Distillery is the brainchild of FollowFox.ai, a small team dedicated to developing a fully open-source AI solutions. It is an advanced AI text-to-image generator that offers unparalleled control over image creation process. They provide users with extensive parameter control, allowing them to adjust parameters for precise image customization. This level of control is nearly unmatched, making Distillery perfect for advanced and power users.
In this post, I will give real-life use case examples and demonstrate how different parameters contribute to the process. Getting used to Distillery takes some time, but after achieving a certain level this tool stands out on top in this already crowded space.
Before diving into an example, here is what Distillery offers as of today:
Before diving into an example, here is what Distillery offers as of today:
Demonstration - Creating promotional material for your business
Distillery is a great tool to support your creative process. It can drastically reduce your time to create promotional materials for your business. In the example below, I have created a quick promotional material for an eyewear shop called ‘Sola’ for Halloween.
The output image may need some tweaks for finer details, but the overall quality is great and is ready to be used for social media. I love how AI added its own interpretation to the image: putting eyewear to the character or adding a banner on the window of the shop. Some generations can come out as messy but even they can lead to surprisingly good ideas you can leverage for your campaign. Those mutations introduce an element of unpredictability and randomness into the creative process. It can provide artists with unexpected and novel combinations, sparking new ideas and pushing the boundaries of traditional artistic approaches.
Image 1: Grim Reaper standing outside his newly opened eyewear shop 'SOLA'
In order to generate the above image we have used few different parameters. Below is the list of of them and their respective explanations:
- Model ‘Mohawk’ - SDXL-based model inside Distillery which is great for creating characters
- LoRA ‘Texta’ - There are 100+ LoRAs available on Distillery. Each of them enhances a specific style. You can mix up to 5 different LoRAs. For this image, we used ‘’texta’’ which improves text creation on images
- Adapt - After the first couple of generations, we chose the best one and used is with adapt function to re-generate new ones. That is how we got to the image above. Here is the explanation of this function from the documentation of Distillery:‘’…allows the model to 'perceive' the input image as a prompt. As the authors of the paper mention, "an image is worth a thousand words," and indeed, it is a powerful tool that comes with a bit of a learning curve but gives a whole lot of new capabilities to the model.’’
- Control - We used the control function for the same purpose as the adapt function. The feature allows users to provide a reference image as an input that will guide the image generation process.
Using Inpainting function to enhance the image
After the initial generation, we started enhancing the image further. There are bunch of ways how that can be done on Distillery, but for now, we mainly focus on inpainting.
Inpaint is a powerful tool that lets you do a new generation only on a specific part of the image. When compared to other tools, Distillery’s inpainting stands out with its quality, ability to follow the outline, and make the new generation look natural.
We combined adapt and inpainting functions to enhance the image with a fire next to the brand name.
Image below explains how the whole process works. First, you create a mask which is an empty white box. You create a new prompt that inpaints only in that box. To enhance our ‘fire’ we provided the given fire image as an input element.
Distillery parameters used: 1. Makemask - Used to cover a part of the image
We can use the same technique to change the face of our character.
- Inpainting - parameter that generates new prompt only in covered part
- Adapt - Same as we explained above
- Negative input - by adding negatives to the prompt ensures certain elements or characteristics are excluded from your generated image.
We can use the same technique to change the face of our character.
Combining uncombinable
When trying to push the limits of Distillery, I tried to combine a painting with our initial image generation with the hope of getting a background that resembles the painting.
Surprisingly, it did super well. Also, the results is from the first tries. I have not tweaked any additional parameters to try and see what will come up.
Discord link - https://discord.com/invite/FSecryv3S8
