Gen-3 Alpha Review - Making Creatives in AI

Content

1. What's new in Gen-3
2. Gambling creatives through Gen-3
3. All the features of RunwayML
4. Free version
5. Paid plans
6. RunwayML Ecosystem
7. Disadvantages and limitations
8. Conclusion

In 2024, Runway reaffirmed its status as a leader in content generation technology by introducing a new neural network for video creation - Gen-3. This tool is the next step in the evolution of generative AI, and promises to revolutionize the video creation process.

What's new in Gen-3?

RunwayML Gen-3 is the latest version of the neural network, which has significantly improved content generation capabilities compared to Gen-2. Gen-3 introduces improved algorithms to create more realistic and detailed images and videos. Key differences between Gen-3 and Gen-2 include:

1. Improved detail: Gen-3 is better able to handle fine details such as textures and complex shapes that may have previously looked blurry or distorted.
2. More natural-looking animations: In Gen-3, animations look smoother and more realistic thanks to improved motion processing algorithms.
3. Support for Complex Prompts: Gen-3 interprets complex text queries more accurately, allowing you to create content with a greater level of control and precision.
4. Performance Stability: The new version has improved generation stability, which reduces the likelihood of artifacts and errors.

These improvements make RunwayML Gen-3 a powerful tool for professionals working with visual content, making it faster and easier than ever to create high-quality images and videos.

Gambling creatives through Gen-3

Last year we already tested generating creatives using Gen-2 neural network and decided to check how Gen-3 has changed during a year of development. We will repeat the same prompts on similar images as last time, as well as create new ones.

All videos will be in 16:9 format, since the trial version doesn't allow you to make 1:1 videos.

Read about Gen-2

Let's start with an assessment of how the neural network coped with generating creative for betting. A static picture of a soccer match with Brazilian flags.

As you can see, the generation quality has improved significantly. Whereas before the image could be slightly blurred, now it looks much more realistic. Additionally, at the end of the creative, the neural network automatically added the Brazilian flag, and this was done without any request.

By slightly supplementing the image with text and generating a new 10-second snippet with a call to action, the creative can be considered complete and ready for use. And with more careful customization of the prompt, even more impressive results can be achieved.

Next up was a creative with a woman and a Brazilian flag, which we also tested last year.

The video turned out realistic again. However, the only drawback was the image of the money - it is still difficult for the neural network to draw fine details. This may be due to the fact that the original picture was also generated by AI.

Next in line is a new static creative generated with the help of a neural network. We uploaded it to the neural network, but got a slightly cropped result. Nevertheless, it is enough for the test.

The creative turned out to be dynamic, and even the lettering was preserved. In the last version there were problems with text distortion, but now Gen-3 recognizes inscriptions perfectly. Thus, in 10 seconds you got a high-quality motion design. The only thing the neural network doesn't handle perfectly yet is detailing.

And lastly, a creative created with a live photo.

At first glance there is nothing special about this creative, but there is one interesting point. You can notice that the neural network recognized that the image shows a photo and started generating a live video, as if fixing the image on the monitor.

This suggests that in the near future, and partially already now, it will be possible to create live-creatives with the help of neural networks.

Verdict

Compared to the previous version, Gen-3 performs significantly better. We can say that 50% of success when working with a neural network depends on your experience and quality of prompts. We haven't tried all the features of Gen-3 yet, but even this limited experience shows how to integrate it effectively into the workflow.

All features of RunwayML

RunwayML is a platform that provides generative design and machine learning tools that enable users to create artificial intelligence (AI)-based content. Here are the main features and functions available on the RunwayML platform:

1. Video and image generation (Gen-1, Gen-2, Gen-3)

• Gen-1: Tools for generating videos and images based on text queries. This is an early version that allowed for simple creatives.
• Gen-2: An improved version with text and video capabilities. Supports higher resolution and improved visualization quality.
• Gen-3: The newest version with improved detail, dynamic effects and the ability to handle more complex queries. Supports animation and texture generation.

2. Video and Image Editing

• Inpainting: Allows you to fill or modify parts of images by removing or adding objects.
• Keyframes (Keyframe): Ability to create and edit keyframes for animation.
• Text Processing: A tool for generating text from data and examples, including creating descriptions, articles, and scripts.

3. Model Training

• Custom Models: Ability to train your own machine learning models based on your data.
• Pre-trained Models: Use pre-trained models to accelerate the development and content creation process.
• Style Transfer: Style Transfer, which allows you to transfer styles from one image to another.

4. API and Integration

• API Access: Allows you to integrate RunwayML capabilities into other applications and workflows through APIs.
• Integrations with other platforms: Supports integration with popular tools such as Adobe Photoshop, After Effects, and other design and editing programs.

5. Video Effects and Filters

• Motion Tracking: Ability to track motion and add effects in real time.
• Color Grading: Tools for color correction and video styling.
• Green Screen: Remove background and replace it with any other background in real time.

6. Text and Sound Generation

• Text Generation: Create text, including dialogs, descriptions, and scripts.
• Sound Generation: Generating AI-based sound effects and music compositions.

RunwayML offers flexible pricing plans that allow users to choose the appropriate level of access to neural network capabilities, depending on their needs.

Free version

RunwayML provides a free version that is great for familiarizing yourself with the platform and basic tasks. As part of the free plan, users receive a limited number of credits to work with the content generation tools.

• 550 pre-registration credits for video generation (~110 seconds)
• 100 image generation

This plan is ideal for those who want to test the capabilities of the neural network before purchasing a paid plan.

Paid Rates

For those who plan to use RunwayML on a professional level, paid plans are available which include:

1. Starter Plan ($12/month): This plan offers more credits and access to advanced models and features, including enhanced video and image tools. Suitable for freelancers and small teams.
2. Pro Plan ($28/month): This plan is geared toward professionals and companies, offering even more credits, priority support, and access to exclusive tools, including new model versions like Gen-3. More options for integration and process automation are also available.
3. Enterprise Plan ($125/month): Customized plan for large companies and projects with special requirements. Includes maximum usage limits, customized support, advanced integration options, and model customization for specific business needs.

Each paid plan is designed for different usage levels, from basic to advanced, allowing users the flexibility to choose and pay only for the features they really need.

RunwayML Ecosystem

One of the main advantages of Gen-3 is its integration with other Runway products. Users can start a project with Gen-3 and then perfect it with the company's other tools. This creates unprecedented content flexibility and saves time.

In addition, Gen-3 provides opportunities for professionals that previously required significant resources. For example, creating promotional videos, visual effects for movies, or even animations for educational purposes can now be done in minutes.

Disadvantages and Limitations

Like any new technology, Gen-3 is not without its drawbacks. While the videos created are of high quality, they may not always match the user's artistic vision. Generating complex scenarios with a large number of variables can sometimes lead to unpredictable results that require manual adjustments.

In addition, the technology still requires significant computing power. Users with limited resources may find it difficult to use Gen-3, especially when working at high resolutions.

Conclusion

Gen-3 from Runway is a step forward in the world of content generation. It offers many opportunities for creative professionals and enthusiasts, simplifying the process of creating high-quality video content. Despite some limitations, this technology promises to be an integral part of the future of digital production.