AI Visual News
Hunyuan Video by Tencent
- AI Video
- Hunyuan
December 3, 2024. China's Tencent released a powerful open-source video generation AI model, Hunyuan Video.
2024/12/07
- mintbear
https://slashpage.com/mintbear/Hunyuan-01-ref
Korea AI International Film Festival Grand Prize Winner, Mateo Online Screening (Event ended / Please wait for the next screening)
- AI Video
- AI Films
2024/11/30
- mintbear
Hunyuan Video by Tencent
Hunyuan VideoAnother Chinese video-generating AI: HunyuanExcellent prompt understanding and production skillsHunyuanVideo Features ListSpecs: 13 billion parameter open source video generation model, 720p * 5s
PerformanceConjugationSample Video01. Text-to-Video02. Image-to-Video03. Avatar Animation03-1. OpenPose_Motion03-2. OpenPose_Face04. Video-to-Audio04-1. Voice Control04-2. Video DubbingWhat does Hunyuan mean?
Status
2024.12
Summary
December 3, 2024. China's Tencent released a powerful open-source video generation AI model, Hunyuan Video.
Category
- Hunyuan
Tag
- AI Video
Dates
2024/12/07
Created by
- mintbear
SP
https://slashpage.com/mintbear/Hunyuan-01-ref
Hunyuan Video
Mintbear 2024.12.07
December 3, 2024. China's Tencent released a powerful open-source video generation AI model, Hunyuan Video .
Mint Bear 🍀🧸
Another Chinese video-generating AI: Hunyuan
Following the powerful video generation AIs Kling and Hailuo, another strong competitor from China, Hunyuan, has emerged. It is showing interesting results based on an overwhelming amount of training data.
It is an open source video generation model released to companies and individuals, and its efficiency and results are overwhelming through the largest parameters. It is difficult to operate in general local environments because it requires installation specifications, and it must be used in replication, etc. The generated video is 1280*720p, 5 seconds.
Excellent prompt understanding and production skills
The ability to understand text prompts and perform video directing is really strong. The reason for the good directing is said to be due to the dual-stream method of composing text and video separately. (Separate explanation to be made)
Keywords: High quality, dynamic, continuous action, artistic direction, concept implementation, physical laws execution
In addition to text-to-video (T2V) mode, image-to-video (I2V) will be supported, and various functions such as generating audio required for avatar and video creation (V2A) will be included.
HunyuanVideo Features List
1.
Text-to-Video (T2V)
2.
Image-to-Video (I2V) – expected in 2025
3.
Avatar Animation
Audio- based animation
Pose- based animation
Expression- based animation
Hybrid conditional animation
4.
Audio Generation (Video-to-Audio, V2A)
Specs: 13 billion parameter open source video generation model, 720p * 5s
Large-scale model: The largest open-source text-to-video generation model to date with 13 billion parameters , with advanced scaling techniques to reduce computational costs by up to 80%.
High-quality video creation: Create 5-second videos in 720p resolution, creating “hyper-realistic” videos with excellent physical accuracy and scene consistency.
Innovative features: Add automatically synchronized sound effects and background music to videos created with the video-audio synthesis feature , and control avatar animations to manipulate digital characters using a variety of input methods, including voice, facial expressions, and body movements.
Performance
According to expert evaluations, the Hunyuan video outperformed commercial models.
After being evaluated by 60 experts on over 1,500 prompts, it achieved a motion quality score of 64.5%.
It outperforms competing models such as Runway Gen-3 and Luma 1.6.
Conjugation
Tencent's Yuanbao AI chatbot app lets you input prompts in both Chinese and English.
It's free for both business and personal users.
The code and weights for the entire system are available on GitHub for research and development.
Sample Video
01. Text-to-Video
7 Prompt：In the gym, a woman in workout clothes runs on a treadmill. Side angle, realistic, indoor lighting, professional.
8 Prompt：Close-up, A little girl wearing a red hoodie in winter strikes a match. The sky is dark, there is a layer of snow on the ground, and it is still snowing lightly. The flame of the match flickers, illuminating the girl's face intermittently.
9 Prompt：Wide shot: A caravan of camels winds its way through the endless golden dunes, resembling a long snake slithering across the earth. The setting sun paints the desert in deep orange hues, while the sky transitions into a gradient of purples and reds. Close-up shot: The aged guide's wrinkled fingers pick up a handful of fine sand, letting it drift away with the wind. His headscarf flutters gently in the breeze, and his weathered face is bathed in the glow of the sunset, his eyes steady and wise. Cinematic detail portrayal.
10 Prompt：In the style of Dunhuang sculptures, A graceful deity, playing a pipa, dancing lightly in a museum, with flowing garments.
11 Prompt：A person with a computer for a head is writing code in front of a computer, in a realistic style.
8 Prompt：Close-up, A little girl wearing a red hoodie in winter strikes a match. The sky is dark, there is a layer of snow on the ground, and it is still snowing lightly. The flame of the match flickers, illuminating the girl's face intermittently.
9 Prompt：Wide shot: A caravan of camels winds its way through the endless golden dunes, resembling a long snake slithering across the earth. The setting sun paints the desert in deep orange hues, while the sky transitions into a gradient of purples and reds. Close-up shot: The aged guide's wrinkled fingers pick up a handful of fine sand, letting it drift away with the wind. His headscarf flutters gently in the breeze, and his weathered face is bathed in the glow of the sunset, his eyes steady and wise. Cinematic detail portrayal.
10 Prompt：In the style of Dunhuang sculptures, A graceful deity, playing a pipa, dancing lightly in a museum, with flowing garments.
11 Prompt：A person with a computer for a head is writing code in front of a computer, in a realistic style.
02. Image-to-Video
Unreleased. Scheduled for 2025.
03. Avatar Animation
03-1. OpenPose_Motion
03-2. OpenPose_Face
04. Video-to-Audio
04-1. Voice Control
1 Prompt: Advanced scene modeling.
2 Prompt: Natural background motion.
3 Prompt: Expressive and vivid facial expressions and gestures.
2 Prompt: Natural background motion.
3 Prompt: Expressive and vivid facial expressions and gestures.
04-2. Video Dubbing
1 Prompt: Birds chirp and tweet.
2 Prompt: Water is rushing down a stream and pouring.
3 Prompt: A car engine revs.
4 Prompt: Footsteps on wood.
2 Prompt: Water is rushing down a stream and pouring.
3 Prompt: A car engine revs.
4 Prompt: Footsteps on wood.
What does Hunyuan mean?
腾讯混元视频 (Téngxùn Hùnyuán Shìpín)
腾讯 (Téngxùn): Tencent company
混元 (Hùnyuán): “Hùnyuán”
1. origin of the universe
2. the world
1. It means that order and creation emerge from a primitive and chaotic state.
2. Everything is fused into one
3. A philosophical concept meaning the original energy or source of the universe.
1. origin of the universe
2. the world
1. It means that order and creation emerge from a primitive and chaotic state.
2. Everything is fused into one
3. A philosophical concept meaning the original energy or source of the universe.
Video (Shìpín): “video”
See more info
