What is Playground-v2

Playground v2 is a diffusion-based text-to-image generative model. 

“PlaygroundAI v2 1024px Aesthetic” is an advanced text-to-image generation model developed by the Playground research team. This model can generate aesthetically pleasing images of high resolution (1024×1024 pixels) based on textual descriptions. It employs two pre-trained text encoders, OpenCLIP-ViT/G and CLIP-ViT/L, providing textual context for image generation. In user research, PlaygroundAI v2 demonstrated superior aesthetic quality in its images, being 2.5 times more favored than Stable Diffusion XL. Additionally, it introduces a new benchmark, MJHQ-30K, for the automatic assessment of the aesthetic quality of generated images.

Advantage of Playground v2

Playground v2 1024px Aesthetic has the following advantages over Stable Diffusion XL (SDXL):

  1. Higher Image Aesthetic Quality: According to user research, images generated by this model are more aesthetically pleasing and 2.5 times more popular than those produced by SDXL.
  2. Same Architecture, Different Implementation: Although it shares the same architecture as SDXL, PlaygroundAI v2 achieves different results through specific optimizations and adjustments.

How to Use Playground v2 model with ComfyUI

Just download the playgroundv2.safetensors model and and put it in the checkpoint directory “\ComfyUI\models\checkpoints”

Use Playground v2 in ComfyUI example

You can use the basic Vincennes workflow to select the model as Playground v2

Here is the comfyUI workflow for loading the SDXL base model and UNETLoader

Leave a Reply