Concurrent Submit | zhenzhen-wan-2.7-global-image-lowprice

Wan 2.7 is API-only - Alibaba stopped shipping open weights after Wan 2.2, so the "global image" models here exist only as cloud calls. This node is the low-price domestic channel's door into Wan 2.7's image stack: text-to-image (wan-2.7-global-t2i) plus two image-editing models (wan-2.7-global-i2i and the pro variant), with up to nine reference images for edits. And it's an image-pool node, which is the headline: it runs in the pack's thirty-worker image pool, so you can queue thirty Wan 2.7 generations and have them all drain in parallel. That's the whole reason to pick the concurrent version over the original.

Inputs that matter

• model - wan-2.7-global-t2i (default), wan-2.7-global-i2i, or wan-2.7-global-i2i-pro.
• prompt - text or edit instruction.
• width / height - 512 to 4096 in steps of 8, both defaulting to 1024. Note the step size: 1024 is fine, 1000 will quietly round or fail - stick to multiples of 8.
• thinking_mode - on by default. The model reasons about the prompt before generating, which usually means better prompt adherence at a small speed cost.
• image1–image9 - references for the i2i modes. Nine slots, so this covers heavy multi-reference edits.
• api_config - this is a low-price channel node: no bare apikey. Connect the pack's zhenzhen-seedance2-low-price-api-settings node, or set SEEDANCE_API_KEY / SEEDANCE_BASE_URL in the environment.
• seed, skip_error - standard. (The seed is a cache-control compatibility seed, per the tooltip.)

Output: a COMFLY_IMAGE_FUTURE task.

How it works

The one thing to get right: this is image-pooled, so you collect it in Concurrent Collect Images (ComflyConcurrent_Image_Await), which has thirty sockets - not the ten-slot video collector. Wire an image task into the video node and it refuses. Under ComflyConcurrent.py the wrapper validates, pushes the real Comfly_wan_2_7_global_image_lowprice call onto the image executor (default 30 workers, COMFLY_IMAGE_CONCURRENCY to change), and returns the future immediately. The collector waits on all tasks, restores slot order, and reports per-slot status; failure_mode: placeholder converts failures to blank images.

Setup

Install the pack once - ComfyUI Manager (search "Comfyui-zhenzhen"), or:

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

No weights. Key it through the low-price settings node or SEEDANCE_API_KEY - the domestic channel (api.seedance.nz) is the discount, roughly 75–80% of overseas pricing, and it's why this node exists.

Gotchas

Two traps. First, the dimensions: multiples of 8 only, and the upper end (4096) is where prices and failure rates climb - 1024–2048 is the sane zone for most work. Second, the key: this node uses the low-price channel config, not your main Zhenzhen key; a rejected key almost always means you connected the wrong settings node. With a 30-wide image pool, a careless batch is thirty paid calls before you see one result - do one, check the cost, then go wide. Timeouts: check your async task page before assuming the job died.