Playground V2.5
Playground V2.5 is a diffusion-based text-to-image generative model, designed to create highly aesthetic images based on textual prompts. As the successor to Playground V2, it represents the state-of-the-art in open-source aesthetic quality. Playground v2.5 excels at producing visually attractive images. It achieves this through advancements in color, contrast and human details.
Technical Details
- •
Model Type: Playground V2.5 operates as a Latent Diffusion Model.
- •
Text Encoders: It utilizes two fixed, pre-trained text encoders: OpenCLIP-ViT/G and CLIP-ViT/L.
- •
Architecture: The model follows the same architecture as Stable Diffusion XL.
- •
Resolution: Playground V2.5 generates images at a resolution of 1024x1024 pixels, catering to both portrait and landscape aspect ratios.
- •
Scheduler Options: The default scheduler is EDMDPMSolver Multistep Scheduler, which enhances fine details. A guidance scale of 3.0 works well with this scheduler.
Playground V2.5 outperforms SDXL, PixArt-α, DALL-E 3, Midjourney 5.2, and even its predecessor, Playground V2.