The Dawn of Video Generation: Preliminary Explorations with SORA-like Models
1Tencent, AI Lab, 2USTC
*Equal contribution
Seeing is believing. To observe and compare the video quality of recent generative video models, we open non-cherry-picked generated videos with over 700 designed prompts from:
- Vertical-domain applications, such as human-centric animation and robotics, etc.;
- Key objective capabilities, such as text alignment, motion diversity, composition, stability, etc.;
- Ten application scenarios, such as film, advertisment, anime, etc.;
- Challenges and future directions, such as complex motion, interaction, concept understanding, etc.;
Compare all model videos:
On the main page, we concatenate the video results from multiple models for comparison.
The input text prompts are listed below the video (Index-Task: Prompt). Notably, you can download all results from the Download button.
Note: Due to certain reasons, some models cannot generate certain cases. In such cases, we choose to pad it with the mask.
Check separate model videos:
We will continue to update the results with model releases and version updates.