SFWaveSpeedQwenTextToImage

Qwen-Image is Alibaba's answer to the modern text-to-image generation - strong at text rendering, good at following detailed prompts, and notably comfortable with Chinese. SF WaveSpeed Qwen Text to Image is the no-fuss wrapper: prompt in, image out, with the Qwen canvas-size system built in. It's the node you grab when you want Qwen's particular look without downloading its (large, encoder-heavy) weights.

If you've read up on Qwen-Image, you know the model expects its specific native resolutions - and this node bakes that in with a size dropdown rather than making you memorize pixel values. That alone saves a round of confused troubleshooting.

It's part of the comfyui-stillfront pack, an API call via WaveSpeed - no local weights, no GPU requirement.

How it works

The node sends your prompt and chosen size to WaveSpeed's Qwen-Image text-to-image endpoint, waits for the result, downloads it, and returns an IMAGE tensor. On the other side of that call is a 20B-class model that's genuinely good at putting legible text into images and understanding long, detailed prompts.

The inputs that matter:

• prompt - supports Chinese and English; Qwen-Image handles both well, and detailed prompts reward you.
• size - the Qwen native canvas presets: 1328x1328 (1:1) (default), 1536x864 (16:9), 864x1536 (9:16), 1472x1104 (4:3), 1104x1472 (3:4), 1536x1024 (3:2), 1024x1536 (2:3).
• custom_size - optional width*height (e.g. 1920*1080) that overrides the dropdown. This is where the model's resolution sensitivity lives: straying from the native sizes can hurt output, so only override when you know why.
• seed - -1 for random, or lock it.
• output_format - jpeg/png/webp.
• enable_sync_mode - default on.

Output is a single image (IMAGE).

Installing it

Pack install plus WaveSpeed key:

cd ComfyUI/custom_nodes
git clone https://github.com/razvanmatei-sf/comfyui-stillfront
cd comfyui-stillfront
pip install -r requirements.txt

Restart, then wire in an SF WaveSpeed Client node with your key (or config.ini / WAVESPEED_API_KEY).

Troubleshooting

• Images look soft or compositionally off at odd sizes. You used custom_size with a non-native value. Qwen-Image was trained on its native canvases; stick to the presets unless you need the exact pixel count.
• Text in the image still garbled. Qwen is one of the better models at in-image text, but short prompts with lots of words ("a menu with these 20 items") will still trip it. Reduce the amount of rendered text.
• "API key is empty." Client node first, then config.ini, then the WAVESPEED_API_KEY env var - in that order.
• Randomness. Default seed is -1, so every run differs. If you're iterating toward a specific look, lock the seed once you find one you like, then vary the prompt.

Where does this sit in the pack? It's the plainest Qwen path - no LoRAs, no references, no editing. For text-to-image with Qwen's text legibility, it's the one to start with. Add the LoRA variant when you want a style, and the Edit nodes when you want to change an existing image.