Alibaba Qwen-Image 2.0 (Official Edition)

Here's the plot twist in the Qwen-Image story: the model everyone's been waiting for - Qwen-Image 2.0 - is the one Alibaba didn't open-source. It cut the 20B base down to 7B, unified generation and editing into a single model, and then launched it API-only in early 2026. No weights on Hugging Face, no local run. So if you want to drive it from ComfyUI, a BizyAir API node like this one is basically the sanctioned way in. The Official Edition means the request goes to the vendor's official endpoint (qwen-image-2.0-official/text-to-image) through the same BizyAir key you use for everything else in this pack.

How it works

Cloud execution, start to finish. Your prompt goes up to api.bizyair.ai, the job runs remotely, and the finished image is downloaded back into ComfyUI as a native IMAGE tensor. Nothing runs on your GPU, there are no checkpoints to place, and every job is metered against your BizyAir credits. The tooltip on the prompt field gives you the one real constraint up front: it supports up to 1300 tokens, so you've got room for a genuinely detailed brief.

Inputs that matter

• prompt - up to 1300 tokens of image description. 2.0 unifies gen and edit, so you can write either a fresh scene or an instruction-style brief.
• negative_prompt - up to 500 characters of what you don't want. Handy, since the original open Qwen-Image never had a strong negative-prompt story.
• size - dropdown, defaults to 2048*2048 (yes, that's big - plenty of headroom for cropping and upscale pipelines).
• n - number of images per run. The tooltip claims 1–6 but the node's slider tops out at 1 in this build; don't go hunting for a batch button that isn't there.
• seed - integer, default 0; set it to reproduce a result.
• watermark - default off. Flip on to have a Qwen-Image watermark stamped in the bottom-right corner, which some workflows want for provenance.
• prompt_extend - default on. This is smart prompt rewriting for more diverse output; switch it off if you need the model to follow your prompt to the letter.
• api_config / skip_error - the pack standard: per-request key/base override, and placeholder-on-error.

Outputs: image, request_id, urls.

Install and get going

Install via ComfyUI Manager (search "BizyAirPlus") or:

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

Restart, register at bizyair.ai, set the key in Settings > BizyAirPlus > API Key (or the BIZYAIR_API_KEY env var), and flip the BizyAirPlus action-bar toggle ON - the node stays inert while that switch is off.

Gotchas

Standard BizyAir setup failures apply: missing key, toggle off, or deps installed into the wrong Python (python -m pip show bizyair-cloudberry to verify). The node-specific trap is prompt_extend - it's great for variety but it will rewrite your careful prompt, so if your negative prompt and your composition aren't surviving contact, turn it off before you blame the model. And remember 2.0 is API-only and billed; iterate at lower resolution when you can.