by THUDM
Lightweight 2B video generation model from Tsinghua University. Most accessible CogVideoX variant, runs on 8GB+ VRAM with quantization. Generates 6-second clips at 8fps. Apache 2.0 licensed.
Your hardware
Detecting...
Measured quality metrics for CogVideoX 2B outputs.
How often humans prefer this model's output (0-100%)
Visual quality and composition rating (5-9 scale)
VRAM estimates at FP16 and FP8 precision. FP8 uses ~40% less memory with minimal quality loss. Grade shows how well each GPU handles the generation workload.
| Scenario | VRAM | RTX 4090 24GB | RTX 3060 12GB | RTX 4060 8GB | MacBook Pro M4 Pro 24GB |
|---|---|---|---|---|---|
| 512×512 · 25 frames | 19.0 GB | S | F | F | B |
| 768×512 · 25 frames | 21.1 GB | A | F | F | D |
| 768×512 · 100 frames | 27.4 GB | B | F | F | F |
| 1280×720 · 25 frames | 29.6 GB | D | F | F | F |
| Scenario | VRAM | RTX 4090 24GB | RTX 3060 12GB | RTX 4060 8GB | MacBook Pro M4 Pro 24GB |
|---|---|---|---|---|---|
| 512×512 · 25 frames | 12.0 GB | S | B | F | S |
| 768×512 · 25 frames | 14.1 GB | S | D | F | A |
| 768×512 · 100 frames | 20.4 GB | A | F | F | D |
| 1280×720 · 25 frames | 22.5 GB | B | F | F | D |
Turbo / LCM distillation
Use distilled scheduler at 4-8 steps for faster iteration
from diffusers import CogVideoXPipeline
import torch
pipe = CogVideoXPipeline.from_pretrained(
"THUDM/CogVideoX-2b",
torch_dtype=torch.float16
)
pipe.to("cuda")
frames = pipe(
prompt="your prompt here",
num_inference_steps=50,
guidance_scale=6.0,
num_frames=49,
).frames[0]
# Save frames or export as videoGet started
Setup instructions for running CogVideoX 2B locally
1. Download the model
Get the checkpoint from HuggingFace
2. Place in:
ComfyUI/models/checkpoints/3. Launch ComfyUI
python main.pyVRAM allocation for 25 frames at 768×512 on RTX 4090 24GB
25 frames at 768×512, 30 steps, FP16.
Download CogVideoX 2B in different precisions. Lower precision = less VRAM but slight quality loss.
| Format | Precision | Size | Provider | |
|---|---|---|---|---|
| safetensors | FP16 | 4.3 GB | official | Download |
Few LoRAs available for CogVideoX-2B.
Frequently asked questions
CogVideoX 2B (2B parameters) requires approximately 21.1 GB of VRAM at FP16 precision for generating 25 frames at 768×512. Video generation typically requires more VRAM than image generation due to temporal attention layers.
Yes, the RTX 4090 (24 GB VRAM) can run CogVideoX 2B at FP16. Expected generation time is around ~1m 28s for a 25-frame clip.
On a reference GPU (RTX 4090 24GB), CogVideoX 2B generates a 25-frame video at 768×512 in approximately ~1m 28s at FP16 with 30 inference steps. Faster GPUs with higher memory bandwidth will reduce generation time.
CogVideoX 2B supports up to 720×480 resolution and 49 frames per generation at 8 FPS. Higher resolutions and frame counts require proportionally more VRAM.
See also