The future looks bright for free and open-source AI image models.
The competition for the best AI image generator has been really intense since 2023. Before the year ended, one of the most beloved platforms for generating AI images, Playground, released a brand new image model, Playground V2.
The creators claim that V2 is 250% better than Stable Diffusion XL. Is it really, though?
Playground is a free-to-use AI image generator platform that uses image models from Stable Diffusion, Dall-E (somehow they disabled the support), and Playground.
Aside from image generation, Playground also supports image editing like inpainting, outpainting, and variations from existing images.
Playground was launched in January 2023 and has since been one of the top picks for AI image generation for its easy-to-use user interface and free-to-use models.
Here are the updates on the V2 model:
Let’s see this new model in action.
Head over to the Playground AI website and create a free account. What you’ll see is a blank canvas at the center, a text prompt and settings on the left panel, and image settings on the right panel.
Select “Playground v2” on the Model drop-down menu.
Here’s an example image prompt:
Prompt: An astronaut riding a horse in space
Now, using the same prompt, let’s generate an image using the SDXL model. Here’s the result:
In case you don’t know what SDXL (Stable Diffusion XL) is yet, it’s the latest image model from Stability AI. Stability AI is one of the world’s leading open-source AI companies.
What the Playground AI team did was ask thousands of users to choose between images generated using the same prompt by SDXL and Playground V2.
Early benchmarks have shown that Playground V2 is preferred 2.5 times more than SDXL.
Here are some example images that were compared between the two models.
Of course, I won’t believe these claims without comparing the models myself. So, I did an actual comparison using the following prompts:
Prompt: A cartoon-like cartoon drawing of a monkey in profile. The lines of that drawing are thick. The monkey looks drunk. Also, the monkey is drooling.
Prompt: A zeta reticulan grey alien in a pressurized suit
Prompt: A handsome Chinese A 5-year-old girl a dancing lion on her head and a halfbody photo, Facing the audience directly, whiteand red, Symmetric facial close-up, Facing theaudience head-on, a stormtroop was a dancinglion, wearing Chinses Costumes, lridescenceclothes, white background
Looking at these results, it’s evident how much better Playground V2’s result is compared to SDXL. However, SDXL occasionally outperforms V2 in details and realism. But overall, V2 is a big leap for Playground’s free model.
If you’re interested in seeing how Playground V2 compares to other AI image generators like Midjourney, Dall-E 3, or Amazon Titan G1, let me know in the comments.
The model is also available on HuggingFace, if you wish to download it. Commercial use is permitted.
Note: Files can be up to 14 GB in size, so you might want to check your disk and hardware compatibility before running them on your local machine.
Access to all the image models is still free, with limitations. You can create up to 500 images per day with a commercial license.
If that’s not enough for you, there is a Pro tier that costs $15 per month. This subscription gives you up to 1,000 image per day and access to other image editing features such as upscales, face restorations, etc.
Turbo Tier costs $45 per month and gives you up to 2,000 image generation credits plus simultaneous image generation capability.
If you want to save money, you can opt for the annual subscription, which gives you a 20% price discount.
Overall, Playground V2 shows promising improvements over the previous version. The fact that it’s free to use makes it more accessible for testing prompts and exploring AI-generated art. As the model continues to be trained on more data, the output quality will likely keep improving.
For now, Playground V2 strikes a good balance between image quality and ease of use, especially for hobbyists and beginners. I’m excited to see how it evolves and whether it can truly match paid models like Dall-E 3 or Midjourney someday. The future looks bright for free and open-source AI image models.
Software engineer, writer, solopreneur