Instructions to use Bl4ckSpaces/Project-Cosmos-V1.0 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use Bl4ckSpaces/Project-Cosmos-V1.0 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("Bl4ckSpaces/Project-Cosmos-V1.0", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
- DiffusionBee
Project-Cosmos-V1.0 (MagR Edition)
Project Cosmos represents the 2026 SOTA in model merging using Magnitude-DARE (MagR). Instead of random dropouts (Classic DARE) or simple blends (Linear), Cosmos intelligently identifies the "strongest" feature updates from NoobAI and implants them into IllumiYume with aggressive rescaling.
The Technology: MagR
- Input Blocks (Anatomy): 90% Drop Rate. We discard 90% of NoobAI's data, keeping only the top 10% strongest structural updates. This creates a "Super-Skeleton" while preserving IllumiYume's stability.
- Output Blocks (Style): 60% Drop Rate. We allow the top 40% of NoobAI's style features to penetrate, rescaled to ensure the "NoobAI Look" dominates the rendering engine without creating noise.
- Rescaling: Remaining vectors are mathematically amplified to compensate for the dropout, resulting in extreme clarity.
Recommended Settings
- Sampler: Euler a / DPM++ 2M SDE
- CFG: 5.0 - 8.0
- Steps: 30+
- V-Prediction: ENABLED
- Downloads last month
- 16