Qwen Image 3.0 / Pro 图像生成/编辑（8 合 1）

Qwen Image 3.0 is Alibaba's image generation line, and this node collapses eight of its variants into one dropdown: domestic and global, standard and Pro, text-to-image and image-to-image. The Pro tier is the one people actually pay for - better prompt adherence and cleaner output - and having all four line/version combos in one node means you can A/B them without rebuilding the workflow. It's the pack's most flexible image generator, sitting alongside the Seedream nodes as the "I need quality now" option.

Everything here is the standard T8mars/ComfyUI_Seedance API-wrapper deal: key from api.seedance.nz, no local compute, requests only.

How it works

The node submits to the image endpoint with one of three sizing strategies you control via sizing_mode:

• auto - no size sent; the model picks.
• ratio - sends metadata.ratio plus resolution (1k or 2k).
• custom_size - sends a top-level size like 1024*1024.

The three modes are mutually exclusive request structures, and the front end only shows the fields each mode uses. For image editing, connect image1 through image3 (1–3 references, required for I2I models, ignored by T2I). The prompt_extend toggle asks the model to expand your prompt upstream, which is a surprisingly effective fix for thin prompts.

The inputs that matter

• model - the eight-way switch (qwen-image-3.0-{t2i,i2i} and -pro-, each with -global- twins).
• prompt - 5 to 2000 characters.
• negative_prompt - optional; this model actually honors it, unlike some 2026 models.
• prompt_extend - on/off prompt expansion.
• sizing_mode - auto / ratio / custom_size, and its dependent resolution, ratio, custom_size.
• n - 1 to 6 images per call.
• seed - -1 omits the seed entirely (truly random); non-negative values forward it. The tooltip says this plainly - use -1 for variety runs.
• image1..image3 - I2I references.

Outputs: image, image_url, task_id, response.

Installing it

ComfyUI Manager → search "ComfyUI Seedance", or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/ComfyUI_Seedance.git
cd ../..
python -m pip install -r custom_nodes/ComfyUI_Seedance/requirements.txt

Restart, key from api.seedance.nz/console, Seedance API Config into api_config.

Common issues

• I2I model ignores your reference - you connected image1 on a T2I model, or left it unconnected on an I2I model (I2I needs 1–3 references).
• Size controls appear but do nothing - you're in the wrong sizing_mode; each mode only reads its own fields.
• Six images all look alike - n multiplies results; without a seed variation, images in one call tend to cluster. Set n=1 and vary the seed, or accept the batch as intended.
• Prompt limit hit - 5–2000 characters is enforced; trim.

Qwen Image 3.0 is also the rare API image model that respects a negative_prompt - worth using. And if you need looser, faster, cheaper image generation for drafts, the pack's Seedance_2_5 video node or its Seedream image nodes are the other doors; Pro-tier Qwen is where you go when the deliverable matters.