Concurrent Submit | Zhenzhen Nano Banana 2 Edit Fal

Same trick as the other Submit twins, different backend. This one wraps Comfly_nano_banana_2_fal - Google's Nano Banana 2 image model reached through fal's queue instead of the direct Google route. Zhenzhen runs multiple lanes to the same model, and the fal lane is the one you switch to when the default group gets congested, throttled, or starts throwing 429/500s. The pack's changelog frames it plainly: May 2026 rewrote the fal polling logic entirely because the old code kept the front end spinning after the backend had already failed.

The node runs in the shared image pool (30 workers), so the task output is a COMFLY_IMAGE_FUTURE - collect it with Concurrent Collect Images (ComflyConcurrent_Image_Await), which gathers up to 30 tasks, waits on all of them, and restores slot order on image_1 through image_30.

What's on the node

• prompt - the edit instruction. Multiline.
• image1 … image8 - reference images for the edit. Up to eight here (the direct Google node offers 14; fal's endpoint caps lower).
• num_images - 1 to 4 images per generation. Remember fal bills per image, so this multiplies your spend.
• resolution - 1K / 2K / 4K, the main cost lever.
• aspect_ratio - auto plus the usual 16:9 → 9:16 set.
• safety_tolerance - Google's filter strictness, 1 (strictest) to 6. This is a real knob and people do use it; just don't expect it to sneak past the vendor's moderation entirely.
• system_prompt - an optional instruction that steers persona and output style.
• output_format - png / jpeg / webp.
• seed - capped at 65535 (fal's limit); 0 means random.
• image_way - image_url (upload to Zhenzhen, get a URL) or base64 (embed in the payload). URL is usually faster for big reference images.
• poll_interval / max_poll_attempts - polling cadence and budget; default 1200 × 3s = one hour timeout.
• api_key - your Zhenzhen token; skip_error if you'd rather get a blank than an exception.

Install and gotchas

Same as every node in this pack - ComfyUI Manager, search Comfyui-zhenzhen:

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

Restart, grab an API key from the site (ai.t8star.org overseas, api.seedance.nz for the domestic discount), and you're off. No model weights live on your disk; the pack does pull a pile of pip deps (aiohttp, opencv-python, openai, transformers…) on first install.

Two things worth knowing before you batch. First, fal-style billing here is pre-deducted and settled later - the price shown on submit isn't necessarily the final charge, which the README spells out for every fal node. Second, the 4K Pro downloads are big enough that cross-border fetches fail; enable the OSS option on your token's edit page (CN or US) if you're chasing 4K through this route. And if you get a timeout, check the async-task page before rerunning - the job may have finished server-side and you'd otherwise pay twice.