AI Vlogs has recently announced a new update to their stable diffusion XL model that they have called version 0.9. This update has several impressive features that are expected to generate the ability to generate hyper-realistic creations for films and advancements in design and industrial use. Its direct paid competition of MidJourney.
Announcement by Stability AI
Stability AI has become very product-focused, which has helped them to sharpen their approach to safety and generate new models. The new version of stable diffusion XL is available today in clip drop. They are pushing it very hard, and the API is coming soon.
Improvements included in the update
There are several significant improvements included in the Stable Diffusion XL 0.9 update that are worth noting:
- Improvements in depth of field, color gamut, and how much color the model is comfortable expressing in any given generation
- Context awareness of faces and organic shapes, such as hands
- Vastly improved text rendering
- Enhancements in the image-to-image prompting, in-painting, and out-painting functionalities
- Additionally, Stable Diffusion XL 0.9 has one of the largest parameter counts of any open-source image model, with 3.5 billion parameters in the base model and 6.6 billion parameters in The Ensemble Pipeline. The final output is created by running on two models and aggregating the results.
One of the key features of SDXL 0.9 is its natural language processing capabilities, which allow users to input prompts using regular language. For instance, it can generate an image of “glowing jellyfish floating through a foggy Forest at Twilight” or a “Frozen castle made entirely of ice cream in a Land of cotton candy clouds and lollipop trees” just by using natural language prompts. This is a vast improvement over the older version (1.5), which requires single or two-word prompts separated by commas, and even model weights to generate AI art.
Stunning Composition Despite Base Model
Despite being just a base model, SDXL 0.9 can still produce stunning compositions. Even though image fidelity is not perfect yet, the quality of the images is impressive. For example, it generated an image of a yellow train with surreal elements, a Jedi female fighting Darth Vader, and a woman eating ice cream. While there are minor issues with extra limbs or imperfect hands, the resulting images are still impressive, especially considering that these were created using only natural language prompts.
Compatibility
Stable Diffusion XL 0.9 is compatible with Linux users with compatible AMD cards that have 16GB of VRAM. However, NVIDIA GPUs are still recommended, and the model will run on modern consumer GPUs. SDXL 0.9 is the most advanced development in the stable diffusion text-to-image suite of models, boasting the largest parameter count of any open-source image model to date. It uses a 3.5-billion-parameter base model and a 6.6-billion-parameter model ensemble pipeline that aggregates the results of two models running in parallel.
Availability
Clip drop has made this product available, and an open-source release is expected mid-July. The Stability AI interface has made it very easy for developers to provide low or no-cost updates to their user base. Stability AI is committed to researching how AI can benefit, rather than replace, human interactions. Concerns of censorship arise with this model, as previous models such as Stable Diffusion 2.0 were censored and were deemed of lower quality due to their anatomical distortions. It is hoped that SDXL 0.9 stays uncensored, as the community is ready to get working on it as soon as it hits public release.
Overall, the Stable Diffusion XL 0.9 update has several impressive features worth noting. Developers and users can expect to enjoy enhancements in depth of field, color gamut, and much more. Stable Diffusion XL 0.9 is set to be at the forefront of real-world applications for AI imagery, and we’re excited to see what the future holds for this powerful model. Although the image fidelity is not perfect yet, it is still an impressive achievement for a base model. As the open-source community is preparing to work on custom models, the future for SDXL art looks bright.
