zhenzhen-qwen-image-3.0-lowprice

Alibaba's Qwen image models are the rare thing in this corner of the hobby: a closed model you can actually afford to play with. Comfly_qwen_image_3_0_lowprice is the zhenzhen pack's gateway to all eight of the Qwen Image 3.0 variants - text-to-image and image-to-image, pro and non-pro, domestic and "global" - through the domestic low-price shop, and it drops the result back as a real IMAGE tensor. If you've been watching the Qwen-Image editing demos and wishing you could use them without renting a cloud GPU, this is the node that gets you there.

The "lowprice" family means it doesn't take an api_key field of its own. You wire an api_config into it from the pack's Comfly_seedance2_low_price_settings node, which points at https://api.seedance.nz and carries your token. One settings node can feed every low-price node in a workflow, which is the intended pattern.

How it works

Async task submission to the shop: your prompt (and optional reference images) get uploaded, the job runs server-side, and the node polls until it's done, downloads the image, and converts it to a ComfyUI tensor. Outputs are image, image_url, task_id, and the raw response.

The inputs that matter

• model - eight choices: qwen-image-3.0-t2i, qwen-image-3.0-i2i, plus -pro- and -global- variants of each. Pro is the higher-quality tier; global routes through the international endpoint. Pick the t2i model for generation, i2i for editing.
• prompt and negative_prompt - the standard pair.
• prompt_extend - default on; the model rewrites your prompt into something richer before generating. Turn it off if you want literal control.
• sizing_mode - auto, ratio, or custom_size. Keep it on auto until you have a reason not to.
• resolution - 1k or 2k.
• ratio - the usual aspect ratios.
• n - how many images, 1–6. Remember this is metered per image; each one costs.
• image1–image3 - up to three reference images for the i2i models. Wire your source image(s) in here for editing.

The seed is cache-control only (it isn't sent to the API - fixed reuses the cached result, randomize forces a fresh run), so don't expect seed-faithful regeneration like you'd get from a local sampler.

Installing it

Standard zhenzhen install: ComfyUI Manager (search "Comfyui-zhenzhen") or

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

restart, and you're done. No model files, no VRAM. You need a token from https://api.seedance.nz in the settings node.

Common issues

• Everything outputs a blank image - skip_error is probably on, which silently swallows failures. Turn it off while you debug.
• n is high and the bill jumps - the API charges per generated image, not per call. The README hammers this: more images, more credits.
• Editing doesn't take - double-check you selected an i2i model, not a t2i one; the t2i models ignore reference images.

It's not a substitute for a good local checkpoint if you have the VRAM and want full control - but the entire value proposition here is the closed-model quality on a pay-per-image basis with zero setup. For a beginner who googled "Qwen image in ComfyUI," this is the fastest door into that model family.