It’s crazy, Gemini can edit pictures with text prompts

Google updated Gemini 2.0 Flash Experimental yesterday  , and its multimodality has been greatly improved. In one sentence, it can continuously edit and adjust a picture with language, and the style can be consistent, and the image will not be distorted . The product manager of Google AI Studio announced that they can also directly analyze the video link .

I tested some cases below, and the results are really amazing. I can continuously make changes to a picture, and I can also directly input a video link and identify what the video in the link is about.

Put a necklace on the beauty

Upload a photo of a beautiful woman, then give some instructions. The first necklace generated is not a pearl necklace. Then give new instructions and the necklace is changed to white pearls. The effect is great!

So, can we combine two photos, such as a real product photo, and then put it on a model, the effect is amazing! !

It can accurately identify two pictures and combine them according to the input requirements. I just made one request: wear the necklace in the first picture on the neck of the girl in the second picture . This is directly stealing the job of photo editing!

YouTube Video Link Q&A

I found a tutorial link for a Google AdSense website approval video on YouTube and fed it directly to Gemini. He spent a minute summarizing what the video was about.

In order to test whether he really understood the video, rather than just extracting the audio track, I continued to ask: How many people appear in the video, what clothes are they wearing?  The answer was very accurate! It can be seen that he really understood the content of the video.

How to use

To access Google AiStudio, you need to have a US IP address , not a home broadband one, otherwise it will not work. The address is as follows:

https://aistudio.google.com/

Select Gemini 2.0 Flash Experimental and choose the output format as Images and text