Google's Gemini 2.0 Flash Can Edit Images with Text Instructions

Google has a model that can edit images using plain text instructions, write legible text inside generated images, and composite multiple images together. The catch: the model is buried inside Google AI Studio under a confusing name, and finding it requires navigating a maze of similar-sounding model options.

How to Access It

Go to aistudio.google.com and select Gemini 2.0 Flash Experimental from the model dropdown. Be careful to pick that exact model variant. The other options with similar names do not have the same image editing capabilities.

What It Can Do

Once you have the right model selected, you can do things that previous image generation models could not handle reliably.

Edit images with text. Upload a photo and give it a text instruction like "put this person in a suit." The model will make the specific change you asked for without regenerating the entire image. You can also tell it to recolor elements, like changing a dragon from orange to green, and it will modify only what you specified.

Composite images together. Take a picture of a croissant and a picture of a person, then tell the model to combine them so the person is holding the croissant. The result looks natural. This is useful for product photography, where you could have a model hold an item without staging a full photoshoot.

Write legible text in images. AI image generators typically produce gibberish when they try to include text. This model can render short phrases and equations correctly inside the image, as long as you specify the exact text you want. If you leave the text content up to the model, it still produces nonsense.

Key Takeaways

• Gemini 2.0 Flash Experimental in Google AI Studio can do targeted image edits from text prompts, a genuinely new capability
• It composites separate images together in ways that look natural, useful for product shots
• Text rendering inside images works when you specify the exact words, but fails when the model generates text on its own
• Google's model naming remains a mess and the tool is harder to find than it should be

Resources

• Google AI Studio -- where to access Gemini 2.0 Flash Experimental for image editing