Kuaishou's Klings AI officially launched the "Subject Library" today, adding "long-term memory" capabilities to the O1 multimodal video model. Users can upload a single character image, and the system will generate 3D perspective completion, multiple lighting variations, and support cross-scene "@character name" one-click invocation. The official claims that "subject consistency is over 96%, completely saying goodbye to AI face swapping."
From Single Image to 3D Memory - Three Steps of the Subject Library
1. Upload: Supports JPG/PNG/RAW, automatic background removal, alignment, and color normalization
2. Completion: AI generates side faces, back views, and local details, providing three options for selection
3. Invocation: Input the prompt word "@character name" to maintain the same face and clothing details in any shot, lighting, or style
AI Smart Description - Let the Model Understand Your Creativity
- The system automatically extracts "hair color + clothing + style" and generates a keyword within 60 words, which users can edit again
- Experiments show that after using smart description, the success rate of generating complex scenes increases by 27%, saving an average of 12 minutes of manual tuning
O1 Model Unified Entry - Seamless Integration of Text, Images, and Video
The Subject Library shares the same underlying latent space with O1's "Text-to-Video", "Image-to-Video", and "Start-End Frame Control", achieving:
- The same character appears continuously across multiple videos, with ID drift less than 0.03
- Supports 48fps/1080p output, with a maximum length of 5 minutes per video
- Official statistics show that since its launch in 2024, Klings AI has been updated more than 30 times, generating a total of 200 million videos
Industry Impact - Short Videos Enter the "Consistency" Era
- Film Preview: Production companies can use the subject library to lock the protagonist's appearance, quickly generate storyboards, and reduce the cost of re-shooting on location
- Cross-border E-commerce: Merchants can upload a model image once and batch-produce multilingual try-on videos, reducing production costs to 1/10
- Virtual Icons: IP holders store V-avatars in the library, allowing fans to generate fan-made videos by "@character", without worrying about "face swapping" anymore
Open Plan and Pricing
- Free Version: Can store up to 5 subjects, with 50 invocations per month
- Pro Version: 29 yuan/month, unlimited number of subjects, includes 600 invocations and 5 minutes of high-definition generation
- Enterprise API: Charged by usage, 0.005 yuan per use, supports private deployment and brand customization
Next Step: Multi-person Collaboration and Real-time Styling
Klings AI revealed that in Q1 2025, it will launch the "Multi-person Subject Library" and "Real-time Styling" - allowing three characters to be locked in the same frame, and switch global styles such as cartoon, retro, and cyberpunk in real time, further penetrating long-form dramas, advertisements, and game animation pipelines.
Editorial Opinion
