NSWaveSpeedQwenTextToImage

Qwen-Image is Alibaba's open-weights image model - you can run it locally, and plenty of people do. This node is WaveSpeed's hosted version of its text-to-image mode, which is the choice you make when you'd rather not hold the weights or the VRAM. Same model family, zero local footprint.

It's part of the Symbiotica pack's neuralsins/WaveSpeed family. The "supports Chinese & English" note in the tooltip isn't marketing fluff - Qwen-Image's vision-language backbone genuinely handles both well, which makes this a rare node for bilingual prompting. The KB's closed-source doc notes Qwen-Image also has a reputation for excellent text rendering in images (its decoder was retuned for legible text - at some cost to fine texture, per the VAE doc).

How it works

prompt in, image (IMAGE) out, POSTed to WaveSpeed's /api/v3/wavespeed-ai/qwen-image/text-to-image. The inputs:

• prompt - the image description, English or Chinese.
• size - 7 presets from 1328x1328 (1:1) through the standard landscape/portrait ratios, default 1:1.
• custom_size (optional) - type 1920*1080 and it overrides the dropdown. Handy for a specific deliverable.
• seed - -1 for random, otherwise reproducible.
• output_format - jpeg (default), png, or webp.

enable_sync_mode defaults on, so the node waits and returns the finished tensor.

Installing

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

Restart or use ComfyUI Manager (search Symbiotica). Wire an NS WaveSpeed Client node carrying your WAVESPEED_API_KEY into client.

Common issues

• You're paying for a model you could run free. The honest tension of this node: Qwen-Image has open weights and native ComfyUI support, so a local install is a real alternative. This node wins when you lack the GPU, want zero setup, or need it in a graph right now - not on price.
• custom_size overrides the dropdown. If both are set, custom wins. Clear it for presets.
• Soft-looking output. The Qwen-Image family's decoder prioritizes text legibility over fine texture, so renders can read as clean-but-smooth. That's the model's known trade, not a node bug.
• Every run is metered. Draft locally if you have the card; keep the API for the runs that need it.

The honest take: it's the pragmatic bridge - Alibaba's capable open model without the setup. If you have the VRAM, the local version is the same model for free; this node exists for everyone who doesn't.