Concurrent Submit | Zhenzhen_nano_banana2_edit_S2A

Nano Banana 2 is Google's image workhorse - the model the community basically cannot shut up about, and the "banana" in "the whole internet is calling the Gemini image API from ComfyUI." This node is the Concurrent Submit twin of Comfly_nano_banana2_edit_S2A: the edit variant of Nano Banana 2, running in S2A mode (sync-to-async), which means every job becomes a queryable async task with a task ID you can chase. And as a Submit node, it runs in the 30-worker image pool, so a whole batch of edits queues and collects in one go.

The inputs that matter

• prompt - required.
• mode - text2img or img2img.
• model - nano-banana-2, nano-banana-pro, or the -2k / -4k pro variants. The author's changelog is worth heeding here: Google's 2K/4K output has been flaky at times, and the 2k/4k models exist as fallbacks for when the standard call downscales.
• aspect_ratio - auto plus the usual set. auto follows your reference image's ratio, which is the right default for edits.
• image_size - 1K / 2K / 4K.
• image1–image14 - up to fourteen reference images. This is the Nano Banana superpower: throw a whole character sheet at it and it keeps everyone straight.
• task_id - the S2A part. After a run you get a task ID you can use to recover the result if the download failed (the README's exact workflow: hit the site's async-task list, grab the image from the URL).
• response_format - url or b64_json. The changelog suggests URL when you're feeding lots of reference images.
• apikey - note the lowercase spelling here; this node takes a direct key.
• seed / skip_error - the usual pair.

The concurrent flow

Output is a task (COMFLY_IMAGE_FUTURE), not an image. Wire your edits into ComflyConcurrent_Image_Await (30 slots) and the collector waits for the batch, returns images in slot order, and appends a status JSON. The batch pattern is the whole point: fourteen reference images and five prompts means five jobs, and you want all five running at once.

Install and gotchas

Part of T8mars/Comfyui-zhenzhen: ComfyUI Manager → "Comfyui-zhenzhen", or git clone https://github.com/T8mars/Comfyui-zhenzhen into custom_nodes, restart. No local models, standard deps, paid token, overseas servers (proxy from China).

Two things the changelog itself warns about: Google's rate limits mean some token groups are more reliable than others (the README recommends the "优质" quality group for 2K/4K, and warns 429/500 errors are Google load, not your fault - and not billed), and the 4K/Pro tier has occasionally shipped white/blank images during Google's congestion. The S2A task ID is your recovery tool for exactly those cases.