Qwen Image 2.0 (BudgetPixel)

Qwen Image 2.0 is the "fast unified generation and editing" rung of Alibaba's Qwen family, and the input that makes it worth your time is negative_prompt - something the base Qwen-Image node doesn't have. You get to say what you don't want in the image, and that single field changes the character of the node: it's no longer just a text renderer, it's an editor that can fuse up to three reference images and keep the output clean. 45 credits per generation, which slots it as the mid-tier workhorse of the pack's Qwen ladder.

It's a hosted API node like everything in the pack: prompt, optional references and negative prompt go to the API, render happens server-side, and the image comes back as a core IMAGE tensor. No local GPU, no model files - the whole job happens in the cloud, which is also the point of the caveats below.

The inputs

• prompt (required) - description or edit instruction.
• negative_prompt - what to avoid. This is the headliner; use it for "no hands visible," "no watermark," "avoid extra people" type constraints.
• reference_images - up to three, for editing and multi-image fusion. Omit for text-to-image. If you're editing, set aspect_ratio to match_input_image to keep the input's frame.
• size - 1K / 1.5K (default) / 2K.
• aspect_ratio - nine options including match_input_image.
• seed - -1 for random, or set for reproducible results.
• num_images - 1–4.

Output is images into Preview Image or Save Image.

How it slots into the family

The pack's Qwen-Image ladder, cheapest to priciest: base Qwen-Image (15 credits) is text-to-image with seed control and the best typography-per-credit ratio. Qwen Image 2.0 (45) adds the negative prompt, editing and fusion. Qwen Image 2.0 Pro (85) is the premium version with enhanced text rendering and "fine-grained realism." Qwen Image 3.0 (40) is the newer balance of speed and quality. If you need editing with a negative prompt but the Pro price feels steep, 2.0 is the sensible middle - fast, multi-image capable, and it holds Qwen's text-rendering strength.

The one trap: if you feed reference_images but forget to set match_input_image, the output can come back at a different aspect ratio than your input and look like a crop. That's the most common new-user stumble, and the tooltip calls it out for a reason.

Install and issues

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

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

Restart, find it under Add Node → BudgetPixel → Image, and set your API key via BUDGETPIXEL_API_KEY or a budgetpixel.json file - no key widget, so shared workflows stay clean.

Pack-wide caveats: a recent ComfyUI for core types, and interrupting a run abandons the local wait while the server-side job keeps running and bills its 45 credits. And since the model is hosted, its content filters are baked in - a negative_prompt can steer within what Alibaba allows, not around it.