Experience Google's impressive AI image generation model

Google has just widely released the feature of generating original images from the Gemini 2.0 Flash language model. With this ability, the model can compose and edit existing images based on text input. Photo: Google .

Unlike other imaging tools that combine diffusion models with large language models (LLM), Gemini 2.0 Flash is multimodal, able to process input and output in multiple formats (text, audio, images, etc.). In theory, this technique improves image quality, allowing the tool to understand context and continue editing within the same conversation. Photo: Google .

To try out the tool, users and developers need to go to Google AI Studio, switch to Gemini 2.0 Flash (Image Generation) Experimental mode. In the Output format section, select Images and text . The interactive area with the tool is in the middle of the screen. Below are some of the main uses of Gemini 2.0 Flash.

Create images and text at the same time. According to Google, Gemini 2.0 Flash supports creating text and images at the same time. For example, you can ask the model to tell a story and draw illustrations. In my experience, the speed of creating images and text is quite fast, averaging 5-10 seconds/image (depending on length and complexity).

Edit photos in the same conversation . Thanks to its contextual understanding, Gemini 2.0 Flash supports feedback and photo editing. If you are not satisfied with the color, object or any detail in the photo, just enter a command and the tool will change it without affecting other elements.

Edit an existing photo . Similarly, simply upload any photo and ask the tool to edit details in the photo, such as changing the color, adding objects, or adjusting the background. Users can give continuous feedback until the tool produces satisfactory results.

Object separation . Gemini 2.0 Flash's object separation ability is quite good, but there are still weaknesses related to human hands. The tool understands Vietnamese meaning, allowing separation and replacement of backgrounds according to many different topics.

Expand/change scene . In this case, the user can ask the tool to shrink the existing image, filling in the gap with a new scene based on the description. Since it is still in the testing phase, the tool sometimes crashes or does not create the desired image.

Create images with a lot of text . According to Google, Gemini 2.0 Flash can create images with long text without spelling mistakes or strange words. This is one of the many weaknesses of other image generation models. However, experience shows that Vietnamese language text is still difficult to read in some places. The tool also cannot translate text in the text without specific suggestions.

Add people to photos . The tool supports adding people to existing photos, with the correct appearance as described. Since it is currently in free trial release, each conversation thread has a limit of about 30,000 tokens. However, users do not have to worry because a question/answer only costs about 300-500 tokens, not too much if you only edit and create basic photos.

Change shooting angle . Users can request to change different shooting angles of the same photo. Of course, the tool supports adjusting different details until creating a satisfactory photo.

Knowledge mining . According to Google, Gemini 2.0 Flash is trained on a large amount of knowledge with reasoning capabilities. For example, you can ask the tool to create a recipe based on existing knowledge, then draw an illustration for easy understanding. Similar to other tools, Google notes that Gemini 2.0 Flash only has general knowledge, not too in-depth or absolutely accurate.

Controversial feature . After its widespread release, many people discovered that Gemini 2.0 Flash can remove watermarks from photos. This ability is not accepted by AI tools like GPT-4o. Since it is still experimental, Google is likely to fix this in the near future. Photo: @deedydas/X .

Source: https://znews.vn/gemini-20-flash-lam-duoc-gi-post1539018.html