I’ve been tinkering with Tensor.art, an image generating platform. My first impression of this platform was that it was a new type of website that I was visiting. Harking back to the early 2000’s when the internet was a less centralised space. Over the last decade, the majority of time I have spent online has been on either social media platforms, search engines or games. This was very different from those. The success of chatGPT may lie in the fact that it is a new kind of use of the digital spaces we visit, one that is more active rather than passive.
Tensor has many things to offer its visitors. Here is the landing page. The many options availabe for explorations include the ability to create your own images by using the large library of tools and models available. One can also train their own LoRA (low rank adaption) a form of base trained models to provide context or style that you may be looking to create. It also allows the use multiple LLMs (large language models) such as stable diffusion, flux as well as those independently created by enthusiasts. One can choose from a wide variety of tools and models due to the open source nature of the platform
Here, I am giving an example of the image generator. I wanted to create an image of a boy riding a bicycle through a forest. I kept the prompt simple to demonstrate how using different LoRAs changed the image created through the same prompt.
I started with a base of Indian woman and combined it with different LoRAs. The images varied based on the second input. Some models worked better with each other while others can interfere with each other to create an output.
Different models can be assigned prompt weights that decide how much of each model is considered in the final outcome. In case of prompt weight being too high, the website will suggest changes for use.
There are lots of modest to choose from and it can be a good tool to create a vary array of images. Here are some examples where the models worked against the prompt to create some hilarious results.