Concurrent Submit | Zhenzhen_nano_banana

If you only install one concurrent Google-image node from this pack, it's probably this one. ComflyConcurrent_Comfly_nano_banana_Submit is the wrapper around the pack's general-purpose Comfly_nano_banana node, and that node is special for one reason: the model dropdown. It lists seven Google image models - nano-banana-2, nano-banana-pro, gemini-3-pro-image, gemini-2.5-flash-image, nano-banana, nano-banana-hd, gemini-2.5-flash-image-preview - so one node can stand in for almost your entire Google image workload. Swap the model, keep the workflow.

It's also a chat-format node, which is why it looks different from the edit twins: instead of aspect_ratio and image_size, you get sampling knobs - temperature, top_p, and max_tokens (up to 32768). Those are real model parameters, not decoration; lower temperature for consistent batch output, higher for variety.

The inputs

• text - the prompt (note: the field is text, not prompt).
• model - the seven-way dropdown above.
• image1 … image4 - up to four reference images, for the edit-capable models.
• temperature / top_p / max_tokens - sampling controls.
• apikey, seed, skip_error - the usual.

The output is a single COMFLY_IMAGE_FUTURE task. Which means: this node does not produce an image. Pair it with Concurrent Collect Images (ComflyConcurrent_Image_Await), which takes up to 30 tasks from the shared image pool, waits for them all, and returns image_1…image_30 in order. That's the whole async pattern of this pack - submit cheap and instant, collect once everything's landed.

Install

Identical to the rest of the pack:

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

Or ComfyUI Manager → search Comfyui-zhenzhen → restart. Grab a token from the Zhenzhen site (ai.t8star.org overseas, api.seedance.nz for the domestic discount) and drop it in apikey.

Things people hit

• Not every model is equal here. The original nano-banana and gemini-2.5-flash-image-preview entries are the ones with a documented "sometimes returns no image" quirk; the README flags the preview model as chat-format only and somewhat unreliable. For anything serious, nano-banana-2 or nano-banana-pro is the safer pick.
• Token group matters. Some models route through the default group, which gets congested; the quality (优质) group is the stable lane when Google tightens concurrency. The README's guidance: default for cheap, quality when you need it to actually succeed.
• Timeouts on chat nodes usually mean the upstream vendor is throttling, not that your prompt was bad. Rerun, or check the async-task page for a completed job before paying twice.