Than the generative modelss are achieving increasingly shocking images and videos is no surprise. A few years ago, producing something similar would have required a great manual effort. But it should not lose sight of a reality: although this technology advances at high speed, it still drags important limitations that hinder its use in certain professional environments.
One of the great challenges for creators is to maintain visual coherence in the results. In ambitious projects, the generative AI usually has difficulty representing consistently to the same character or object in different planes or environments. Although changes may seem subtle, they are not always easy to manage. Of course, there are those who work to solve it.
Runway Gen-4 arrives with coherence as a key piece
Runway returns to the load. The Startup based in New York has presented Gen-4its new video generation model. The great novelty: a significant improvement in visual coherence, one of the weak points of models such as Openai Sora. With this advance, the idea is that creators can build longer narratives without fighting with visual jumps between plans.
Gen-3 Alpha, the previous model, already offered a notable level of detail and control. But with Gen-4, the proposal rises level. It is now possible to generate characters, locations and consistent objects throughout multiple scenes. Once the appearance, style and atmosphere are defined, the model maintains those elements unchanged, frame by frame, and in multiple scenes.
Gen-4 allows you to use visual references combined with instructions to generate images and videos with styles, locations and topics coherent. In addition to this continuity, the model seeks to highlight for its ability to create dynamic videos, with realistic movements and a remarkable consistency in characters, objects and style. According to Runway, his understanding of the world is “the best of his class.”
The best example is found in the video that accompanies this article, one of the many shared by Runway to show Gen-4 capabilities. If we refine our eyes, it is evident that these are synthetic scenes, both in the image and in movement. Even so, it represents a remarkable leap for audiovisual productions that are willing to accept this type of aesthetics.
Gen-3 Alpha had already made its appearance in productions such as ‘House of David’, by Amazon Prime Video, and in Puma ads. Now, GEN-4 begins to deploy in Runway’s payment accounts, allowing to generate five-second and ten seconds clips In 720p resolution. Prices range from $ 15 to $ 95 per month, with an Enterprise option for companies, whose cost has not been revealed.
Images | Runway
In WorldOfSoftware | The fever for the chatgpt images is “melting” the OpenAi GPUs: the company has already begun to limit access
