Zhenzhen Nano Banana Pro Edit Fal
Nano Banana Pro at 4K through FAL — mind the download size
- image1
- image2
- image3
- image4
- image5
- image6
- image7
- image8
- images
- image_urls
- response
Nano Banana Pro is Google's flagship image model - 4K native output, "thinking" before it generates, up to fourteen reference images - and Comfly_nano_banana_pro_fal ("Zhenzhen Nano Banana Pro Edit Fal") is the FAL-route node for it. It's structurally identical to the Banana 2 FAL node: same eight reference slots, same num_images batching, same 1K/2K/4K resolution switch, same FAL polling machinery. The difference is which flagship you're calling and how expensive its output is.
It's a FAL-class node in the T8mars/Comfyui-zhenzhen pack, the "Comfly" API-wrapper family that calls the author's paid reseller service. Key goes into api_key directly; billing is pre-deduct and settles after. Submit → poll → download, results as IMAGE.
The inputs that matter
prompt- required. Pro's "thinking" behavior rewards detailed instructions; it reasons before generating.image1–image8- reference images for edit jobs.num_images- 1 to 4 per call.aspect_ratio-autoplus the standard set.resolution- 1K (default), 2K, 4K.output_format- png/jpeg/webp.safety_tolerance- 1 to 6; default 4.system_prompt- an optional instruction to steer the model's persona/style.image_way-image_urlorbase64for references.seed- real FAL seed (0–65535, 0 = random).poll_interval/max_poll_attempts- defaults 3s × 1200 = a one-hour timeout ceiling.
Outputs
images- the result(s) asIMAGE.image_urls- hosted URLs as a string.response- the raw payload.
Installing it
cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen
# restart ComfyUI
Or ComfyUI Manager → "Comfyui-zhenzhen", then a key from the author's site into api_key.
The honest part
The one thing that actually distinguishes this node from its siblings is the 4K download problem. The README is upfront about it: Pro 4K output is a big file, and cross-border transfers fail often enough that the author added an OSS (object storage) setting on the website - you edit your token, scroll to the bottom, and pick a CN or US OSS region so the result has a download route that works for you. If your downloads keep dying mid-transfer, that's the fix, not reinstalling anything. Same for the usual suspects: white 4K images and 429/500s are Google load problems (premium group on the site is the real cure), and every prompt and reference you send goes through a reseller's server. Great node; just budget for its file sizes.
Inputs (21)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | — | |
| image1opt | IMAGE | — | |
| image2opt | IMAGE | — | |
| image3opt | IMAGE | — | |
| image4opt | IMAGE | — | |
| image5opt | IMAGE | — | |
| image6opt | IMAGE | — | |
| image7opt | IMAGE | — | |
| image8opt | IMAGE | — | |
| api_keyopt | STRING | — | |
| num_imagesopt | INT | 11–4 | — |
| aspect_ratioopt | COMBO | auto | 11 options: auto, 21:9, 16:9, 3:2, 4:3, 5:4, +5 |
| resolutionopt | COMBO | 1K | 3 options: 1K, 2K, 4K |
| output_formatopt | COMBO | png | 3 options: png, jpeg, webp |
| safety_toleranceopt | COMBO | 4 | 6 options: 1, 2, 3, 4, 5, 6 |
| system_promptopt | STRING | Optional system instruction to steer model persona and output style. | |
| seedopt | INT | 00–65535 | 0 = random seed. FAL seed max is 65535. |
| image_wayopt | COMBO | image_url | 2 options: image_url, base64 |
| poll_intervalopt | INT | 31–30 | — |
| max_poll_attemptsopt | INT | 120010–3600 | Default 1200*3s = 1 hour timeout. |
| skip_erroropt | BOOLEAN | false | 开启后,节点失败时不报错、返回默认空结果。 |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| images | IMAGE | — |
| image_urls | STRING | — |
| response | STRING | — |