Zhipu Qingying: Zhipu AI Video Generation Tool 169

Zhipu Qingying is the latest AI video generation tool launched by Zhipu AI. It supports both text-to-video and image-to-video modes, capable of quickly generating 6-second video content at a resolution of up to 1440x960 in just 30 seconds.

Zhipu Qingying is built upon a multimodal large model technology foundation, with its core video generation model being CogVideoX, which adopts a novel DiT architecture that can compress video information more efficiently while integrating text, temporal, and spatial dimensions. In addition, Zhipu Qingying offers a high degree of controllability, allowing users to control the generated video content through language.

Highlights and Features of Zhipu Qingying:

Text-to-Video: Users input text descriptions or scripts, and the AI generates video content based on these textual inputs.

Image-to-Video: Users upload one or more static images, and the AI analyzes the image content to generate dynamic video.

Add Background Music: Supports adding background music to the generated video, including styles such as quiet, light and cheerful, melancholic, epic, and humorous.

Rapid Generation: Compared with other similar products on the market, Qingying can significantly shorten video generation time. Users only need to wait about 30 seconds to obtain a 6-second video, compressing inference time from the previous minute-level to second-level, greatly improving efficiency.

High Controllability: Through refined techniques and its self-developed end-to-end video understanding model, Qingying has improved its ability to follow user instructions, making the generated video closer to the user's creative intent. By describing the scene, camera language, and details in detail, users can achieve satisfactory results with fewer "trial and error" attempts.

Diverse Styles: Users can choose from different styles for creation, including but not limited to cartoon 3D, black and white, oil painting, and cinematic, meeting the demands of personalization and creative expression.

Accessible to All: The Qingying AI agent is open to all users, whether ordinary users or developers, making it easy to use. Enterprises and developers can also integrate text-to-video and image-to-video capabilities into their own applications through API calls.

Built-in Soundtrack: Qingying also provides a soundtrack feature, where user-generated videos can be automatically paired with music, simplifying the video creation process and enabling even users without professional skills to quickly produce vivid and engaging video content.

Technical Foundation: The technology behind Qingying is supported by Zhipu AI's self-developed video generation large model CogVideoX, which adopts a DiT structure similar to Sora, capable of integrating text, temporal, and spatial information. The optimized model's inference speed is 6 times faster than its predecessor.

Technical Features of Zhipu Qingying:

• Rapid Generation: Completes the generation of a 6-second video in just 30 seconds.

• Efficient Instruction Following: Even with complex prompts, Qingying can accurately understand and execute them.

• Content Coherence: The generated video can well replicate motion processes in the physical world.

• Flexible Camera Movement: For example, the camera can smoothly follow the movement of three dogs in the frame.

Industry Applications of Zhipu Qingying:

• Advertising Production

• Film Editing

• Short Video Production

• Creative Content Creation

How to Use Zhipu Qingying to Generate Videos?

1. Access the Platform: Visit the Zhipu Qingyan official website at chatglm.cn/video, register, and log in to your account.

2. Choose the Generation Mode: Select the "AI Text-to-Video" or "AI Image-to-Video" function in the Qingying AI agent.

3. Input Prompts:

• Text-to-Video: Input detailed text descriptions or scripts. The more detailed the prompt, the higher the quality of the generated video. For example, instead of inputting "a little boy drinking coffee," it is better to input "camera panning, a little boy sitting on a park bench holding a steaming cup of coffee. He is wearing a blue shirt and looks very happy, the background is a tree-shaded park, sunlight filtering through the leaves onto the boy."

• Image-to-Video: Upload one or more static images and select the image style. You can use the AI agent provided by Qingying to generate prompts, ensuring accurate and detailed descriptions.

4. Select Video Style and Parameters: Choose the desired video style, including cartoon 3D, black and white, oil painting, cinematic, etc. You can also set the emotional atmosphere and camera movement.

5. Add Background Music: Qingying supports adding background music to the generated video, including styles such as quiet, light and cheerful, melancholic, epic, and humorous.

6. Generate Video: Click the Generate Video button, and the video result will be generated in about 1 minute of queue time. Qingying can complete the generation of a 6-second video within 30 seconds.

7. Optimize Prompts:

The prompt structure should be as clear as possible. A complex formula can include camera language, lighting, subject description, subject motion, scene description, and mood/atmosphere. For example: "camera panning, a little boy sitting on a park bench holding a steaming cup of coffee. He is wearing a blue shirt and looks very happy, the background is a tree-shaded park, sunlight filtering through the leaves onto the boy."

Compared with other AI video generation tools, Zhipu Qingying can generate 6 seconds of video content within 30 seconds, which is much faster than many other AI video generation tools. For example, Kling AI, currently in the first tier, generally takes 2 to 5 minutes to generate a single 5-second video.

During the initial testing period, Zhipu Qingying offers API access for enterprises and developers, while providing free trial services to all C-end users.