GPT Image 1.5

GPT Image 1.5 is the "wait, it got better" entry in OpenAI's image lineup - the December 2025 refresh that improves fidelity over the original GPT Image 1. In the Runware pack it's a near-identical node with the newer model behind it: same required inputs, same IMAGE output, same wiring. If you already know the GPT Image 1 node, you know this one. The question is only whether the newer model is worth the per-image premium, and for text-heavy or detail-critical work, it usually is.

Both nodes live under Runware/Image/openai, which makes the whole version question moot in practice: drop this node in, and Runware routes the request to the current GPT Image 1.5 backend. The KB's closed-source-models essay frames the broader arc - these closed image models keep trading quality against censorship and cost, and the community keeps measuring the gap against open FLUX. This node is squarely on the "closed, costs money, renders text beautifully" side of that trade.

What you actually set

Identical to GPT Image 1:

• positivePrompt - required, and this model rewards literal detail, especially on any in-image text.
• width / height - default 1024×1024, set them to the aspect you want.
• referenceImages - IMAGE socket for image-to-image, since the model is natively multimodal.
• providerSettings.openai.quality - auto, high, medium, low. high for the kind of fine detail where 1.5's upgrade shows.
• providerSettings.openai.background - auto, opaque, transparent. Transparent is a quiet workhorse option for compositing.
• providerSettings.openai.moderation - auto or low. Auto runs the content filter; low loosens it if the filter is being annoying about a legit prompt.
• numberResults - up to 4 seeded variations.

The one output, image (IMAGE), wires into any Preview/Save node like a local generation. Whole workflow is two nodes.

Install

Same as every node in the pack. ComfyUI Manager → search Runware → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

Then a Runware API key (runware.ai/api-keys) into ComfyUI Settings → Runware API key or RUNWARE_API_KEY. No models to download - the pack's whole pitch is that the cloud holds the weights.

Where people get burned

• Same moderation as its sibling. moderation: auto means OpenAI-grade filtering on a prompt that sailed through locally. Expect the occasional polite refusal.
• Cost. 1.5 is the premium tier of an already paid service. quality: high plus multiple results is where the title-bar cost readout starts to sting. Check it.
• Don't expect a different node experience. If 1.5's output isn't beating 1 for your use case, there's no hidden feature here - it's the same knobs with a newer brain.

Bottom line: if you already pay for the pack and text fidelity matters, reach for 1.5 over 1. If you're on a budget and the prompt is simple, the older node will do and costs less.