Concurrent Submit | Zhenzhen_nano_banana_edit

This is the concurrent twin of Zhenzhen's first-gen Nano Banana edit node - the one that existed before Nano Banana 2 and Pro took over. It's not the newest toy, but it's still the cheapest way to push a batch of image edits at Google's gemini-2.5-flash image family through the reseller, and for simple swaps - "change the shirt to red," "put it on a beach" - the first Banana still holds up fine.

What the "Concurrent Submit" part buys you: the plain Comfly_nano_banana_edit blocks the whole graph until the cloud answers. This one drops the request into a background pool of 30 image workers and returns a COMFLY_IMAGE_FUTURE task immediately. You then feed the task into Concurrent Collect Images (ComflyConcurrent_Image_Await), which waits on up to 30 tasks and hands back image_1…image_30 in order. Graph stays responsive, all 30 jobs run in parallel, and you get a status string per slot to see who died.

The inputs

• prompt - what to generate or edit. Multiline.
• mode - text2img or img2img. One honest caveat from the pack's own changelog: the size/ratio selector only actually applies in text2img mode. In img2img the model follows the reference image, so don't fight it.
• model - nano-banana or nano-banana-hd.
• aspect_ratio - 1:1 default, plus 16:9, 4:3, 9:16, 21:9, 9:21 and the rest.
• image1 … image4 - up to four reference images for edits.
• apikey, response_format (url/b64_json), seed, skip_error - the usual suite.

Output is a single task of type COMFLY_IMAGE_FUTURE. Nothing else. If you drop it into a preview node directly you'll get an error that basically says "this isn't an image yet" - the Await node is mandatory, not optional.

Install

The pack-wide instructions apply unchanged. Easiest is ComfyUI Manager → search Comfyui-zhenzhen → install → restart. Manual, if you prefer:

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

No downloads beyond pip deps; you're paying for API calls, not weights. You need a Zhenzhen token (ai.t8star.org overseas, api.seedance.nz for the domestic discount) in the apikey field.

Trouble spots

• Congested default group. The default token group has a habit of getting slow when Google tightens its concurrency; if you see long stalls, the README suggests switching your token to a quality/优质 group - or jumping to the fal twin of this node, which runs a different lane.
• 429 / 500 errors are Google throttling, not your config. Rerun; the pack's own docs say most failures are the upstream vendor, especially during peak hours.
• Timed out ≠ failed. These are paid asynchronous jobs - a timeout just means the client gave up polling. Check the site's async-task list, grab the result by task ID, and don't resubmit until you've confirmed it actually died.