Gemini 2.5 Flash Image: Production-Ready with New Aspect Ratios

Gemini 2.5 Flash Image is now production-ready and supports 10 aspect ratios (from 21:9 to 9:16), $0.039 per image pricing, and image-only output. You can use it in Google AI Studio (Gemini API) or Vertex AI and set the ratio directly in your request.

What’s new in a sentence

You’re no longer stuck with square: you can generate or edit images in commonly used aspect ratios for social, ads, and product imagery without clumsy workarounds.

What’s new in Gemini 2.5 Flash Image at a glance

• 10 aspect ratios: 21:9, 16:9, 4:3, 3:2, 1:1, 9:16, 3:4, 2:3, 5:4, 4:5.
• Production-ready: available via Gemini API in AI Studio and on Vertex AI for enterprise.
• Image-only output: tell the API to return just images when that’s your goal.
• Pricing: $0.039 per image (with $30 per 1M output tokens).
• Community-tested: examples include character-consistent renders and in-app live edits.

Supported aspect ratios and when to use each

Landscape (21:9, 16:9, 4:3, 3:2)

• 21:9: cinematic banners, hero sections, YouTube channel art.
• 16:9: YouTube thumbnails, web hero images, slide covers.
• 4:3: editorial images, presentation content.
• 3:2: product cards and “above-the-fold” blog images.

Square (1:1)

• Instagram grid, catalog tiles, marketplace listings.

Portrait (9:16, 3:4, 2:3)

• 9:16: Reels/TikTok/Shorts covers, Stories, vertical ads.
• 3:4: posters, app store promo art.
• 2:3: lifestyle full-body shots, magazine-style portraits.

Flexible (5:4, 4:5)

• Pinterest, product detail images, portrait-leaning ads where text overlays need breathing room.

How to set aspect ratios in code and avoid the square trap

In AI Studio (no code)
Select the Image model, choose Build, and set Aspect ratio to your target (e.g., 16:9). For edits, the model usually preserves the input image ratio if it drifts, add a line to your prompt: “Do not change the input aspect ratio.”

With the Gemini API (Python)

from google import genai
from google.genai import types

client = genai.Client()

resp = client.models.generate_content(
    model="gemini-2.5-flash-image",
    contents=["Create a product hero image: matte black earbuds on marble, soft daylight"],
    config=types.GenerateContentConfig(
        response_modalities=["IMAGE"],
        image_config=types.ImageConfig(
            aspect_ratio="16:9"
        )
    )
)
# Save first image part
img = resp.parts[0].as_image()
img.save("earbuds_hero_16x9.png")

Common gotchas

• Mixed inputs: If you upload multiple images with different ratios, the model may follow the last image’s ratio.
• Default square: No input image? The model may default to 1:1 unless you set aspect_ratio.
• Prompt clarity: For edits, explicitly state “keep original framing and aspect ratio.”

Pricing & limits you should actually plan for

Pricing

• $0.039 per image (billed via output tokens: ~1,290 tokens per image; $30 per 1M output tokens).
• Other modalities follow standard Gemini 2.5 Flash pricing.

Limits/behaviors to note

• Public docs and forum notes point to 1024-px output as the current default max for generation; plan to upscale externally if you need 2K+ banners.
• Input constraints: up to 3,000 images per prompt, max 7 MB each, common formats (png, jpg, webp).
• Preview model retirement: gemini-2.5-flash-image-preview and the older 2.0 preview image model are slated for retirement on Oct 31, 2025—switch to gemini-2.5-flash-image.

Editing, blending, and “image-only output” tips

Targeted edits
Ask for precise changes: “remove the coffee stain,” “blur background to f/2.8 look,” “match brand hex #0B5FFF for the label.” Keep each change as a separate sentence small, clear instructions get cleaner results.

Multi-image fusion
Upload a product cutout and a room scene, then prompt: “Place the product on the table, 3/4 angle, soft window light, reflections consistent with scene.”

Image-only output
When you don’t need text responses, request response_modalities=["IMAGE"]. It keeps outputs tidy and reduces parsing overhead in pipelines.

Mini case study: one shot, three channels

We started with a 1:1 pack shot (e-commerce).

• Step 1 (1:1 → 16:9): “Recompose for website hero; product center-right, negative space on left for headline; daylight look; marble surface.”
• Step 2 (16:9 → 9:16): “Re-crop for vertical story; maintain product scale ~60% height; add subtle background texture; keep brand blue accent.”
• Step 3 (Final polish): Export PNGs, then add text overlays manually for crisp typography.

Time saved: we skipped manual cropping and masking for each format.

Pros and cons

Comparison / Pros and Cons Table

Frequently Asked Questions (FAQs)

Is Gemini 2.5 Flash Image free?
You can try it in AI Studio, but production use is billed $0.039 per image at current rates.

How do I change the aspect ratio?
Set aspect_ratio in your request (e.g., "16:9"). In edits, add “Do not change the input aspect ratio.”

Does it add a watermark?
Yes SynthID (invisible) is embedded in generated/edited images for provenance.

Can it generate images of people?
Docs indicate people generation support with updated safety filters in the current public release.

What about the preview model I used last month?
Migrate to gemini-2.5-flash-image; preview image models are scheduled to retire Oct 31, 2025.

Max resolution today?
Expect ~1024-px generation; plan upscaling for print/retina heroes.

The Bottom Line & Implementation Checklist

Bottom Line
Gemini 2.5 Flash Image is live for production with 10 aspect ratios, straightforward API control, and $0.039/image pricing. Use it to standardize asset generation across web, social, and ads, then upscale when you need bigger canvases.

Checklist

• Switch model name to gemini-2.5-flash-image
• Set aspect_ratio explicitly (16:9, 9:16, etc.)
• Add “keep aspect ratio” language to edit prompts
• Budget for $0.039/image and output tokens
• Note SynthID watermarking in your brand guidelines
• Plan external upscaler for >1024-px deliverables

Featured Snippet Boxes