggle with. The Official Edition keeps that capability but makes you responsible for the dimensions, which matters when the output is going into a layout, a video frame, or any place where the aspect ratio isn't negotiable.

Why you'd reach for it

Because sometimes "1K, auto" isn't good enough. The Channel Edition's defaults are fine for a one-off, but if you're generating poster art at an exact size, or producing a batch where every image must be the same resolution, you want the controls. The schema spells out the rules, and they're worth internalizing:

• dimensions must be multiples of 16 in both directions
• maximum side length 3840 pixels
• aspect ratio 3:1 or less
• total pixels between 655,360 and 8,294,400

The defaults are 1080×1920 (portrait), and quality defaults to medium - with the natural reading that higher quality costs more per image. The tooltips are the author's own documentation here, and they're unusually precise: get a dimension wrong and the API will bounce the request.

How it works

Fill in prompt, set width and height within the constraints, pick quality, queue. The node sends the request through BizyAir to GPT Image 2, polls for the render, and returns image (IMAGE - a normal tensor you can save, preview, or pipe onward), request_id (STRING - trace/usage ID), and urls (STRING - output file location). Nothing runs locally; the only inputs you can get wrong are the numeric ones.

The optional api_config (BIZYAIR_OPENAPI_CONFIG) input overrides base_url and api_key for a single request, and skip_error makes a failed call output a placeholder instead of aborting the whole workflow.

Install and setup

Same one-time install as every node in this pack:

cd ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAirPlus.git
cd BizyAirPlus
python -m pip install -r requirements.txt

Restart ComfyUI (or install "BizyAirPlus" via ComfyUI Manager). No model files to download - the bizyair-cloudberry and bizytrd pip packages auto-install on startup. Click the BizyAirPlus button to ON, create an account at bizyair.ai, enter the key when prompted (or Settings > BizyAirPlus > API Key), and queue.

Common issues

• Request gets rejected - you broke the dimension rules. Multiple-of-16 is the one people forget first; check the constraints before blaming the node.
• Upscaling instead of generating big - if you need more than 3840px on a side, this node won't do it. Generate at the max and upscale locally; that's the workflow the constraint is quietly steering you toward.
• Quality vs cost - quality isn't free. Start at medium and only step up when the output genuinely needs it.

Standing caveat for the pack: BizyAir auto-installs and sends prompts to the cloud. Legitimate SiliconFlow product, but keep the switch OFF for local-only work.