SDXL 1.0 Prompt Interpretation Consistency
I created 12 SDXL 1.0 AI images using the following prompt.
Test AI text to image prompt: “Watercolor, old dog sitting looking at a war memorial in the rain, his back facing us, over the shoulder shot, rainy gloomy atmosphere. Red poppies. Soft watercolor, complex contrast, pastel colours, masterpiece.”
The reason for this AI text to image creation test was to determine how faithfully Stable Diffusion XL 1.0 interprets a prompt.
You can see from the combined screenshot above the SDXL 1.0 results are very good. All 12 images depicted what was asked for “old dog sitting looking at a war memorial in the rain, his back facing us, over the shoulder shot, rainy gloomy atmosphere. Red poppies” whilst respecting most of the style requested “Watercolor, Soft watercolor, complex contrast, pastel colours, masterpiece”. The style wasn’t a soft watercolor, it was closer to a vibrant watercolor.
Although the images overall are very good, there’s a few with minor issues. Where a war memorial had a soldier/person as part of the memorial, the soldier/person tended to be turned away. On a couple of images this resulted in a statue looking at a wall! On a few images the placement of the dog (the ‘camera’ angle) could’ve been better. One image has a dog with cross protruding from its head, another has the dog behind a memorial. Despite these minor issues the results are very good.
The results were then compared to a similar DALL-E 3 test.
Continue Reading Best AI Prompt Interpretations, DALL-E 3 or Stable Diffusion XL 1.0?
