Superside Nano Banana V2 Edit Node

Nano Banana V2 is the same Google editing family as the Pro node, with a longer row of controls bolted on - and this Superside wrapper exposes most of them. Where the Pro node keeps things simple (image, prompt, resolution), V2 hands you a seed, a safety-tolerance dial, web-search grounding, and a thinking level, on top of the same up-to-six-reference editing. Same closed model, same fal.ai plumbing, same "only a wrapper can put this in your graph" reality.

The practical question is which Nano Banana node to reach for, and the answer is mostly "V2 if you need reproducibility or grounding, Pro if you want the least friction." V2's extra knobs are real but they're exactly the kind of thing you'll set once and stop touching.

The inputs that matter

• prompt, image_1, api_key - the essentials, same as Pro.
• image_2 through image_6 - extra reference images for scene-level edits.
• seed - the headline addition over Pro. V2 lets you pin a run and reproduce it; useful when you find a keeper and want the same take at a higher resolution. (Note the schema default is null/random.)
• resolution - this one goes further than Pro: 0.5K, 1K (default), 2K, or 4K. The 0.5K floor is a nice cheap-iteration option the Pro node doesn't give you.
• thinking_level - none (default), minimal, or high. This is how much reasoning the model does before editing. "High" genuinely helps with complex instructions ("make it look like late afternoon but keep the shadows soft") and genuinely costs more time per call.
• enable_web_search (default off) - lets the model pull in web context. Sounds fun; think twice. It means the model can ground on real-world references mid-edit, but for product work that's usually not what you want, and it's another thing that leaves your machine.
• safety_tolerance (default 4, 1–6) and limit_generations (default on) - moderation and quota dials. Leave them.
• num_images, aspect_ratio, output_format, sync_mode - same as Pro.

Outputs: images and description, same contract as Pro - the description string is worth reading on V2, because with thinking enabled it tends to explain the edit in a way that catches misread instructions early.

How it works and the drift warning

Same upload→call→download pattern as every fal node in this pack. And the same color-drift caution applies, arguably more on V2: the README explicitly calls out that Nano Banana passes accumulate a warm/red cast, and this pack's White Balance and Color Match nodes exist to fix exactly that. If you're chaining multiple edit passes, budget a color-correction step at the end of the chain.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/Superside/comfyui-superside-nodes
pip install -r requirements.txt

Restart ComfyUI, find it under Superside, paste a fal key into api_key (blank falls back to a FAL_KEY env var).

The honest take

For a straight "edit this image" task, Pro is the leaner node and the extra dials on V2 mostly don't matter. Reach for V2 when you actually need one of the three additions: a pinned seed for reproducibility, 0.5K for cheap iteration, or thinking_level for instructions complex enough that a shallow read will get them wrong. Otherwise you're just paying the V2 call rate for dials you're not turning.