CogVideoX1.5-5B

CogVideoX1.5-5B is an open-source text-to-video generation model analogous to the commercial model QingYing. It is designed to create video based on text prompts, supports the English language, and also offers image-to-video generation (version CogVideoX1.5-5B-I2V). The model is available on platforms such as Hugging Face, ModelScope, and WiseModel.

Features:

• Video resolution: 1360×768 (minimum side 768 pixels, maximum ≤1360, aspect ratio must be a multiple of 16).
• Duration: 5 or 10 seconds (default 5 seconds).
• Frame rate: 16 frames per second.
• Number of frames: Must equal 16N+1, where N ≤10 (default 81).
• Prompt length limit: Up to 224 tokens.
• Supported data types: BF16 (recommended), FP16, FP32, FP8, INT8. INT4 is not supported.

Inference Requirements:

• VRAM: In BF16 mode: At least 10GB (diffusers). In INT8 mode: At least 7GB (diffusers with optimizations).
• GPU: Supports NVIDIA Ampere architecture or newer (A100/H100 for optimal speed).
• Inference speed: On A100: ~1000 seconds for 5-second video (BF16). On H100: ~550 seconds for 5-second video (BF16).

---

The model is a component of the image generation pipeline, consisting of:

• Text encoder: ~4.8B parameters,
• Transformer: ~5.6B parameters,
• VAE: ~216M parameters.

Total: ~10.5B parameters

---

The 'VRAM requirements' line is calculated based on the size of the weights in the specified precision, with an additional +10% coefficient for inference:
  model_parameters_count * bytes_per_parameter * 1.1 / 1024^3 
where model_parameters_count is number of model parameters, bytes_per_parameter is number of bytes used to store one parameter. For 16 bit = 2, 8 bit = 1, 4 bit = 0.5.

Announce Date: 02.11.2024
Parameters: 5.6B
VRAM requirements: 5.4 GB using 4 bits quantization, 10.8 GB using 8 bits quantization, 21.6 GB using 16 bits quantization
Developer: Z.ai
License: The CogVideoX License

Use our pre-built public endpoints for free to test inference and explore CogVideoX1.5-5B capabilities. You can obtain an API access token on the token management page after registration and verification.

Rent your own physically dedicated instance with hourly or long-term monthly billing.

We recommend deploying private instances in the following scenarios:

• maximize endpoint performance,
• enable full context for long sequences,
• ensure top-tier security for data processing in an isolated, dedicated environment,
• use custom weights, such as fine-tuned models or LoRA adapters.

Contact our dedicated neural networks support team at nn@immers.cloud or send your request to the sales department at sale@immers.cloud.