Stability AI released a high-performance image generation model 'SDXL 1.0' & a ready-to-use web application was also released, so I tried using it
Stability AI, the developer of the image generation AI `` Stable Diffusion '', has released the image generation model `` Stable Diffusion XL 1.0 (SDXL 1.0) ''. SDXL 1.0 is a further enhancement of '
ANNOUNCING SDXL 1.0 — Stability AI
https://stability.ai/blog/stable-diffusion-sdxl-1-announcement
Stability AI Announces Stable Diffusion XL 1.0, Featured on Amazon Bedrock — Stability AI
https://stability.ai/press-articles/stable-diffusion-xl-1-featured-amazon-aws-bedrock
AWS-Expands-Amazon-Bedrock-With-Additional-Foundation-Models-New-Model-Provider-and-Advanced-Capability-to-Help-Customers-Build-Generative-AI-Applications
https://press.aboutamazon.com/2023/7/aws-expands-amazon-bedrock-with-additional-foundation-models-new-model-provider-and-advanced-capability-to-help-customers-build- generative-ai-applications
SDXL 1.0 can generate an image of 1024 × 1024 pixels, and in addition to improving `` light source and shadow processing '' compared to existing models, image generation AI such as `` hand '' and `` three-dimensional composition '' It is possible to generate images that are not good at.
In previous image generation models, the technique of inserting a word such as 'masterpiece' in the prompt was used to improve the quality of the generated image, but in SDXL 1.0, high quality is possible without including words such as 'masterpiece'. It can generate images. It also improves prompt comprehension, such as distinguishing between 'The Red Square' and 'red square'.
The SDXL 1.0 model data has already been released, and it has been announced that the SDXL 1.0 API will be available for Amazon's AI service 'Amazon Bedrock'. Furthermore, at the link below, it is possible to generate images using SDXL 1.0 for free, so I actually generated images.
Clipdrop - Stable Diffusion
When you access the link destination, the following screen will be displayed, so enter the prompt in the input field and click 'Generate'. This time, I entered 'A girl hacking in a cyberpunk city.'
A notification saying 'You can shorten the waiting time by subscribing to a paid plan' is displayed, but I want to use it for free, so I clicked 'Skip'.
After a few moments, four images will be generated according to your prompts. To download the image, select your favorite image from the thumbnails at the bottom of the screen and click 'Download' OK.
Four images generated by 'A girl hacking in a cyberpunk city.' Even short prompts produced high quality images.
Next, I generated four images with the prompt 'Eating sushi with bare hands. Close-up of the face.' Stability AI appeals that the depiction of 'hands' has improved, but it seems that it is still difficult to draw hands naturally.
Finally, the image generated with the prompt 'Suited panda dances in front of Mt Fuji.' One naked panda was generated, but the element 'panda dances in front of Mt. Fuji' is accurately expressed in every image.
The distribution location of SDXL 1.0 model data is summarized in the following GitHub repository.
GitHub - Stability-AI/generative-models: Generative Models by Stability AI
https://github.com/Stability-AI/generative-models
Related Posts:
in Review, Software, Web Application, Posted by log1o_hf