并发提交｜Wan 2.7 海外图像生成/编辑（3 合 1）

This is the 并发提交 (concurrent submit) version of the Wan 2.7 Global image node - a three-in-one covering wan-2.7-global-t2i, wan-2.7-global-i2i, and wan-2.7-global-i2i-pro, all on the pack's overseas route. Wan 2.7 has no open weights (Alibaba stopped shipping after Wan 2.2), so like everything in this pack you're calling it through the api.seedance.nz gateway. And "Global" means the overseas infra, which for image tasks is the looser-audit escape hatch the pack's README is honest about.

What sets this node apart from the pack's other image nodes: thinking_mode. It's a T2I-only toggle that enables the upstream reasoning pass - the model mulls over your prompt before rendering, which is the same "thinking" behavior you see in the closed frontier image models. Leave it on for quality, flip it off when you want cheaper/faster drafts.

The inputs split by model, which is worth knowing before you're confused:

• model - wan-2.7-global-t2i, wan-2.7-global-i2i, or wan-2.7-global-i2i-pro. Default is T2I.
• prompt - up to 5,000 chars for T2I, 2,048 for I2I. The node validates the limit against the selected model.
• width / height - T2I only, 512–4096, stepped by 8. Ignored by the edit models.
• thinking_mode - T2I only, default on.
• image1 … image9 - I2I only, and I2I requires 1 to 9 of them. Ignored by T2I. The i2i-pro variant is the higher-grade edit path.

Because it's a concurrent submit node, its single output is future (type SEEDANCE_IMAGE_FUTURE), not an image. You connect that into 并发接收图片（30 路） alongside up to 29 other image futures and the receiver waits on all of them in parallel and hands back the finished images in slot order.

How the concurrent part works

Every node with 并发提交｜ in its name keeps the same inputs as its base node but doesn't render anything itself - it validates your settings synchronously, then drops the job into a background thread pool and hands you a handle. The pack runs separate pools for images (up to 30 workers) and videos (up to 10); you can shrink them via SEEDANCE_IMAGE_CONCURRENCY / SEEDANCE_VIDEO_CONCURRENCY if you're batching politely. Per-slot skip_error keeps one bad prompt from sinking the whole batch, and the receiver's failure_mode (raise vs placeholder) decides whether a failure stops everything or leaves a placeholder in that slot. You can even mix futures from different nodes into one receiver - a Wan image future in slot 1, a Seedream future in slot 2, whatever.

Installing it

Pack install as always: ComfyUI Manager → search ComfyUI Seedance, or:

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

Restart, look under Seedance/并发提交. Nothing to download - requests and an api.seedance.nz key is the whole story.

Where people get burned

• T2I vs I2I confusion. Width/height and thinking do nothing on edit models; reference images do nothing on T2I. The node hides nothing, so you have to remember which fields apply to your selected model.
• The future output isn't an image. Wire it to the receiver node, not to Save Image. This trips up everyone once.
• Batching is metered per call. Ten images in parallel is ten paid renders. Use n-style batching on the cheap models for drafts and save parallel runs for when you actually need variety.

Wan 2.7's reputation in the community is mixed - great fine detail in close-ups, weaker on heavy motion - but that's for video; as a still-image generator via this node it's solid, and thinking_mode gives you a genuine quality lever that most of this pack's image nodes lack.