by NVIDIA
7B diffusion model from NVIDIA's Cosmos platform for physical AI and world modeling. Generates physically plausible videos from text descriptions. Part of NVIDIA's Physical AI initiative.
Your hardware
Detecting...
Measured quality metrics for Cosmos Diffusion 7B 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 | 25.9 GB | B | F | F | F |
| 768×512 · 25 frames | 26.3 GB | B | F | F | F |
| 768×512 · 100 frames | 27.2 GB | B | F | F | F |
| 1280×720 · 25 frames | 27.5 GB | B | F | F | F |
| Scenario | VRAM | RTX 4090 24GB | RTX 3060 12GB | RTX 4060 8GB | MacBook Pro M4 Pro 24GB |
|---|---|---|---|---|---|
| 512×512 · 25 frames | 13.7 GB | S | B | F | S |
| 768×512 · 25 frames | 14.0 GB | S | D | F | A |
| 768×512 · 100 frames | 14.9 GB | S | D | F | A |
| 1280×720 · 25 frames | 15.2 GB | S | D | F | A |
Turbo / LCM distillation
Use distilled scheduler at 4-8 steps for faster iteration
from diffusers import DiffusionPipeline
import torch
pipe = DiffusionPipeline.from_pretrained(
"nvidia/Cosmos-1.0-Diffusion-7B-Text2World",
torch_dtype=torch.float16
)
pipe.to("cuda")
frames = pipe(
prompt="your prompt here",
num_inference_steps=35,
guidance_scale=7.5,
num_frames=57,
).frames[0]
# Save frames or export as videoGet started
Setup instructions for running Cosmos Diffusion 7B 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 Cosmos Diffusion 7B in different precisions. Lower precision = less VRAM but slight quality loss.
| Format | Precision | Size | Provider | |
|---|---|---|---|---|
| safetensorsRecommended | BF16 | 14.0 GB | official | Download |
Frequently asked questions
Cosmos Diffusion 7B (7B parameters) requires approximately 26.3 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.
Cosmos Diffusion 7B can run on the RTX 4090 with sequential offloading, though video generation will be significantly slower than native fit.
On a reference GPU (RTX 4090 24GB), Cosmos Diffusion 7B generates a 25-frame video at 768×512 in approximately ~4m 38s at FP16 with 30 inference steps. Faster GPUs with higher memory bandwidth will reduce generation time.
Cosmos Diffusion 7B supports up to 1024×576 resolution and 57 frames per generation at 24 FPS. Higher resolutions and frame counts require proportionally more VRAM.
See also