Nano banana

Nano Banana Prompting Guide

Facts in this skill are verified against Google's official Gemini API docs, last checked 2026-08-04. See "Verification & Sources" at the bottom. Model IDs and prices change - re-verify before quoting them in a deliverable.

Model Lineup

Nano Banana is Google's marketing name for Gemini's native image generation. Four models are exposed via the Gemini API, all of them now Stable (GA):

Marketing name	Model code	Status	Updated	Best for
Nano Banana 2	gemini-3.1-flash-image	Stable (GA 2026-05-28)	Feb 2026	Default workhorse. All-round quality, 0.5K-4K, web + image search grounding, 14 aspect ratios, up to 14 reference images, video-to-image
Nano Banana 2 Lite	gemini-3.1-flash-lite-image	Stable (GA 2026-06-30)	Jun 2026	Volume and interactivity: sub-2s latency, cheapest per image, fast local edits. 1K only
Nano Banana Pro	gemini-3-pro-image	Stable (GA 2026-05-28)	Nov 2025	Studio-quality precision, dense/accurate text, complex graphic design, factual data visualizations, brand consistency
Nano Banana (legacy)	gemini-2.5-flash-image	⚠️ Deprecated - shutdown 2026-10-02	Oct 2025	Nothing new. Migrate existing code

Defaults - pick in this order:

1. Nano Banana 2 (gemini-3.1-flash-image) - Google's own recommendation as the go-to model, and the right default for almost everything: thumbnails, product shots, multi-reference composition, infographics, 4K finals.
2. Nano Banana 2 Lite (gemini-3.1-flash-lite-image) - when you need many images fast and cheap, or a snappy interactive edit loop (colour swaps, stickers, background changes). Half the price of Nano Banana 2 at 1K.
3. Nano Banana Pro (gemini-3-pro-image) - when text must be flawless, layout is complex, or brand/factual accuracy is the point. Roughly 2x Nano Banana 2 per image.

⚠️ Migration warning (state as of 2026-08-04)

• gemini-2.5-flash-image is deprecated, shutdown 2026-10-02. Google's guidance: transition to Nano Banana 2 Lite for better quality, faster generation and lower price (source). If a user's code or notes reference it, flag this.
• The -preview IDs are gone. gemini-3.1-flash-image-preview and gemini-3-pro-image-preview were shut down 2026-06-25 (deprecations). Never emit a -preview image model ID.
• Imagen is deprecated too - imagen-4.0-* shut down 2026-08-17. Use Nano Banana models for all image generation.

---

Quick Start: Generate or Edit an Image in 60 Seconds

Generate: Describe your scene in a full sentence with style and specs:

A professional product photo of a black leather wallet on white marble.
Soft studio lighting, macro lens, shallow depth of field. 2K, 3:2 aspect ratio.

Edit: Stay in the same conversation and describe only what changes:

Turn 1: "A red convertible parked on a coastal road at sunset."
Turn 2: "Change the color to midnight blue."
Turn 3: "Add surfboards on the roof rack."

The model carries the full visual context through the conversation - you describe only what changes, not the entire scene again.

For prompt formulas, model-specific guidance, and advanced techniques - read on.

---

Language of Prompts

Always write Nano Banana prompts in English, regardless of the language the user is writing in. The model was trained predominantly on English and produces noticeably better results with English prompts.

Workflow:

• Talk to the user in their language (Polish, German, French, etc.).
• Write the final prompt in English.
• If the user describes what they want in another language, translate to English before presenting the prompt.

Exception: text that should appear inside the image (e.g., a Polish poster headline, a Japanese label) stays in the target language. The surrounding prompt instructions stay in English.

Official best-performance language list (source): EN, ar-EG, de-DE, es-MX, fr-FR, hi-IN, id-ID, it-IT, ja-JP, ko-KR, pt-BR, ru-RU, ua-UA, vi-VN, zh-CN.

Polish is not on that list. Practical consequence: Polish prompt text is a bad idea (write English), and Polish text rendered inside an image is not a guaranteed-quality path. It usually works - Nano Banana 2 advertises "improved i18n text rendering" (model card) and Polish diacritics are Latin - but for a deliverable, treat it as needing a proof render. If the diacritics come out wrong (ą, ę, ś, ż, ź, ć, ń, ł), the reliable fallback is: generate the image without text, then set the headline as a real text layer in post. The same applies to any language outside the list.

Two-step trick for text-heavy images (straight from the official docs): "When generating text for an image, Gemini works best if you first generate the text and then ask for an image with the text." So write the exact copy in a first turn, then ask for the image in a second turn - don't make the model invent the copy and lay it out in one shot.

---

Golden Rules

1. Creative Director, not keyword vomit - Write full sentences describing a scene, not tag lists.
2. Edit, don't re-roll - If an image is 80% right, ask for specific changes. Preserve the 80% that works.
3. Explain the logic - Nano Banana 2, Lite and Pro are all thinking models (Pro card); Pro has Thinking on by default. They reason through intent. Say why, not just what.
4. Specify preservation explicitly - Always state what must not change during edits.
5. Provide context and purpose - "For a Brazilian gourmet cookbook" helps the model make better artistic decisions.
6. Use positive framing - Describe what you want, not what you don't want.

---

Core Prompt Structure (6 Factors)

Every strong prompt covers these six factors:

[SUBJECT] - who or what is in the image
[ACTION] - what is happening
[SETTING/LOCATION] - where the scene takes place
[COMPOSITION] - camera angle, framing, shot type
[LIGHTING/ATMOSPHERE] - quality and mood of light
[STYLE/MEDIUM] - art style, look and feel

Text-to-Image Template

[Subject + adjectives] doing [action] in [setting].
[Composition/camera angle]. [Lighting/atmosphere].
[Style/medium]. [Technical specs: resolution, aspect ratio].
[Constraints or text to render if any]

Example:

A stoic robot barista with glowing blue optics preparing an espresso
in a minimalist Scandinavian cafe. Wide shot, rule of thirds.
Soft morning light through large windows. Cinematic 3D, photorealistic.
2K, 16:9, shallow depth of field.

Image Editing Template

[Action verb] + [specific element] + [desired result].
KEEP UNCHANGED: [list of what must stay the same].
[Style/quality requirements].

Action verbs: Replace, Remove, Add, Change, Transform, Colorize, Translate, Enhance, Restore

Example:

Replace the white top with a black t-shirt.
KEEP UNCHANGED: face, hairstyle, pose, background, lighting.
Photorealistic, natural shadows. Seamless integration.

---

Reference Images (Multi-Image Composition)

The Gemini 3 image models accept up to 14 reference images in a single generation (source). Capacity per model, per the official limitations list:

Capacity	Nano Banana 2 (gemini-3.1-flash-image)	Nano Banana Pro (gemini-3-pro-image)
Objects with high fidelity	up to 10	up to 5
Characters with consistency	up to 4	(Pro leads on fidelity, count not stated)
Total reference images	14	14

Nano Banana 2 Lite is not the model for this. Its card states it is "not optimized for multiple reference inputs or multi-turn sequential editing" - if the job needs several references or a long edit chain, use Nano Banana 2.

Legacy Nano Banana (gemini-2.5-flash-image) works best with up to 3 input images (limitations).

Role-assignment pattern (works for any model):

Image 1 [CHARACTER/IDENTITY]: Keep this person's face exactly
Image 2 [POSE/COMPOSITION]: Use this framing and body position
Image 3 [STYLE/AESTHETIC]: Match this color palette and mood
Image 4 [LIGHTING]: Replicate this lighting setup
Image 5 [ENVIRONMENT/BACKGROUND]: Use this setting

Example multi-reference prompt:

Combine the subject from Image 1 (keep face identical) with the setting
from Image 2. Apply the lighting style from Image 3.
Seamlessly blend all elements, match perspective and shadows.
Photorealistic. 2K, 3:2 aspect ratio.

---

Multi-Character and Object Consistency

To maintain visual consistency across multiple characters in a single image (group photo, comic panel, storyboard):

Scene features [CHARACTER A] (tall woman, red coat, short dark hair)
and [CHARACTER B] (older man, grey beard, blue jacket).
Keep both characters visually identical to the reference images.
[Scene description, action, setting, composition]

Sequential consistency (storyboarding):

Use conversational editing - stay in the same chat and reference earlier images:

Continue this story from the previous image. The main character -
same as Image 1 - is now doing [new action] in [new location].
KEEP IDENTICAL: character appearance, clothing, hair, distinguishing features.
CHANGE: [what is different in this scene].

360-degree character view:

Documented in the official guide (source): iteratively prompt for additional angles in the same chat, referencing the first image to maintain consistency. For complex poses, include a reference image of the desired pose.

---

Text Rendering and In-Image Translation

Text rendering:

Specify text in quotes, with font style and placement:

Create a poster. Main headline: "AUTUMN SPECIAL" in bold gold serif at top.
Subtitle: "Limited Time Only" in clean modern badge style, right side.
Footer: "Offer ends Oct 31" in small clean text at bottom.
Ensure all text is perfectly spelled, ultra-sharp, high contrast.

Pro and Nano Banana 2 produce noticeably sharper, more reliable text than the original Nano Banana. For dense layouts (menus, infographics, multi-language posters), prefer Pro.

In-image translation:

The official docs demonstrate translating an existing image's text into another language while preserving the design (source, Spanish infographic example):

Update this image to [TARGET LANGUAGE]. Do not change any other
elements of the image. Match the original font style and weight.
All translated text must be perfectly legible.

The docs do not enumerate a supported language list - verify against the current docs if a specific script matters.

---

Conversational Editing

Iterate naturally in the same chat. The model holds the full visual context, so describe only what changes:

Turn 1: "A corporate headshot of a woman in a navy suit, white background, studio lighting."
Turn 2: "Change the background to a blurred modern office."
Turn 3: "Add glasses."
Turn 4: "Make the lighting warmer, golden hour feel."

Use this instead of starting from scratch when one element needs adjusting.

---

Image Search Grounding (Nano Banana 2 and Pro)

Nano Banana 2 and Pro can ground generations on Google Search results for factually accurate visuals (source, Grounding section). Nano Banana 2 additionally integrates Google Image Search grounding alongside web search, and it works with Thinking on or off.

Not supported by Nano Banana 2 Lite or legacy Nano Banana (2.5) - if grounding matters, those two are out.

Documented limitation: on Nano Banana 2, Search grounding "does not support using real-world images of people from web search at this time" (limitations). Don't promise a grounded likeness of a real person.

Use cases:

• Weather charts with current data
• Diagrams of real-world places (floor plans, maps, building layouts)
• Reference visuals for products, vehicles, or landmarks the model may not have seen

Visualize the current weather forecast for Tokyo for the next 5 days
as a modern weather chart with temperature values, condition icons,
and day labels. Clean professional style, 16:9, 1920x1080px.

Grounding is billed per search query on top of the image price (5,000/month free across the Gemini 3 family, then $14 / 1,000 queries) - see references/pricing.md.

---

Video-to-Image (Nano Banana 2 only)

Since 2026-05-28, gemini-3.1-flash-image accepts a video as multimodal context - either a public YouTube URL or a file uploaded via the Files API - and generates a still from it (source). The model reads the frames for visual themes and key events, then combines them with your text prompt.

This is the single best fit for the YouTube workflow: thumbnail straight from the finished video, instead of describing the video from memory.

Supported only on Nano Banana 2. Lite, Pro and legacy 2.5 do not take video input.

Prompt patterns:

Generate a YouTube thumbnail from this video.
Pick the single most dramatic moment as the base composition.
Add the headline "[3-4 WORDS]" in bold sans-serif, white with a heavy black outline, top-left.
Keep the presenter's face large, sharp and unobstructed. High saturation, mobile-readable contrast.
16:9, 1K.

Generate a cinematic poster that captures the key themes of this video.
No text. Leave clean negative space in the lower third for a title added later.

Generate a summary infographic of what this video covers: 4 numbered sections,
short labels only, flat design, generous whitespace. 4:5, 2K.

Practical notes: ask for one specific frame-moment rather than "the best moment" when you know what you want; the model does not guarantee a frame-accurate crop - it synthesizes a new image informed by the video, so treat it as art direction, not extraction. If you need the literal frame, pull it with ffmpeg and use it as a reference image instead.

---

Storyboards for Video Workflows

Nano Banana is documented as the starting point for storyboard frames that can be handed to Veo (Google's video model) (source, "Bonus: Comic strips and storyboards" + link to Veo guide at the end of the page).

Storyboard workflow:

1. Generate the first key frame in Nano Banana with full character/scene specification.
2. Use conversational editing to produce subsequent frames - reference the previous frame for consistency.
3. Export the frames and use them as image inputs to Veo for motion generation.

Frame prompt template:

Storyboard frame [N of total]. [Subject description with all consistency details]
doing [specific action] in [setting].
[Precise camera angle and framing - e.g. "low angle, wide shot"].
[Lighting suitable for video, e.g. "soft even diffuse light"].
Style: [art direction consistent with other frames].
Clear foreground/background separation.

Veo motion-from-keyframes specifics are out of scope for this skill - see the official Veo guide for the handoff details.

---

YouTube Thumbnails

Structure:

[Subject positioning] + [Expression/action] + [Text overlay specs]
+ [Background treatment] + [Contrast/color requirements] + [Technical specs]

Key principles:

• High contrast - mobile-readable, minimum 30% contrast between elements.
• Bold text - 3-5 words max, large bold sans-serif with outline or shadow. Use Pro or Nano Banana 2 for text accuracy.
• Strong emotion - excited/surprised expressions tend to increase CTR.
• Rule of thirds - face on one side, product/text on the other.
• Avoid bottom-right - YouTube timestamp overlay covers this area.

Example:

YouTube thumbnail for a tech review video.

SUBJECT: Person with excited expression, eyes wide, positioned left 2/3 of frame
PRODUCT: [PRODUCT NAME] on right side, subtle glow effect, ~30% of frame
TEXT: "[HEADLINE 3 WORDS]" in bold sans-serif, white with black outline, top center
BACKGROUND: Blurred tech-themed, high contrast with subject
COMPOSITION: Rule of thirds, visual flow from face to product to text

TECHNICAL: 1280x720px, 16:9, high saturation, mobile-optimized

Set output to 1K (or 0.5K on Nano Banana 2) for thumbnails - 2K/4K is overkill for YouTube's 1280x720 spec and costs more.

---

Infographics

Pro is best for infographics because of text accuracy and Search Grounding for factual data. Nano Banana 2 also works.

Best types:

• Timeline infographics
• Comparison charts (side-by-side)
• Process / How-to diagrams
• Statistical KPI cards
• Educational concept diagrams

Framework:

[Type] infographic: "[TITLE]"

STRUCTURE:
- [Section 1 description]
- [Section 2 description]
- [Section 3 description]

DESIGN:
- Colors: Primary [HEX], Secondary [HEX], Text [HEX], Background [HEX]
- Typography: [font style], headlines [size]pt, body [size]pt
- Layout: generous whitespace, [padding]px padding

TECHNICAL: [resolution], [aspect ratio], high contrast, all text ultra-sharp

STYLE NOTE: Flat design, clean lines. Avoid drop shadows, 3D effects, and gradients on charts.

For data-grounded infographics (e.g., "show GDP growth of top-5 economies last quarter"), enable Search Grounding by using Nano Banana 2 or Pro and including a phrase like "use real, current data".

---

Technical Specs

How resolution and ratio are requested

Output size is not something you write in prose - it is a parameter. In the current docs it lives in response_format on the Interactions API:

{
  "model": "gemini-3.1-flash-image",
  "input": "...prompt...",
  "response_format": { "type": "image", "aspect_ratio": "16:9", "image_size": "2K" }
}

Default behaviour: the model matches the output size to the input image, or generates a 1:1 square if there is no input image. Stating "16:9" in the prompt text helps composition but does not set the canvas - set aspect_ratio.

Note: the sampling parameters temperature, top_p and top_k were deprecated across the Gemini API on 2026-07-21 (changelog). Don't hand users prompt-tuning advice built on them.

Resolutions (output)

Model	0.5K (512px)	1K	2K	4K
gemini-3.1-flash-image	✅	✅ (default)	✅	✅
gemini-3.1-flash-lite-image	-	✅ only	-	-
gemini-3-pro-image	-	✅	✅	✅
gemini-2.5-flash-image (legacy)	-	✅ (up to 1024x1024)	-	-

Sources: 3.1 Flash card, 3.1 Flash Lite card, aspect-ratio tables.

Aspect ratios

• Nano Banana 2 - 14 ratios: 1:1, 1:4, 1:8, 2:3, 3:2, 3:4, 4:1, 4:3, 4:5, 5:4, 8:1, 9:16, 16:9, 21:9. The extreme banner ratios (4:1, 8:1 and their verticals) are exclusive to this model - useful for web headers and tall mobile strips.
• Nano Banana 2 Lite - 10 ratios: 1:1, 3:2, 2:3, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9.
• Nano Banana Pro - the same 10 as Lite (no extreme banner ratios).
• Legacy 2.5 - the same 10, single resolution.

Handy exact pixel sizes at 1K on Nano Banana 2: 16:9 = 1376x768, 9:16 = 768x1376, 4:5 = 928x1152, 1:1 = 1024x1024. At 2K and 4K these scale x2 and x4.

Watermarking

All generated images carry a SynthID watermark (source). Nano Banana 2 Lite additionally states "SynthID (Always On) + C2PA" watermarking on its model card - it is the only model in the family where the docs name C2PA Content Credentials. For the others, don't promise C2PA.

Capabilities matrix

Capability	NB 2 (3.1-flash-image)	NB 2 Lite (3.1-flash-lite-image)	NB Pro (3-pro-image)	Legacy (2.5-flash-image)
Image generation + editing	✅	✅	✅	✅
Conversational / multi-turn editing	✅	⚠️ fast local edits only	✅	✅
Multiple reference images	✅ up to 14	⚠️ not optimised	✅ up to 14	up to 3
Video input (video-to-image)	✅	❌	❌	❌
Thinking	✅	✅ (minimal / high)	✅ (default on)	❌
Search grounding (web)	✅	❌	✅	❌
Search grounding (image)	✅	❌	❌	❌
Function calling	❌	✅	❌	❌
Batch API	✅	✅	✅	✅
Flex / Priority inference	❌	not listed	✅	✅
Context caching	❌	❌	❌	✅
Structured outputs	❌	❌	❌	✅
Input token limit	131,072	65,536	65,536	65,536
Output token limit	32,768	4,096	32,768	32,768

Sources: per-model card pages linked at the end.

Access points

• Gemini API (generativelanguage.googleapis.com) - direct programmatic access. The image docs are now written against the Interactions API (client.interactions.create(...), response_format).
• Google AI Studio (aistudio.google.com) - free interactive playground for all four models; the right place to iterate on a prompt before spending API budget.
• Vertex AI - enterprise deployment.

Consumer surfaces (Gemini app) are out of scope for prompt-writing here but use the same underlying models.

---

Pricing (snapshot)

Detailed pricing in references/pricing.md. Headline numbers (source, verified 2026-08-04):

Model	Per-image output (Standard, paid tier)
gemini-3.1-flash-lite-image	$0.0336 (1K only) - cheapest
gemini-3.1-flash-image	$0.045 (0.5K), $0.067 (1K), $0.101 (2K), $0.151 (4K)
gemini-3-pro-image	$0.134 (1K and 2K), $0.24 (4K)
gemini-2.5-flash-image (legacy)	$0.039 (up to 1024x1024)

Cost intuition: Lite is roughly half of Nano Banana 2 at 1K, and Pro is roughly double it. Note the legacy 2.5 model is more expensive than Lite for a worse result - another reason the migration is a straight win.

No free tier for image generation models. Free Tier shows "Not available" for all four. The free Google AI Studio playground exists but counts against per-account quotas, not a billing-free image quota.

Batch API halves the per-image cost on all four models (up to 24h turnaround).

---

Common Issues and Fixes

Problem	Solution
Text blurry or misspelled	Switch to Nano Banana 2 or Pro (Lite is not a text model). Generate the copy in a first turn, then ask for the image. Add: "All text ultra-sharp, perfectly spelled, high resolution"
Polish/other off-list diacritics render wrong	Language is outside Google's best-performance list. Proof-render first; if broken, generate without text and add the headline as a real text layer in post
404 model not found on an image call	The -preview IDs were shut down 2026-06-25. Use gemini-3.1-flash-image / gemini-3-pro-image
Lite output looks flat vs. expectations	Lite is a speed/cost specialist at 1K with no grounding and weak multi-reference handling. Move the job to Nano Banana 2
Face changes during edit	Add: "KEEP UNCHANGED: face must be 100% identical to original"
Colors don't match brand	Use HEX codes: "background color: #0066CC"
Composition shifts	Add: "Preserve exact composition, framing, and layout"
Inconsistent style	Specify one dominant style: "Professional photography ONLY. No filters."
Character drifts across images	Supply reference image + "Character appearance is LOCKED to Image 1"
Translation looks wrong	Add: "Match original font weight, match text box size, do not rearrange layout"
Need accurate factual visual	Use Nano Banana 2 or Pro with "use real, current data" - triggers Search Grounding
4K image too expensive	Generate at 2K, upscale offline; or use Batch API for 50% discount
Need PNG output	Save the returned bytes locally with .png extension - models return image bytes, not URLs

---

What NOT to Do

• Tag soup: dog, park, 4k, realistic, sunny, professional
• Vague descriptions: "nice image", "good quality", "make it better"
• Cliche boosts: "trending on artstation, masterpiece, 8k ultra" - the model ignores these
• Re-rolling from scratch when 80% is already right - edit conversationally instead
• Missing preservation instructions during edits
• Mixing conflicting styles in one prompt
• Over-prompting - 500+ word prompts confuse rather than clarify
• Celebrity names - use descriptive attributes instead ("woman with sharp cheekbones and red hair")
• Picking Pro for thumbnails - Pro is for high-fidelity work; thumbnails don't need it and pay ~2x more per image
• Emitting -preview model IDs - all of them are shut down. Four live IDs only: gemini-3.1-flash-image, gemini-3.1-flash-lite-image, gemini-3-pro-image, gemini-2.5-flash-image (legacy, dying 2026-10-02)
• Reaching for Lite on a hard job - many references, long edit chains, grounded facts, or anything above 1K all rule it out
• Setting the canvas in prose only - "make it 16:9" without aspect_ratio gives you a square or an input-matched size

---

Quick Reference Card

Always include:

• Subject + action + setting
• Lighting conditions
• Art style / medium
• Technical specs (resolution, aspect ratio)
• What to preserve (for edits)

Model picks:

• Default for anything → Nano Banana 2 (gemini-3.1-flash-image)
• Thumbnail from a finished video → Nano Banana 2 (only model with video input)
• Bulk runs, interactive edit loops, sub-2s latency, lowest cost → Nano Banana 2 Lite (gemini-3.1-flash-lite-image)
• Studio-quality design / dense accurate text / brand & factual precision → Nano Banana Pro (gemini-3-pro-image)
• Legacy code on gemini-2.5-flash-image → migrate before 2026-10-02

For ready-to-use prompts by category, see references/prompt-examples.md. For pricing breakdown by tier (Standard, Batch, Flex, Priority), see references/pricing.md.

---

Verification & Sources

Every numeric claim, model ID, capability flag, and price in this skill is sourced from Google's official Gemini API docs at ai.google.dev/gemini-api/docs, verified on 2026-08-04.

• Image generation guide: https://ai.google.dev/gemini-api/docs/image-generation
• Nano Banana 2 card (gemini-3.1-flash-image): https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-image
• Nano Banana 2 Lite card (gemini-3.1-flash-lite-image): https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite-image
• Nano Banana Pro card (gemini-3-pro-image): https://ai.google.dev/gemini-api/docs/models/gemini-3-pro-image
• Nano Banana legacy card (gemini-2.5-flash-image): https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash-image
• Pricing page: https://ai.google.dev/gemini-api/docs/pricing
• API changelog (GA dates, deprecated sampling params): https://ai.google.dev/gemini-api/docs/changelog
• Deprecations and shutdown dates: https://ai.google.dev/gemini-api/docs/deprecations
• SynthID watermark: https://ai.google.dev/responsible/docs/safeguards/synthid
• Veo handoff (storyboards): https://ai.google.dev/gemini-api/docs/video

Key dates from those sources, worth re-checking on each audit:

Date	Event
2026-05-28	gemini-3.1-flash-image + gemini-3-pro-image GA; video-to-image launched
2026-06-25	-preview image model IDs shut down
2026-06-30	gemini-3.1-flash-lite-image (Nano Banana 2 Lite) GA
2026-07-21	temperature / top_p / top_k deprecated API-wide
2026-08-17	Imagen 4 models shut down
2026-10-02	gemini-2.5-flash-image shut down

Things this skill does not claim because the official docs do not confirm them:

• A "100+ languages" count for text rendering. The docs list 15 best-performance locales (Polish not among them) and say nothing about a total.
• C2PA on any model except Nano Banana 2 Lite, whose card names it explicitly.
• A formal "Veo + Lyria + Flow 4-stage pipeline" - docs only link to the Veo guide; the rest is third-party narrative.
• A character-count limit for Pro's reference images - Google states Pro leads on fidelity (5 images high-fidelity) without giving a character number.
• Default model claims for the consumer Gemini app - this skill targets the API.

If you discover a claim above is wrong, the fix is to update both this section and the relevant body section with the corrected fact and its source URL.