Keling 3.0 Introduces Advanced Shot Composition and Multilingual Video Generation

Keling has launched a significant update, advancing its video generation capabilities from version 2 to 3.0. Early access users have reported substantial improvements, particularly in shot composition and multilingual support, which they describe as setting a new benchmark for video models.

The new Keling 3.0 model demonstrates enhanced control over video sequences and integrates complex linguistic elements. One example highlighted a 15-second video of a rock band at a music festival, generated from a single prompt, featuring six distinct shots with varied sizes and camera movements. Another showcased a multilingual dialogue sequence where multiple characters spoke different languages, with accurate pronunciation and contextual alignment, all from a simple prompt. These examples suggest a significant leap in the model's ability to follow instructions and manage intricate video elements.

Video editing software showing a multi-shot rock band video and multilingual audio tracks.

Enhanced Shot Composition Features

Keling 3.0 allows users to generate videos between 3 and 15 seconds, with options to specify the number and duration of individual shots. The platform offers two primary methods for shot composition: intelligent and custom.

Intelligent shot composition automatically generates a series of shots based on a user's prompt. For instance, a simple instruction to depict a human fighting a mech, emphasizing "shot/reverse shot" and the human's "indomitable spirit," resulted in a dynamic video. The system independently filled in the specific shots, illustrating the size disparity between the combatants and the human's resilience.

Custom shot composition provides more granular control, enabling users to define each shot's details. An example used the music festival scenario, where the user pre-visualized specific shots—bassist, drummer, lead singer, audience, and a wide shot with fireworks—and then specified shot size, content, and sound effects to produce the desired video.

Keling 3.0 also addresses a long-standing challenge in video models: the smooth execution of shot/reverse shots. This feature is not limited to dialogue scenes; it has been successfully applied to action sequences like a tennis match. This capability is also useful for creating video demonstrations, such as a short commercial for VR glasses, where Keling can generate a rough visual and auditory effect display.

Dynamic shot/reverse shot sequence of a tennis match, illustrating Keling 3.0's capability.

Furthermore, Keling 3.0 supports video generation directly from storyboards. Users can create an AI-generated storyboard, then input it along with a prompt into Keling 3.0 to produce a complete video. Examples included a "backrooms"-style scene and a Nordic-themed video, both demonstrating high-quality visual and auditory output.

Advanced Multilingual Capabilities

A key highlight of Keling 3.0 is its advanced language ability. Beyond generating dialogue in multiple languages, the model can also adapt tones and accents to fit specific contexts and narratives. This allows for the creation of unique educational content, such as "evil cultivation" vocabulary videos, where words are introduced through humorous, context-rich scenarios.

One such example demonstrated the word "morose" through a Valentine's Day scene in Hong Kong, where a woman's Cantonese phrase, "好cheap啊，约会都冇rose" (So cheap, no roses for the date), phonetically linked to "morose" as the man is left feeling gloomy. Another example used a Game of Thrones-inspired setting with a Northeastern Chinese dialect to teach "ambition," where "俺们必胜" (we will win) phonetically represented the word.

This linguistic versatility can be combined with shot composition, enabling the model to generate quick-cut sequences of characters speaking different languages to various customers, demonstrating its utility in diverse applications.

Abstract representation of multilingual communication with glowing speech bubbles and diverse linguistic symbols.

Keling 3.0 Omni for Video Modification

Alongside the Keling 3.0 update, the Omni model has been upgraded to Keling 3.0 Omni. While Keling 3.0 focuses on video generation, 3.0 Omni specializes in video modification. Both models share core capabilities such as 15-second generation duration, intelligent shot composition, and custom shot composition. They also include a subject voice reference option, allowing users to upload audio or select voices from historical videos.

Key distinctions include Omni's reference video length constraint of 3-10 seconds for modifications. Both models support extracting subjects from videos, though 3.0 Omni's video reference function currently does not support subjects extracted from videos.

Users are advised to use 3.0 Omni for video modification and reference-based generation, while Keling 3.0 is recommended for text-to-video, text-plus-image-to-video, or first/last frame generation. The combined capabilities of both models are expected to cover a wide range of video production scenarios. An early user created a short, plot-driven film in under four hours using both Keling 3.0 and 3.0 Omni, showcasing their integrated potential.

Filmmaker's workstation with dual monitors displaying video generation and modification software.

Outlook for AI Video Production

The Keling 3.0 update is described as a comprehensive advancement. Currently, access is limited to "black gold members," with a wider release anticipated soon. This update is expected to simplify and democratize AI video creation, bringing it closer to professional filmmaking standards. Features like improved shot composition, the ability to extract subjects and sounds for reuse, and direct video modification through Omni are poised to significantly boost AI video production efficiency. This development suggests a future where AI tools could empower a broader range of creators, potentially ushering in a new era for AI video.