What’s New in Gemini’s Image Editing

According to Android Central, Google is rolling out an update that allows Gemini to utilize Google DeepMind’s imaging technology for transformative edits directly within the chat experience. Essentially, you can send Gemini a photo and request edits like removing distractions, adjusting lighting, changing skies, or expanding a background. It will produce polished results with clearer previews and step-by-step suggestions (Android Central).

Behind the scenes, this update combines Gemini’s language understanding with Google’s latest generative image model, making complex edits feel conversational. While Google hasn’t shared a specific marketing name for this entire pipeline, they have publicly detailed two key components that drive these upgrades:

• Imagen 3, Google DeepMind’s most advanced text-to-image model for photorealism and fine detail, introduced at Google I/O 2024 (DeepMind).
• SynthID, Google’s watermarking technology that invisibly tags AI-generated images to support authenticity and facilitate downstream detection (DeepMind).

Why This Matters

Google already offers impressive one-tap edits in Photos, like Magic Editor and Magic Eraser. With the Gemini update, that power shifts into a conversational interface: instead of fiddling with sliders, you can describe the changes you want and refine them using natural language. Here are a few advantages:

• Fewer steps. Simply tell Gemini what you want to change, then refine with follow-up phrases like “make the sky warmer” or “pull the camera back a bit more.”
• Better context. Gemini can understand the scene, so instructions like “make the lighthouse the focal point” or “match the lighting to golden hour” are more likely to align with your vision.
• Share-ready outputs. Results will be watermarked with SynthID and can include disclosures as needed, promoting responsible sharing.

What You Can Do Now

Capabilities may vary by region and account, but early users can generally expect the following types of edits when they attach a photo in Gemini and ask for adjustments:

• Remove objects, wires, and photobombers.
• Relight a subject for more balanced portraits.
• Replace skies or adjust the atmosphere for different weather and time-of-day vibes.
• Expand the frame (outpainting) to fix tight crops.
• Apply styles like cinematic, vivid, or retro looks.

If you’re familiar with Google Photos’ Magic Editor, you’ll notice some similarities. The main difference lies in how you initiate the edit. Magic Editor is tap-and-drag, while Gemini is prompt-and-iterate, with a helpful assistant that suggests options and keeps the conversation flowing. Google highlighted both approaches at I/O 2024 when teasing Ask Photos and other AI-first experiences built on Gemini and DeepMind’s latest models (Google) and (Google I/O recap).

How It Works Under the Hood

Google typically doesn’t disclose every component, but the details that are available provide a clear picture:

• Understanding your request. Gemini 1.5, featuring long-context reasoning, interprets your prompt and analyzes the photo content to plan an edit (Google).
• Generating pixels. DeepMind’s Imagen 3 model facilitates high-fidelity synthesis and inpainting for photorealistic changes, while managing fine textures and lighting (DeepMind).
• Safety and provenance. SynthID applies an invisible watermark to AI-generated outputs, and Google’s policies restrict certain sensitive edits, particularly around realistic faces and copyrighted logos (DeepMind) and (Gemini policies).

Availability and Pricing

Google is rolling out updates in phases. Access may vary based on country, device, and account type. Some advanced generative features have historically rolled out first to Gemini Advanced or Google One AI Premium subscribers (Google One AI Premium). If you don’t see the new editing prompts yet, make sure to update the Gemini app and check back as server-side rollouts continue.

How to Try It

1. Update the Gemini app on Android or iOS, or visit gemini.google.com on the web.
2. Start a new chat, tap the attachment icon, and add a photo.
3. Describe your edit using clear language, e.g., “remove the trash can and warm up the lighting.”
4. Utilize Gemini’s suggestions for refinement, or explore alternative styles.
5. Export the result. Gemini typically saves a new copy and leaves your original photo unchanged.

How It Compares to Google Photos’ Magic Editor

Think of these tools as complementary:

• Magic Editor in Google Photos excels in precise, touch-friendly tweaks with on-device cues and cloud support. It’s ideal for quick fixes and arranging elements (Google Photos).
• Gemini excels when you want to explore creative possibilities conversationally, try multiple variations, or combine edits into a single request. It’s also useful during a broader planning or storytelling discussion with Gemini.

Bottom Line

By merging Gemini’s reasoning with DeepMind’s imaging research, Google is transforming complex photo edits into an intuitive back-and-forth interaction. This means less fiddling and more creativity, along with safer sharing thanks to watermarking and policy safeguards. If you currently rely on Magic Editor, you don’t need to choose—just expect to ask Gemini for the heavier editing tasks.

FAQs

Do I need a paid plan for these edits?

While many features of Gemini are free, some advanced generative tools may require Gemini Advanced or Google One AI Premium in certain regions. Check your account settings for eligibility here.

Are edited images watermarked?

Google applies SynthID watermarks to AI-generated content to support authenticity and detection. These may also show on-screen disclosures when you share or export edits (DeepMind).

Can Gemini edit faces?

There are restrictions around highly realistic face edits, public figures, and sensitive content. If an edit is blocked, try broader scene changes or ensure consent for personal photos. See Google’s policies here.

Is this different from Magic Editor in Google Photos?

Yes. Magic Editor is tap-based, while Gemini serves as a conversational assistant that understands instructions and suggests edits. Both can be used on the same device.

Does it work offline?

No. Generative edits require an internet connection, as they run in the cloud.