No worries if Kelin charges now, another free video generation tool has arrived. Previously, the highly anticipated video generation model Vidu by Shengshu Technology is now officially live globally. Users can register and log in directly with their email, without the need to queue for approval. Upon successful registration, users will receive 80 credits.
This AI video generator is not only comprehensive in functionality but also easy to operate. Users can effortlessly create high-definition videos of 4 or 8 seconds, with a resolution up to 1080P, meeting various high-standard video production needs.
Key Highlights of Vidu:
Fast Generation: Vidu achieves the industry's fastest inference speed, generating a 4-second video clip in just 30 seconds, twice as fast as the industry standard.
High Realism: Whether it's anime or realistic style, Vidu can generate vivid and lifelike images, with natural and smooth character movements, and no breakdown in scenes with large movements.
Character Consistency: Vidu supports character consistency, allowing users to upload a character image and specify that character to perform any action in any scene, making the creation of memes and emojis effortless.
Multiple Styles Supported: In addition to realistic styles, Vidu also supports anime style video generation, with a Miyazaki-like art style, rich in imagination.
Direct Text-to-Video Conversion: Due to innovative underlying architecture, Vidu's creations feel more seamless, with videos generated continuously from start to finish, without any frame insertion traces.
Wide Application Scenarios: From game production and film post-production to education and training, Vidu provides strong support.
Compared to products like Kelin and Luma, Vidu's main feature lies in its introduction of character consistency and anime style as its two major unique functions.
Here, AIbase directly tested previously generated flat illustrations, which are difficult to convert into videos on platforms like Kelin, often resulting in broken faces.
The operation interface of Vidu is simple: just upload the image and select its purpose. Here, I chose to use it as the starting frame without altering the original background, and then clicked to generate.
Prompt: A boy joyfully splashing in a puddle, with the rain getting heavier.
Test results are as follows:
It can be seen that Vidu is relatively stronger in anime style video generation compared to Kelin, with normal character movements and no significant breakdowns. Except for the last frame which slightly deviated from the prompt, the previous parts are basically usable.
To verify how strong Vidu is in anime, AIbase also tested a "traditional challenge": an ancient-style anime character. Yesterday, this image was tested on Kelin and Luma, and the results were not satisfactory. Ancient-style anime characters have always been a weak point for video generation models.
Prompt: The boy reaches up to adjust his hat, suddenly smiling.
Test results are as follows:
It can be seen that the process of the ancient-style character moving is overall coherent, and the hands and face are not significantly broken. However, the character is slightly uglier compared to Kelin, which is better at maintaining the integrity of ancient-style illustrations when converting to video.
It is worth noting that Vidu does not support multiple tasks simultaneously like Kelin. If your previous video has not been completed, the next video generation task cannot be started.
After testing two videos, AIbase attempted to operate again and was prompted that there were too many tasks, with a maximum of one task ongoing at a time. Is there a limit of only two generations per day for free users?
Although the official claims that Vidu can generate a 4-second video clip in just 30 seconds, in AIbase's actual testing, the generation time for one video is at least 2-3 minutes.
Those interested can try it themselves, product address: https://top.aibase.com/tool/viduguanwang
Vidu is developed by a team led by Professor Zhu Jun from Tsinghua University, based on a fully self-developed U-ViT architecture. This architecture is the first in the world to integrate Diffusion and Transformer, proposed earlier than the DiT architecture used by Sora.
The innovation of Vidu lies in its ability to achieve direct and continuous text-to-video conversion, avoiding traditional multi-step processing such as frame insertion and splicing, making the generated videos smoother and more natural in perception.
Shengshu Technology was founded in March 2023. The team of Shengshu Technology consists of core members from the Tsinghua University Institute for Artificial Intelligence, who have a deep background and strength in the research and development of multi-modal general-purpose large models. Since its establishment in 2023, the company has received recognition from well-known industrial institutions such as Ant Group and Qiming Venture Partners, and has completed several hundred million yuan in financing, becoming the most highly valued entrepreneurial team in the domestic multi-modal large model track.
