Google's Gemini 2.0 Flash Can Now Edit Images In Natural Language
Gemini 2.0 Flash now has native image generation capability.
Google’s Gemini 2.0 Flash model now lets you make image edits using natural language natively. Unlike earlier multimodal systems that relied on pairing separate models (such as using a language model together with Imagen 3 for image generation), Gemini 2.0 Flash handles multimodality by generating images directly within the same system that processes text. This eliminates the need for inter-model communication, reducing latency significantly.
Since Gemini 2.0 Flash no longer depends on Imagen 3, it offers faster responses and smoother interactions. Plus, you can even embed longer text directly onto images!
Check out this example where I transformed Google Deepmind’s CEO, Sir Demis Hassabis, into a long-haired dude.
Here’s another example showing Gemini adding a chocolate drizzle to plain croissants.
This is mind-blowing because no other aspect of the original image was changed except for the added chocolate drizzle — which, by the way, looks incredibly realistic.
Here’s how it works
To get started, head over to Google’s AI Studio, log in with your Google account, and set the model to Gemini 2.0 Flash Experimental. Make sure also that the Output format is set to “Images and text”.
Then, upload your image file by clicking on the “+” button at the bottom-right corner of the prompt field. To illustrate, here’s a playful edit I made to an image of a fox. I dressed him in a puffer jacket because he might feel chilly up there in the icy mountains!
As you can see, Gemini allows you to precisely guide the AI to update only specific parts of the image. It doesn’t generate a completely new image from scratch but instead modifies only what you explicitly mention in the prompt.
Compared to image generators like Grok 3 and Gemini 2.0, Flash has a clear advantage when it comes to precision and consistency. With Grok 3, if you generate an image and then request edits, the AI creates a completely new image instead of precisely updating the original.
To better illustrate what I mean, let’s use Grok to generate a sample image.
Prompt 1: An image of a rounded perfume bottle with amber color liquid inside, put on a brown table and ambient lighting
Prompt 2: Add text “Generative AI”
Do you see what I mean?
While the final result might look similar, you’ll often notice odd differences or unexpected elements appearing in the image. Gemini, however, lets you guide the AI directly to the specific portions of the image you want to change. Plus, you can continue making precise follow-up modifications on the same image without losing consistency.
Gemini 2.0 Flash Experimental is also capable of adding text to the photo; check out this example where I asked it to add the words “Generative AI Publication”.
The letters are sharp, legible, and well placed. Judging by this result, Gemini might just have the best text-rendering capabilities among current image models — competitors like Midjourney and Flux don’t achieve this level of sharpness or accuracy.
Practical examples and use cases
Let me show you some of the coolest examples I found on the internet where conversational image editing truly shines. Look at the example below where X user A E A E used Gemini 2.0 Flash to colorize a black and white manga.
The final result looks incredibly well made; it even added elements that were not mentioned in the prompt, like the clouds in the background. Nothing in the final image screams AI-generated.
Another cool use-case shown by X user Kurawa Dono is combining two images together. You can upload an image of the product and a model and then ask the AI to make the model hold the product.
If you’re selling products online, this method significantly streamlines your workflow, eliminating the need for extensive photoshoots or manual Photoshop edits.
Here’s another one: you can do style transfer by uploading any image and copying its style to generate a new one. This one was perfectly demonstrated by X user Robert Riachi in his post.
Style transfer isn’t new in the AI image generator world, but doing it in natural language is a completely new and fun experience.
Final Thoughts
Honestly, I thought Google was already lagging behind in the AI race. OpenAI recently released exciting new products like GPT-4.5 and their brand-new AI SDK. Meanwhile, China had another “DeepSeek moment” with the release of ManusAI, making Google’s recent progress seem somewhat less impressive by comparison.
But I was wrong.
Google is very much back in the race. The release of Gemma 3 27B and Gemini 2.0 Flash with native multimodal image generation on the same day is incredibly impressive.
I was blown away by Gemini’s ability to modify images directly through natural language prompts. It’s basically inpainting on steroids, offering a level of precision and flexibility I haven’t seen before.
As an AI enthusiast and a developer who regularly works with AI tools, I can’t help but feel excited by all the possibilities these new models unlock. The opportunities for creativity and innovation are huge.
I honestly can’t wait to get access to the API and start experimenting — building new, interesting, and hopefully useful things. For now, I highly encourage you to try Gemini 2.0 Flash Experimental yourself, play around with the features, and discover your creative use cases.
Hi there! thanks for making it to the end of this post! My name is Jim, and I’m an AI enthusiast passionate about exploring the latest news, guides, and insights in the world of generative AI. If you’ve enjoyed this content and would like to support my work, consider becoming a paid subscriber. Your support means a lot!