CogVideoX1.5-5B (and 5B / 2B / I2V variants)
CogVideoX1.5-5B generates 1360x768 video at 16 fps in frame counts of 16N+1 up to 161 frames (about 10 seconds), with a matching I2V checkpoint.
- Up to
- 10 seconds
- Resolution
- 1360x768
- Weights
- Downloadable
- API
- Yes
- Commercial use
- Depends which variant you download
Earlier CogVideoX-5B and CogVideoX-2B produce 720x480 at 8 fps up to 49 frames. The family is notable for very low VRAM floors via diffusers quantization.
CogVideoX-2B is Apache 2.0; CogVideoX-5B is under a custom CogVideoX LICENSE that permits commercial use only after registration, caps service usage at 1 million visits a month, and is governed by PRC law with disputes in Beijing courts.
- You must register for a commercial licence. The CogVideoX LICENSE requires you to register and obtain a basic commercial licence.
- Permission required above a user threshold. Commercial service usage is capped at 1 million visits per month, beyond which you must contact the business team.
Specification
- What it takes in
- text, image
- What it makes
- video
- Maximum duration
- 10 seconds
- Maximum resolution
- 1360x768
- Aspect ratios
- not established
- Controls
- text prompt, reference image (I2V checkpoints), frame count in 16N+1 steps, guidance scale, seed, BF16/FP16/INT8/FP8 precision, CPU offload and VAE tiling, LoRA and SFT fine-tuning
- Weights
- Downloadable
- API
- Yes
- Pricing
- Free open weights (license differs by variant, see licenseNotes). Self-hosting is cheap: CogVideoX1.5-5B is documented at 'from 10GB' with diffusers optimizations, CogVideoX-5B down to about 5GB minimum, and CogVideoX-2B I2V from about 4GB, at the cost of speed. Zhipu also sells hosted video generation through its open.bigmodel.cn platform.
- Released or updated
- 2024-11-08
Supported aspect ratios is not established here. That means no page we could retrieve stated it, not that the model lacks one.
Strengths
- Lowest documented VRAM floors in the category (from ~4-10GB depending on variant)
- Documented frame counts and fps, so duration is computable rather than guessed
- CogVideoX-2B is genuinely Apache-2.0
- Mature diffusers and fine-tuning support
Weaknesses
- 5B and I2V variants require registration on a Chinese platform for commercial use and are capped at 1M monthly visits
- PRC governing law and national-security clause are non-starters for some buyers
- 8 fps output on the original 5B/2B models looks like a slideshow
- Visual quality is now well behind 2025-2026 releases
- Apache-2.0 applies only to the weakest model in the family
Best for
- Very low-VRAM local generation and research, with CogVideoX-2B as the option when a truly permissive license is mandatory.
Where the sources disagree
Recorded rather than resolved. Picking the tidier number would hide the fact that the vendor's own pages do not agree.
- Many roundups describe the whole CogVideoX family as Apache-2.0. Only CogVideoX-2B is; the 5B and CogVideoX1.5 models are under the custom CogVideoX LICENSE with a registration requirement and a 1M monthly visit cap.
What to understand before using it
Explainers naming the mechanisms this model's own documentation mentions.
See it from the vendor
Shotcraft hosts no sample output. Judging a model means watching its own showreel with your own eyes, and a clip re-hosted here would be a clip we chose.
Nearby in the directory
Adobe Firefly Video Model
The Adobe Firefly Video Model generates video from a text prompt, optionally guided by an image used as the first or final frame. Adobe's Firefly Serv
Video generation
Allegro
A small text-to-video model comprising a 2.8B VideoDiT and a 175M VideoVAE, released under Apache-2.0 in October 2024. It generates 88-frame clips at
Video generation
Amazon Nova Reel 1.1
Amazon Nova Reel generates video from text prompts or from text plus an image, and is available only through Amazon Bedrock. AWS documents output at 1
Video generation