Enhancements in Action: Exploring Image Integration in Runway's Gen-2 AI Video Generation Tool
Last week, I experimented with another AI tool for text-to-video, Pika Labs, to explore the current landscape of AI video generation tools. After comparing Pika Labs' and Runway's tools, I concluded that Runway's tool surpassed Pika Labs' due to its seamless functionality, preview options, and the AI's proficiency in comprehending and generating from text. However, I haven't utilized Runway's tool since July, and there have been updates worth exploring, especially as I gear up for my upcoming AI commercial.
During my initial attempts with this tool in June, users were constrained to the AI-generated output, as the tool incurred credit costs, and new users were only allocated 125 free credits. While creating my most recent video in July, Runway introduced a free preview button. This enhancement allowed users to view four still images without using credits, enabling them to choose the best one for further generation or alter their prompt and preview until they found an image aligning with their vision. This update significantly improved the video creation process, allowing me to create my film exactly as envisioned without forcing the given generations to work as I had done previously.
Today, the tool has been rebranded as "Text/Image to Video," providing users with options to use text alone, allowing the tool creative freedom; an image as the starting point for video generation; or an image paired with a description outlining the scene and its dynamics. However, it's important to note that the free preview option is exclusive to the text-only input, as the preview is considered the image for other input options. Additionally, users can now customize the general intensity of their video's motion (on a scale of 1-10), as well as the camera movement and intensity. Users can specify horizontal (left or right), vertical (up or down), rolling (left or right), and zoomed (in or out) movements, enhancing the ability to create envisioned content quickly.
To test these new features for my upcoming fragrance commercial, I inputted a photo of a perfume bottle into the "image + description" option with the prompt "A glamorous shot of a perfume bottle with sand drifting by and lightning striking in the sky." I adjusted the speed to setting 4 and opted for a zoom-in similar to a typical fragrance commercial.
The resulting generation, as visible below, adhered to the zoom and speed settings I specified but overlooked my prompt regarding sand movement and lightning. While the output could serve in a pinch due to its glamorous appearance, the specific elements I mentioned are crucial for creating a dramatic fragrance commercial and are essential for an accurate representation.
If Runway incorporated a free preview button for image inputs, this issue could be resolved, allowing users to confirm prompt inclusion before using credits. For now, I rate the image prompt addition to this tool relatively low, as it struggles with "image + description" prompts and lacks free preview capability. Nevertheless, I remain optimistic that future improvements will address these issues. I intend to revisit the image prompt option once I finalize the vision for my fragrance. If it proves unsuccessful again, I'll resort to the text-only prompt for my commercial.
Sources:
Comments
Post a Comment