Concurrent Submit | Zhenzhen Nano Banana Pro Edit Fal

Google's Nano Banana Pro - the flagship Gemini image model, the one with native 4K output and actual "thinking mode" reasoning - but reached through fal's queue rather than the direct Google lane, and wrapped in Zhenzhen's concurrent submit system. If you've ever sat through a 4K Pro generation that took minutes and blocked your whole graph, this is the node that lets you queue a dozen of them and go do something else.

Like every Submit twin in this pack, it doesn't return an image. It hands you a COMFLY_IMAGE_FUTURE task and runs the actual generation in the shared image pool (30 workers). The matching Concurrent Collect Images (ComflyConcurrent_Image_Await) node waits on up to 30 tasks and gives you image_1…image_30 back in order. The submit is instant; the waiting happens in the collector, where it's at least happening in parallel.

The inputs that matter

• prompt - the instruction. Pro rewards detail; this is the model that thinks before it draws.
• image1 … image8 - reference images. Pro's real party trick is heavy multi-reference editing (Google advertises up to 14; this fal endpoint exposes 8 slots).
• resolution - 1K / 2K / 4K. The biggest cost and quality lever. 4K is where this model earns its reputation - and where the file sizes get dangerous (see gotchas).
• num_images - 1 to 4 per call.
• aspect_ratio - auto plus the standard ratios.
• safety_tolerance - Google's moderation strictness, 1–6. Lower = stricter.
• system_prompt - optional persona/style steering.
• seed - fal max 65535, 0 = random.
• output_format - png / jpeg / webp.
• image_way - image_url or base64 for the references.
• poll_interval / max_poll_attempts - default 1200 × 3s = a 1-hour timeout budget.

Install

Standard for the pack - ComfyUI Manager, search Comfyui-zhenzhen:

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

Restart, put your Zhenzhen token (ai.t8star.org overseas; api.seedance.nz for the domestic discount) into api_key. Nothing downloads; it's an API reseller pack.

Where this specific node bites

• 4K downloads fail cross-border. This is the pack's own documented headache: Pro's 4K image files are big enough that fetching them overseas fails intermittently. The fix is on the site, not in ComfyUI - edit your token, scroll to the bottom, and enable OSS (object storage) for CN or US. If you get a failure, the terminal's last error will include a download URL you can grab directly.
• The quality group is basically mandatory at 4K. The README is blunt: the default group has been congested since Google cut concurrency, and 4K through default risks white/blank outputs. Switch your token to a quality (优质) group for Pro.
• Pre-deducted fal billing. The displayed price is an estimate; the final charge settles after the job. And timeouts - check the async-task page before resubmitting, because a timed-out client may still have a finished job server-side.