Abstract:
On November 6, Google officially launched the video generation model Gemini Omni 1.2 Flash on Thursday. The model supports native 15-second video generation and editing through multiple rounds of dialogue. The price is 30% lower than the previous generation.
Gemini Omni 1.2 Flash
Google said that Gemini Omni 1.2 Flash is being rolled out through Gemini API, Google AI Studio, Google Cloud Console, Flow, Google Vids, Gemini applications for Google AI Ultra subscribers, and Google Ads. Each generated video is watermarked with an imperceptible SynthID to indicate its origin.
The length of video clips generated by the model can now be set from 3 to 15 seconds, up from the previous maximum of 10 seconds. At the same time, the price of 720p video generation has been reduced by 30% to $0.07 per second. Users can also edit the generated video through ongoing conversations.
One of its standout features is what Google calls "Conversational Editing." Users can gradually adjust the generated video through multiple rounds of dialogue, such as changing the background, adjusting the lighting, or modifying the aspect ratio, without regenerating the entire video. Google says this feature supports editing of full 15-second videos.
Combined with hierarchical workflows, this feature is also expected to reduce production costs. Users first generate a first draft at 360p or 720p resolution, then correct individual details through subsequent editing, and finally only upconvert the final film to increase the resolution to 1080p or 4K.
Google also stated that the model’s ability to follow reference materials has also been significantly improved. According to the official announcement, users can enter character images, product photos, short video clips of up to 3 seconds, and audio prompts. The model can maintain the consistency of characters' faces, clothing, text on product packaging, and brand color matching under different shooting angles.
The ability to follow material has been significantly improved
Google takes a virtual try-on segment as an example: with just one model photo and product images of three sets of clothing, the model can generate a 15-second continuous scene that includes a 360-degree turn, while maintaining the identity of the model and the cut and drape of each garment.
At the same time, overall visual consistency has also been improved. Google said that compared with the previous generation model, the new model has less visual deviation and deformation during screen movement and lens switching.
In addition, Gemini Omni 1.2 Flash can generate smoother and more natural conversational speech, the synchronization effect of mouth shape and speech is more accurate, and it also has the ability to distinguish multiple different speakers.
The model is also better able to handle complex, continuous camera movements, such as FPV drone flight shots or fast zoom shots with strong visual impact, and significantly reduces unnecessary jump cuts.
Comments