Qwen Generate/Edit Image

Alibaba's Qwen-Image family is a fascinating case in this ecosystem: the open-weight models are the community's default instruction editor, but the flagship line - Qwen-Image-2.0 - launched API-only. YogurtQwenGenerateImage is the door into that API half: it calls Alibaba's DashScope (百炼) service for text-to-image and multi-image editing, and returns proper IMAGE tensors to your graph.

The inputs, with the ones you'll actually set first:

• api_key - DashScope key. Leave blank and it tries the pack's qwen/dashscope config or the DASHSCOPE_API_KEY env var.
• model_name - defaults to qwen-image-2.0-pro. The tooltip recommends the qwen-image-2.0 series for models that support the sync interface.
• prompt - generation or editing instruction.
• system_prompt - locally prepended to the prompt (sent as a single user message).
• size - output resolution like 1024*1024, or auto to let the model decide.
• n - number of images (1–6, model-dependent cap).
• negative_prompt - reverse prompt, which most modern API image models quietly don't support well - set it, but don't bet on it.
• prompt_extend - toggle Qwen's intelligent prompt rewriting.
• watermark - whether to add the Qwen-Image watermark.
• seed, retry_count, timeout, proxy_url - the usual.

Optional: image plus image1–image4 for multi-image editing, history, and extra (raw JSON merged into the request's parameters or top level). Outputs: image, images, num_images, text, history.

The context you should bring

From the knowledge base: Qwen-Image's Edit line became the default open instruction editor (Apache 2.0, no NSFW policing), but Qwen-Image-2.0 cut the model to 7B, unified generation with editing, and went API-only. This node is that API - so you're getting the current flagship's quality without the weights, and you're paying per call. The multi-image image1–image4 inputs are the editing story: describe a change across several reference images in one prompt.

Install and keys

Ships in ComfyUI-YogurtNodes. ComfyUI Manager → search ComfyUI-YogurtNodes → install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/yogurt7771/ComfyUI-YogurtNodes.git
cd ComfyUI-YogurtNodes
pip install -r requirements.txt

Restart ComfyUI, find it under "Yogurt Nodes/LLM". Requires the openai package (DashScope's OpenAI-compatible endpoint) and a DashScope key.

Where people get burned: the base_url defaults to the Beijing region (dashscope.aliyuncs.com) - if your account is provisioned elsewhere, or you need the international endpoint, set it explicitly. And if you're coming from the open-weight Qwen-Image-Edit local workflow, remember this is a metered service: n multiplies cost, so leave it at 1 for tests.