Week 17 of 2024: A Review of AI Developments


CogniWerk Editor

Date: 26.04.2024

As we find ourselves a quarter into 2024, it's astounding to reflect on the rapid advancements in artificial intelligence this year. Making it seem as if two full years of progress have been condensed into one. Notably, GPT-4 Turbo was launched, marking a significant stride in large language model performance. Other LLMs also showed substantial gains. Amidst these developments, Emad Mostague stepped down as CEO of Stability, and Adobe integrated enhancements pioneered by the open-source community into its software suite. Also parts of Adobe’s Firefly seemed to be trained on Midjorney images (Link).
Think of it what you want. 
But our focus today now shifts to the innovations on, which have been also numerous and notable this year.

Stable Cascade and SD3
In February, Stability released the Stable Cascade model, and shortly thereafter, we implemented it on our platform for using. Following this, we obtained API access to SD3, planning to make it available to our users in the current iteration.

Responsiveness Enhancements
We've also enhanced our mobile responsive presentation, enabling first generations via mobile devices—a development spurred by an increase in visitors from social media platforms. We continue to refine the user experience, and we welcome any feedback or suggestions you may have.

Given that our interface differs from many user-friendly AI programs, we are focusing on making our configuration options more accessible through simple tutorial animations in our interface. These will be expanded over the coming weeks to help users more effectively utilize our tools.



Provided Models
We have significantly altered the structure and quantity of our models to provide you with a better selection of styles and capabilities, which also load more quickly. 

  Tutorial Cw Models - Bring Your Own Model
With the integration of features, we now offer a "bring your own model" option. Civitai is the largest platform for fine-tuned models and a key element within the open-source community. Users can now load LoRA models directly through our interface into their generations, significantly enhancing the quality of their outputs. Simply copy the model link from into our text field and use it directly in our interface. Please note that loading the model may take some time. However, in the future, you will find your loaded models in the "Imported Models" section.

  Tutorial Import Models  

Image to Image
Another tool that significantly impacts the control within your generations is the Image to Image feature. Here, you can upload an existing image as a reference and adjust it from 0 to 100%.

We remain dedicated to providing the best possible experience with the open-source world and presenting technically sophisticated solutions in an accessible format. Let us know how we can improve or what you already like about our services. Feel free to link to us on social media and continue to succeed with our services at