Qwen Image 2 · Text to Image

First, a warning that will save you twenty minutes: the numbers in this family are Alibaba's, and they are not in order. You'll find Qwen Image 2, Qwen Image 2.1, Qwen Image 2512, Qwen Image 3.0 and a 3.0 Pro all sitting next to each other in the SpicyAPI menu, and "2.1" being newer than "2" doesn't mean it's the same kind of model.

The distinction that matters here: Qwen Image 2 is the API-only line. The 2.1 and 2512 nodes are hosted versions of models whose weights exist - you could run them locally if your card could take it. Qwen-Image-2.0 unified generation and editing into a smaller 7B model and launched API-only, with no open weights at all. So this node isn't a convenience over a local option, it's the only way to use this model. That also explains its shape: it's deliberately smaller and cheaper than its relatives, reading prompts of up to 800 characters in Chinese or English natively and returning one image per request.

The inputs

Three of them, and that's the entire node.

• prompt - up to 800 characters, tested in both Chinese and English. Shorter budget than the 2.1/3.0 nodes, so this is not the one to use for an essay of scene description. Be dense instead: subject, composition, style, lighting.
• aspect_ratio - 1:1 (default), 4:3, 3:4, 16:9 or 9:16. Five shapes, so if you need something exotic, that's the 2.1 or 3.0 node instead.
• seed - optional, and fixed after generate by default. Same seed and settings, same result; and because every run is billed, the pack leaves it fixed so queueing the same graph twice reuses what you already paid for rather than buying a second image.

No prompt expansion, no resolution tier, no background toggle. If you're used to the 2.1 node's option list, this one feels stripped - it is.

Outputs

images and task id. IMAGE type, so Save Image is your next node, and the task id is the SpicyAPI job handle you keep for support.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/spicyapi-ai/comfyui-spicyapi

Restart ComfyUI. Nothing to download, nothing to install, no GPU involved - the pack depends on nothing beyond what ComfyUI already ships. Grab a key from spicyapi.ai/console/keys, paste it into Settings > SpicyAPI, restart once so the node list is built from your account's catalogue, and you're running. Headless machine? SPICY_API_KEY in the environment does the same job without putting the key in a browser session.

Worth knowing

Choose by schema, not by version number. The fastest way to pick between the five Qwen image nodes is to read their input lists. Long prompt and transparent backgrounds? 2.1 text-to-image. LoRA slots? One of the LoRA variants. Text rendering in signage and branding? The 3.0 line. This one is the small, fast, five-shape, 800-character option at $0.035 an image.

It's per image, so batches are where the money goes. Set Max cost per run in the SpicyAPI settings before you start experimenting - the node quotes each run before submitting and aborts above your ceiling with nothing charged. That setting is the difference between a metered API being fun and being a regret.

Cancelling stops the wait, not the job. If you interrupt, the task keeps running on SpicyAPI, is charged if it succeeds, and its result sits in your console for up to 14 days. Errors are pack-wide and code-based: no key gives you the obvious message, 40310 is an unverified email, 40004 means no deployment serves the combination you selected.