The detailed configuration of the base model’s architecture is as follows:<\/p>\n
\n
\u2192 Scroll horizontally to see all columns<\/p>\n
\n\n\n| Model<\/th>\n | Dimension<\/th>\n | Input Dimension<\/th>\n | Output Dimension<\/th>\n | Feedforward Dimension<\/th>\n | Frequency Dimension<\/th>\n | Number of Heads<\/th>\n | Number of Layers<\/th>\n<\/tr>\n<\/thead>\n |
\n\n| 14B<\/td>\n | 5120<\/td>\n | 16<\/td>\n | 16<\/td>\n | 13824<\/td>\n | 256<\/td>\n | 40<\/td>\n | 40<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n In evaluations conducted by the publishers, multidimensional verification spanning 14 primary dimensions and 26 sub-dimensions was performed using 1,035 internal prompts. These results were calculated using human preference-based weighting, and it has been confirmed that the model exhibits superior performance compared to existing models.<\/p>\n Additionally, the following features are cited as technologies supporting the quality of the base model:<\/p>\n \n- High-Quality Visuals and Motion Dynamics: Capable of generating videos with high-quality visuals and significant motion dynamics.<\/li>\n
- Advanced Text Generation: It is reported to be the first video model capable of generating both Chinese and English text within videos.<\/li>\n
- High-Performance Video VAE: Adopts a proprietary
Wan-VAE<\/code>, making it possible to efficiently encode and decode 1080P videos of arbitrary length while preserving temporal information.<\/li>\n- Resolution Flexibility: Supports video generation at both 480P and 720P resolutions.<\/li>\n<\/ul>\n
This adapter, LongLive-Plug-Wan2.1-T2V-14B-few-step<\/code>, achieves efficiency through a reduced number of generation steps by applying distillation techniques to these powerful base models.<\/p>\nStrengths and Use Cases<\/h2>\nThis model excels at performing fast video generation with a reduced number of steps while maintaining the performance of the base model Wan-AI\/Wan2.1-T2V-14B<\/code> in text-to-video tasks. It is designed as an adapter leveraging PEFT, LoRA, and distillation techniques, and is intended to be used in combination with the corresponding CFG LoRA<\/code>. Furthermore, it contributes to efficient video generation by taking advantage of the base model’s capabilities, such as generating both Chinese and English text, visual expressions with high motion dynamics, and 480P or 720P resolutions.<\/p>\nHow It Differs from Similar Models<\/h2>\nWhile LongLive-Plug-Wan2.1-T2V-14B-cfg<\/a>, which has already been introduced on this site, aims to perform distillation for conditional-only video generation using Classifier-Free Guidance (CFG) to reduce the execution of unconditional inference branches, this new LongLive-Plug-Wan2.1-T2V-14B-few-step<\/code> differs in that it functions as a LoRA adapter designed to achieve video generation in fewer steps. Both target the same base model, Wan-AI\/Wan2.1-T2V-14B<\/code>, and it is recommended to use the few-step and CFG adapter weights together at a ratio of 1 : 0.5<\/code>.<\/p>\n<\/p>\n Distributed Files<\/h2>\nWeight files published in Efficient-Large-Model\/LongLive-Plug-Wan2.1-T2V-14B-few-step<\/a>, listed by this site from the Hugging Face API. Sizes are the actual file sizes.<\/em><\/p>\n |