Qwen Image Text-to-Image

Qwen Image is Alibaba's image model line, and the thing most people don't realize until they try it: it prompts in Chinese and English equally well. If your workflow runs on English prompts that an SDXL-era CLIP encoder kept mangling, or you actually write in Chinese, this node quietly removes a whole category of frustration. At $0.02 an image, it's also one of the cheapest ways to get a proper MMDiT-quality render without owning the GPU.

It runs through the WaveSpeed AI API via the ERPK Collection, the community pack that wraps the service for ComfyUI. Cloud-billed, no local weights.

How it works

You type a prompt, the node sends it to WaveSpeed's Qwen Image endpoint, and ~5–8 seconds later you get an image back as a ComfyUI image IMAGE tensor - ready for Save, Preview, or more downstream nodes. No URL wrangling on your end; the node downloads the result and converts it for you.

The model combo is where the real choice hides:

• Qwen Image (20B MMDiT) - the full-size model, default.
• Qwen Image 2512 (7B) - smaller, but with better text rendering. If your prompt needs legible words in the image (signage, titles), this is the pick.

The inputs that matter

• prompt - Chinese or English. Describe the scene plainly; Qwen handles natural language well, so verbose is fine.
• model - the 20B vs 7B choice above.
• width / height - 256 to 1536, step 8, default 1024×1024. Note the ceiling: unlike Seedream in this pack, Qwen Image tops out at 1536, so don't plan 2K jobs here.
• seed - a real random seed (-1 for random). Unlike the video nodes, this one genuinely reproduces.

There are also output_format (jpeg default), enable_sync_mode (wait for completion inline), and enable_base64_output (get base64 instead of a URL under the hood) - all leave-them-alone until you have a reason.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/eRepublik-Labs/comfyui-nodes-erpk.git erpk
cd erpk
pip install -r requirements.txt

Or ComfyUI Manager → search erpk → ERPK Custom Nodes. WaveSpeed API key with credits via right-click canvas > ERPK Settings (Settings > ERPK > API Keys).

Common issues

• "Prompt is required" - the node validates before billing, so an empty box is a free error message.
• Images topping out at 1536px - that's the API's limit for this model, not a settings bug. Use Seedream for bigger canvas work.
• Text in the image comes out garbled - switch to the 7B 2512 model; that's exactly what it's tuned for.
• Cost math - $0.02/image is cheap, but it adds up during iteration. Seed locking plus small batches keeps the meter readable.

For a bilingual, cheap, "just works" text-to-image node that drops a real tensor into your graph, this is the one. When you need more resolution or typography, the Seedream V4.5 node in this same pack is the natural upgrade path.