Concurrent Submit | Zhenzhen_qwen_image

This is the concurrent twin of Zhenzhen's Comfly_qwen_image node - Alibaba's Qwen image model, sold by the reseller at a price the README brags is "全网最低" (lowest on the web). It's a straightforward text-to-image generator with a genuinely useful party trick: you can set an arbitrary custom resolution, which most of these closed API models won't let you do.

The Concurrent part means it doesn't sit there blocking your graph while the cloud works. You feed it a prompt, it hands back a COMFLY_IMAGE_FUTURE task, and the real generation runs in the shared image pool (30 workers). Finish wiring, then collect everything at once with Concurrent Collect Images (ComflyConcurrent_Image_Await) - up to 30 tasks, returned as image_1…image_30 in slot order. Batch of twenty prompts? One submit node, one collect node, and the graph is never idle-waiting on call #3.

The inputs

• prompt - the text prompt. Multiline.
• size - preset resolutions: 512x512, 1024x1024, 768x1024, 576x1024, 1024x768, 1024x576, or Custom. That last one is the selling point.
• Custom_size - when size is Custom, type your own, e.g. 1280x720. Qwen's API accepts a real arbitrary resolution here, which is rare among the closed image models.
• model - locked to qwen-image.
• num_images - 1 to 4 per call. Billing is per image; the README is explicit that 4 images costs 4× the API spend.
• num_inference_steps - 2–50, default 30. More steps, more cost/time, marginal quality gains past ~30.
• guidance_scale - 0–20, default 2.5. Push up for stricter prompt adherence.
• negative_prompt - actually supported here; use it.
• enable_safety_checker, output_format (png/jpeg), seed, api_key, skip_error - the rest of the panel.

Output: one task of type COMFLY_IMAGE_FUTURE. Remember, no Await node = no image.

Install

Pack-standard. ComfyUI Manager → search Comfyui-zhenzhen, or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

Restart, paste a Zhenzhen token into api_key (ai.t8star.org overseas; api.seedance.nz for the domestic discount). No weights download - it's a paid API reseller, and your prompts leave the machine by design.

Gotchas

• Custom size isn't always honored by the API - that's a Qwen quirk, not this node. If the output comes back at a preset resolution, your custom value was likely outside what the endpoint accepted.
• Steps and cost. At 4 images × 50 steps, you're burning the cheap price into a not-so-cheap bill. Keep steps around 30 unless you've tested that more actually helps.
• Qwen ≠ Google. If you're comparing outputs to Nano Banana, expect different aesthetics and weaker text rendering; Qwen's edge here is price and resolution flexibility, not leaderboard-topping quality.