Open source and free: Image AI Stable Diffusion XL 1.0 released
[15:46 Fri,28.July 2023 by blip]
The open source AI image generator Stable Diffusion XL has recently been made freely available in the official 1.0 version. The new text-to-image model is said to have - as was already apparent with the previous version 0.9 - significant improvements in image generation. SDXL works in two stages with a base model based on 3.5 billion parameters and a refiner model with 6.6 billion parameters.
SDLX 1.0 generated images from Stability AI
The artificially generated 1024x1024 images are said to look more photorealistic and provide high visual quality, with more sophisticated image compositions. According to Stability AI, SDXL 1.0 can produce good results even with shorter prompts. The model is also better at inserting text into images, it said. Although errors still occur, in many cases the words have become more coherent and readable.
Likewise, it has made progress in the representation of people, among other things the hands are supposed to look less deformed - however, we cannot confirm the latter for the (limited) base model available online after a quick test (the camera labeling also leaves something to be desired).
Still too many fingers are generated online
Stable Diffusion XL offers several styles for image generation: No style, Enhance, Anime, Photographic, Digital art, Comic book, Fantasy art, Analog film, Neon punk, Isomteric, Low poly, Origami, Line art, Craft clay, Cinematic, 3D model and Pixel art.
Online interface ClipDrop
The SDXL model (base + refiner) is available for download at HuggingFace under a CreativeML Open RAIL++-M license. Those who don't want to install the image generator themselves can access the base model online at ClipDrop or DreamStudio. Stablity even provides developers with (paid) ClipDrop API access.
;)
;)
;)