Google's Generated AI Film Production Program has reached a milestone. The tech giant has confirmed exclusively for CNET that Flow Creator has created over 100 million AI videos in its program. Thanks to the advanced AI video model Veo 3's Flow, users can generate video clips and edit them together to create scenes.
It's been 90 days since Google was surprised by Flow at its annual I/O Developers Conference. Elias Roman, senior director of product management at Flow at Google Labs, has since been spent on “hustlings just to keep up with demand.”
The flow is a departure from Google's previous generation AI work. For years, the company's AI initiatives have focused on Gemini, an all-in-one chatbot. AI flooded the product with AI, like search AI overviews and Gmail's AI-generated summary. NoteBookLM, a research assistant tool with an AI audio generator that can convert documents into personal podcasts, is constantly rolling out new features.
Industry leaders have spent billions of dollars trying to win races to develop the most advanced AI for the average Google searcher, developers, and even artists and creators. 100 million AI videos are a key milestone for the company and help to show what the future of creating AI-enabled features looks like.
Enter AI flow
To compete with Midjourney and stable diffusion, Google created a harvest of an AI image model known as Imagen (pronounced “Imagine”). That previous generation media model was suitable for amateur and enthusiast creators rather than professionals, and did not dominate the creative space of AI. It all changed in VEO 3.
Google dropped its latest AI video model, the Veo 3, at the I/O Conference in May. VEO 3 jumped the competition with a slightly obvious but first-ever advancement in the industry. AI video with synced AI generated audio. The model attracted a lot of attention online, with Google reporting over 40 million AI videos just seven weeks later.
“What Veo 3 allowed was a much wider range of people to create very engaging videos. “To make Foley do it [ambient sounds]sound effects, soundtracks, dialogs, etc., don't let them think about each of the modalities.
VEO 3 is one of several AI models that can be used in filmmaker tools. Flow was built for professional creators and filmmakers. This is a step beyond the simple image and video generation available with Gemini. Google intentionally moved away from the original ImageFX nomenclature and built its interface, Roman said, and wanted a flow to combine the most advanced image and VEO model with Gemini, which was used in VEO training, “basically speaking native veo.”
Flow is one way to combine all these AI models and pieces, integrating Google's various generated AI models for seamless video creation and editing.
Different flows from Veo and Imagen
Flows were built focused on consistency, the ability to maintain visual identity from one clip to the next. If you have a 90-second video of a character drinking coffee in a cafe, it is not desirable to change the length of the hair or color of the eyes every 8 seconds between scenes. That consistency is important for professional projects and is difficult to achieve. Roman called it “Achilles Heal in AI Video.”
Flow has several tools that help you maintain that consistency, and in my tests you can give you a new level of control over what previously lacked in Google AI tools. The best way to explain the flow is the upscale version of the simple video generator interface, with the option to export multiple clips to a timeline like a simplified version of Premier Pro.
AI tools aren't automatically drawn out to automatically use by target audiences, but they are often upgraded in the hopes of becoming more useful to professional creators. Generation AI is a controversial issue in the creative industry, especially when it comes to wholesale creation of text, images and videos. AI enthusiasts may praise the creativity and speed of AI models, but creators continue to express legitimate concerns about AI training and deployment. This is why publishers and artists filed lawsuits against AI companies that allegedly infringing copyright. That's why workers in data-rich industries face job safety concerns as executives try to cut costs.
Another problem with AI is the types of images you can create. Last year, users discovered that Gemini could create images of people of color in Nazi soldier uniforms. Google apologised for what the company calls “inaccuracies in some historical image-generating depictions,” and said it was working to quickly improve those depictions.
(Google guidelines prohibit the creation of abusive and illegal AI content. Roman said improving the implementation of safety policies will be supported by technical updates and real-world use and reporting.
Going forward, Roman said Flow is working to add new features such as expanding the capabilities of VEO 3, improving consistency, and custom-made voices for character work. The North Star of the Project makes creation more accessible to more people.
“We can lower the barriers to prevent wider people from telling stories through videos, and raise the ceiling on what stories can be told through videos,” Roman said. “Like Wild Street interviews and the Yeti ASMR bloggers, some of them will be funny and stupid, and some of them will be really powerful.”
How to use Google flow for AI videos
Flow, part of Google Labs and accessible via the AI Test Kitchen, can be paid to Google AI subscribers with a Pro plan of $20 per month and an Ultra plan of $250 per month (currently discounted at $125 for three months). According to the Google Labs Privacy Notice, “human reviewers say they will 'read, annotate, and process' to improve the AI model of lab interaction and tool output. (Lab data is stored by default for up to 18 months, and the company advises you not to upload or send sensitive information. Google's general Privacy Center has details.)
I spent some time testing the flow, generating clips and sewing together using the scene builder. Some tools are available only to flow users.
From ingredients to video: There are several ways to prompt to generate self-practical text to video and video clips to images/frames. Materials to video materials are new and worth exploring. Using this method, when you upload a specific photo and add a text prompt, Flow will fragment the part into pieces. For example, you can upload a male photo, a product photo of a particular jacket, or a scenic background, and then Flow can combine them to animate the video.
Clip expansion and smoothing transitions: Expansion helps to lengthen the clip. In the Scene Builder timeline, drag the edges of a single clip's frame to the desired length. If you want to generate a new video and require a smooth transition, we recommend that you proceed to the end of the first clip and press the plus button at the top of the marker to save the final frame to your library. You can then use the image at a frame-to-video prompt to maintain its consistency from clip to clip.
Slowly edit: If you are editing a frame or image in another document, you can upload the marked-up image to flow and tell the model to implement the changes. You can also do that with the image you drew. It also brings those doodles to life. This is a feature under development – a new prototype for this is currently in work – but stretching the functionality of such a flow is definitely fun.
Prompt with Gemini: There is no way to automatically create and/or improve prompts directly in Gemini (hope for changes in future updates), but you can create the perfect prompt using a chatbot. If you're struggling to make more detailed ideas come true, get Gemini to help you.
For more information, check out our top AI Image Generators and our guide to writing the best AI Image Prompts.
