Feb 27, 2024 21:00:00

A demo of 'SDXL Lightning' where Stable Diffusion XL follows prompt input and generates images at high speed is now available

A demo app for the high-speed image generation model '

SDXL-Lightning ' developed by ByteDance was created by AI engineer Alex Volkov. The demo app uses the image generation AI high-speed processing API hosted by `` fal.ai '' and inputs simple sentences by combining the high-speed execution API of the large-scale language model ``Mixtral'' provided by `` Groq ''. By simply doing this, you can quickly generate prompts optimized for image generation AI and instantly generate high-quality images.

ThursdAI Feb 22nd - Groq near instant LLM calls, SDXL Lightning near instant SDXL, Google gives us GEMMA open weights and refuses to draw white people, Stability announces SD3 & more AI news
https://sub.thursdai.news/p/thursdai-feb-22nd-groq-near-instant

This week was incredible for AI speed breakthroughs!

Preparing for @thursdai_pod later today, I combined super fast LLM (<300ms @GroqInc ) + super fast SDXL (thanks to friends at @fal_ai_data ) with infinite craft by @nealagarwal to get near instant visualizations of the… pic.twitter.com/ oQ7Lc8LlOR

— Alex Volkov (Thursd/AI) (@altryne) February 22, 2024

You can try SDXL Lightning from the demo page below.

SDXL Lightning - by fal.ai
https://fastsdxl.ai/

You can see how fast it is generated by watching the movie below.

I tried using 'SDXL Lightning' where Stable Diffusion XL automatically generates images at lightning speed - YouTube


The page is simple, if you enter a prompt in 'Prompt' and a seed value in 'Seed', the generated image will be automatically displayed below the input field.

If you rewrite the seed value or randomly switch the seed value by clicking the icon on the right of the seed value input screen, images will be generated at lightning speed.

Inference takes 100 to 200 milliseconds. You can see that it takes about 1 second to generate an image and display it in the browser, which shows that the image can be generated quite quickly.

Of course, when I rewrote the prompt, the new image was immediately generated.

Images are generated at such high speed that images are generated each time you type at the prompt. In the movie below, you can type the prompt 'japanese anime blue hair two girls eat ramen outside snow winter night' into the input field, and you can see how the generated image changes as the prompt gets longer. .

When you input a prompt with 'SDXL Lightning', an image is generated and changes in real time - YouTube


The code for the SDXL Lightning demo is hosted in the GitHub repository below.

GitHub - fal-ai/sdxl-lightning-demo-app: A demo application using fal.realtime and the lightning fast SDXL API provided by fal
https://github.com/fal-ai/sdxl-lightning-demo-app