Concurrent Submit | zhenzhen-mai-image-2.5-fal

MAI Image 2.5 is Microsoft's closed image model, and this node is the text-to-image flavor of it hosted on fal through the Zhenzhen reseller. The text-to-image variant is the simplest node in the whole pack's image family: a prompt, a ratio, a count, done. Its whole reason to exist as a Concurrent Submit node is volume - when you're iterating on concept art or a character sheet, being able to queue thirty prompts into the shared image pool and collect them as one batch is the difference between a night of babysitting and a single click.

The inputs that matter

• prompt - required, the only required thing.
• num_images - 1 to 4 per job. Remember this is charged per image, so 4 is 4 billings.
• aspect_ratio - auto plus 1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3. Good coverage for posters vs social.
• output_format - png / jpeg / webp.
• poll_interval / max_poll_attempts - fal polling control.
• api_key - direct key paste.
• seed - ComfyUI cache-control seed (fixed reuses the cached run; randomize forces a new one). Not a native MAI parameter.
• skip_error - on, and failures return an empty placeholder instead of killing the batch.

That's the whole list. No style slot, no references - this is the plain generator, and for once "just a thin wrapper" is a feature.

The concurrent flow

Output is a task (COMFLY_IMAGE_FUTURE), not an image. The pattern is the pack's standard one: line up a stack of these Submit nodes with different prompts (or different seeds), wire every task into ComflyConcurrent_Image_Await (30 slots), and the collector waits for all of them, returns the images in slot order, and gives you a status JSON. Set the collector's failure_mode to placeholder if you'd rather a dead slot return an empty image than throw the batch.

Install and gotchas

It's part of T8mars/Comfyui-zhenzhen - ComfyUI Manager → search "Comfyui-zhenzhen", or git clone https://github.com/T8mars/Comfyui-zhenzhen into custom_nodes, then restart. Nothing local to download; standard deps. Paid Zhenzhen token, overseas servers (proxy needed from China), fal pre-charge-then-settle billing. And because the whole point is running a lot of these, keep an eye on the wallet - thirty parallel generations is a visible line item even at reseller prices.