import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("cyberdelia/CyberRealisticXL", dtype=torch.bfloat16, device_map="cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]CyberRealistic XL
CyberRealistic XL is a photorealistic image generation model based on Stable Diffusion XL 1.0 (SDXL). It is designed to produce highly realistic, cinematic-quality outputs with minimal prompt engineering. This checkpoint is optimized for portraits, fashion, and editorial-style scenes, but also handles a wide range of creative prompts with remarkable fidelity.
🧠 Model Details
- Model Type: Image generation (Text-to-Image)
- Base Model: Stable Diffusion XL 1.0
- Format:
safetensors - Latest Version: v5.7
- Creator: Cyberdelia
- File Size: ~6.31 GB
- License: CreativeML Open RAIL++-M License
✨ Features
- ⚡ Ultra-Photorealism – Exceptional detail in faces, lighting, and textures.
- 🧩 Effortless Prompting – Works well with short prompts, but supports complex ones too.
- 🪞 Skin & Eye Fidelity – Improved clarity in eyes, skin detail, and facial structure.
- 🔧 Built-in VAE – No need for an external VAE; optimized out of the box.
🛠️ Recommended Settings
| Parameter | Recommended Value |
|---|---|
| Sampling Steps | 25–30 |
| Sampler | DPM++ 3M SDE Karras, Euler A |
| Resolution | 1024x1024 (native), 1536x1536 with Hires.fix |
| Hires Upscaler | 4x_NMKD-Siax_200k or 4x_NickelbackFS_72000_G |
| CFG Scale | 7.0–8.0 |
🧾 Example Prompts
(masterpiece, best quality), ultra-detailed, realistic photo of a 22-year-old woman, natural lighting, depth of field, candid moment, color graded, RAW photo, soft cinematic bokeh
(masterpiece, photorealistic), editorial fashion photo, close-up, dramatic side lighting, textured skin, shallow depth of field, soft shadows
📸 Example Outputs
- Downloads last month
- 46,943