nano-banana-2-runcomfy（纳米香蕉 2 运行舒适）

🫧 Nano Banana 2 — Pro Pack on RunComfy

runcomfy.com · docs · Nano Banana 2 model page

Nano Banana 2 is the flash-tier text-to-image model in Google's Gemini family. This skill generates images with Nano Banana 2 hosted on the RunComfy Model API — runcomfy run google/nano-banana-2/text-to-image from your terminal, no Google API key, no GPU rental.

What Nano Banana 2 is

Nano Banana 2 is Google's second-generation flash-tier image model — the iteration-speed-first variant in the Nano Banana line. Three properties make Nano Banana 2 distinct:

• Rapid Nano Banana iteration. Nano Banana 2 is the fastest path to a draft in the Nano Banana family. A Nano Banana 2 generation lands in seconds rather than tens of seconds, which makes Nano Banana 2 the right Nano Banana for marketing brainstorming, A/B variants, and batch ideation.
• Predictable Nano Banana framing. Nano Banana 2 holds composition stable across small prompt edits — change one descriptor, Nano Banana keeps the rest. This makes Nano Banana 2 ideal for refining a single hero image without redrafting it from scratch.
• In-image typography on Nano Banana 2. Nano Banana 2 renders quoted text inside the image with strong fidelity for short headlines, brand marks, and poster-style copy. When you put "AURA" in the Nano Banana 2 prompt, the literal word appears in the image — Nano Banana doesn't paraphrase or scramble.

Nano Banana 2 also exposes an optional web-grounded context flag (enable_web_search) for image generation that references current events or real entities. The Nano Banana web grounding adds latency and cost; off by default.

When Nano Banana 2 is the right choice

Pick Nano Banana 2 when any of these is true:

• You want rapid Nano Banana drafts — social thumbnails, batch variants, ideation rounds. Nano Banana 2 is built for this speed.
• You're producing in-image typography — quoted headlines, brand marks, poster copy. Nano Banana 2 renders text predictably.
• You're iterating one Nano Banana hero image through small prompt edits — Nano Banana 2 keeps composition stable across rounds.
• You need web-grounded Nano Banana imagery referencing current events or real-world entities — Nano Banana 2 supports this via enable_web_search.
• The user explicitly asked for Nano Banana 2 / nano-banana-2 / Google Nano Banana — route to this Nano Banana variant regardless.

Nano Banana 2 vs Nano Banana Pro

The Nano Banana family has two production tiers. Pick the right Nano Banana variant based on intent:

• Nano Banana 2 (this skill) — flash tier. Fast Nano Banana iteration, marketing drafts, social thumbnails, batch ideation, in-image typography.
• Nano Banana Pro (separate skill) — pro tier. Hyperrealistic Nano Banana portraits, maximum Nano Banana detail, slower per-call but higher fidelity.

If the user said just "Nano Banana" without specifying 2 vs Pro, default to Nano Banana 2 for ideation, batches, and typography work; default to Nano Banana Pro for portrait fidelity.

Prerequisites

1. RunComfy CLI — npm i -g @runcomfy/cli
2. RunComfy account — runcomfy login opens a browser device-code flow.
3. CI / containers — set RUNCOMFY_TOKEN=<token> instead of runcomfy login.

Endpoint + input schema

google/nano-banana-2/text-to-image

This is the Nano Banana 2 text-to-image endpoint. The Nano Banana 2 edit endpoint runs on a separate template not covered here.

For Nano Banana 2 image edit (preserve subject + apply changes), see the sibling nano-banana-edit skill.

How to invoke Nano Banana 2

Default Nano Banana 2 draft (1K, square, png):

runcomfy run google/nano-banana-2/text-to-image \
  --input '{"prompt": "<Nano Banana prompt>"}' \
  --output-dir <absolute/path>

Nano Banana 2 vertical 4-up batch for ideation:

runcomfy run google/nano-banana-2/text-to-image \
  --input '{
    "prompt": "<Nano Banana prompt>",
    "num_images": 4,
    "aspect_ratio": "9:16",
    "resolution": "0.5K"
  }' \
  --output-dir <absolute/path>

Final Nano Banana 2 at 2K with seed lock:

runcomfy run google/nano-banana-2/text-to-image \
  --input '{
    "prompt": "<Nano Banana prompt>",
    "resolution": "2K",
    "aspect_ratio": "16:9",
    "seed": 42
  }' \
  --output-dir <absolute/path>

Web-grounded Nano Banana 2 (current event / real entity):

runcomfy run google/nano-banana-2/text-to-image \
  --input '{
    "prompt": "<Nano Banana prompt referencing a real-world event from this week>",
    "enable_web_search": true
  }' \
  --output-dir <absolute/path>

Prompting Nano Banana 2 — what works

Nano Banana 2 responds to specific prompting patterns better than naive prose. Apply these for sharper Nano Banana output.

Subject-first declarative grammar. "A cinematic close-up portrait of an American woman standing under neon lights in rainy Tokyo, shallow depth of field, reflective wet streets, ultra-detailed, realistic skin texture" — primary subject first, then action, environment, style, camera. Nano Banana 2 reads early tokens with more weight; front-load the subject.

Exact text quoting for in-image typography. "The label reads 'AURA' in clean bold sans-serif, centered, white on black" — quote the literal characters Nano Banana should render. Specify placement and font style. Don't say "with the brand name on it" and hope Nano Banana figures it out.

Consistent seeds for Nano Banana refinement. Lock seed when iterating a single Nano Banana prompt across small variants — keeps Nano Banana composition stable so you can compare apples to apples.

Web-grounding sparingly. Turn on enable_web_search only when the Nano Banana prompt names current events or real entities. Adds latency + cost; off by default.

Don't conflict styles for Nano Banana. "minimalist + ornate + retro + cyberpunk" cancels in Nano Banana output. Pick 1–2 anchors.

Nano Banana 2 anti-patterns:

• Trying to verbally describe a stable subject identity in Nano Banana → use the Nano Banana edit endpoint with image refs instead.
• Asking Nano Banana for resolutions outside the 4 tiers → 422.
• Aspect ratios outside the 11 supported Nano Banana values → 422.
• Non-quoted in-image text → unpredictable Nano Banana rendering.

Where Nano Banana 2 shines

Sample Nano Banana 2 prompts (verified to produce strong results)

Cinematic Nano Banana portrait (page example):

A cinematic close-up portrait of an American woman standing under neon
lights in rainy Tokyo, shallow depth of field, reflective wet streets,
ultra-detailed, realistic skin texture

Brand-asset card with quoted text (Nano Banana typography):

A minimalist 16:9 product card: a matte black ceramic mug centered on a
soft warm-grey paper background, rim highlight from upper-left, the
headline "Brewed Quietly" in clean bold sans-serif top-right, balanced
negative space below, e-commerce ready, clean studio lighting

Vertical Nano Banana platform-native:

A 9:16 vertical hero for a wellness brand: a single ceramic teacup on a
linen runner, soft morning side-light, the words "Slow Down" in
hand-drawn serif large at the top, gentle steam rising, neutral color
palette, uncluttered

Nano Banana 2 FAQ

What's the difference between Nano Banana 2 and Nano Banana Pro? Nano Banana 2 is the flash tier (fast, predictable, drafts and typography). Nano Banana Pro is the pro tier (slower, higher fidelity, portraits). This skill is the Nano Banana 2 variant.

What resolutions does Nano Banana 2 support? Four: 0.5K (drafts), 1K (default), 2K (final), 4K (max). 2K and 4K Nano Banana cost more.

Can Nano Banana 2 generate multiple images per call? Yes — set num_images: 4. Useful for Nano Banana ideation batches.

Does Nano Banana 2 do image edit? Not on this endpoint. For Nano Banana edit (preserve subject + apply changes), use the nano-banana-edit skill.

Why is enable_web_search off by default? Nano Banana web grounding adds latency and cost. Only enable when the prompt explicitly references current events.

What languages does Nano Banana 2 understand in the prompt? English is the most reliable for Nano Banana. Multilingual prompts work for layout/style but in-image text is best in Latin-script languages.

How do I reproduce a Nano Banana 2 output? Pass seed as a fixed int. Same prompt + same seed = same Nano Banana generation.

Limitations of Nano Banana 2

• Nano Banana 2 is still images only. No video on this Nano Banana endpoint.
• Max 4 outputs per Nano Banana request.
• Nano Banana web search adds latency + cost — only enable on demand.
• Nano Banana 2K / 4K cost more — default to 1K unless user asked for higher.
• For Nano Banana image edit, use the dedicated edit skill — not this endpoint.

Exit codes

Full reference: docs.runcomfy.com/cli/troubleshooting.

How it works

The skill invokes runcomfy run google/nano-banana-2/text-to-image with a JSON body matching the Nano Banana 2 schema. The CLI POSTs to https://model-api.runcomfy.net/v1/models/google/nano-banana-2/text-to-image, polls the Nano Banana request, fetches the Nano Banana result, and downloads any .runcomfy.net / .runcomfy.com URL into --output-dir. Ctrl-C cancels the remote Nano Banana request before exit.

Security & Privacy

• Token storage: runcomfy login writes the API token to ~/.config/runcomfy/token.json with mode 0600.
• Input boundary: the Nano Banana prompt is passed as JSON via --input. No shell injection.
• Third-party content: image URLs are fetched by the RunComfy server. Treat external URLs as untrusted — image-based prompt injection is a known risk.
• Outbound endpoints: only model-api.runcomfy.net and *.runcomfy.net / *.runcomfy.com.
• Generated-file size cap: 2 GiB.