WaveSpeedAI Google Nano Banana Pro Edit Ultra

The regular Nano Banana Pro Edit node already tops out at 4K. This one asks a reasonable question: why stop there? Nano Banana Pro Edit Ultra is the same multi-reference editor with a resolution ceiling of 8K, priced at $0.15 (4k) and $0.18 (8k) per the tooltip. It's the node for the person who doesn't want to think about whether the output will survive a print, a crop, or a delivery spec. Everything else about it is the standard Nano Banana Pro Edit experience: up to six reference images, ten aspect ratios, PNG-for-transparency output.

What it is

The high-resolution sibling in the pack's Google pair - same inputs as Nano Banana Pro Edit, but resolution only offers 4k and 8k (no 1k/2k economy tier), and it defaults to 4k rather than the smaller node's 1k. If you know the final image has to be enormous, you skip a render-and-rethink cycle by starting here.

How it works

Identical mechanism: client from the WaveSpeedAI Client node, a prompt plus up to six image URLs, and a POST to the Nano Banana Ultra edit endpoint. The result downloads back as an IMAGE tensor. The one thing to plan for is downstream memory - an 8K image as a tensor is big, and whatever node you wire it into needs to handle the size.

The inputs that matter

• prompt - the edit instruction in plain language.
• image_1..6 - reference images; image_1 is required, the rest optional.
• aspect_ratio - ten ratios from 1:1 to 21:9.
• resolution - 4k ($0.15) or 8k ($0.18). The gap between tiers is small; if 4K suffices, there's little reason to pay up, and if it doesn't, 8K is cheap enough to not sweat.
• output_format - png (transparency) or jpeg.
• enable_sync_mode - default on.

Output name is output_image, type IMAGE.

Install & setup

cd ComfyUI/custom_nodes
git clone https://github.com/razvanmatei-sf/razv-wavespeed.git
cd razv-wavespeed
pip install -r requirements.txt

or ComfyUI Manager → "razv-wavespeed" → Install, restart. API key from wavespeed.ai in config.ini (from config.ini.tmp), the Client node, or WAVESPEED_API_KEY. Deps: requests, pillow, numpy<2.0.0, torch.

Common issues

• 8K is a lot of tensor - if ComfyUI slows down or the save node complains, downscale the output before further processing. The generation was the point; the pixels are now a local resource problem.
• Big output, no visible gain - if your source references are low-res, 8K output is upscaled information, not new detail. Feed it good references.
• "No output received" - unreachable image URL or timeout on the longer 8K jobs.
• 401 - key issue.

Be honest about your ceiling. Most hobby work never needs 8K, and the plain Edit node at 2K covers a lot of it at half the price of anything here. Ultra earns its keep when output resolution is a hard deliverable - print, large-format display, or a client who'll ask "is that 8K?" The 3-cent gap between 4K and 8K means you'll rarely regret choosing Ultra when the canvas is already huge.