Here's A New AI Image Generator That Combines The Power Of Stable Diffusion and Midjourney

In the world of AI image generators, two names stand out: Midjourney and Stable Diffusion. Midjourney can generate images with exceptional quality, but it’s not possible to run on your machine. Stable Diffusion is famous for being open-source, but the quality of the base model is still distant from Midjourney.

Now here’s an AI model that combines the level of control of Stable diffusion with ease of use and quality similar to Midjourney—it's called Distillery.

What is Distillery?

Distillery is an AI art generation service that is fully open source, created by FollowFox, which is an AI venture studio focused on small AI models. According to the company, they are committed to disclosing all their backends and models to the community.

Distillery is open-source

FollowFox has released its latest model, Cosmopolitan, which is based on Stable Diffusion 1.5 (SD 1.5). This powerful model has proven to be highly capable for general use cases and can be accessed for free on Distillery in Discord.

In line with its open-source philosophy, the company makes its training methods and codes publicly accessible.

The development process involved several steps: selecting and creating a dataset, fine-tuning the model, and mixing different versions of the model to achieve the desired result. The team clearly put in a ton of hard work to make Distillery a reality.

You can check out the open-source model in CivitAI to run the model on your local machine.

If you’re interested in knowing how Distillery’s back-end works, check out this article.

Here’s How You Can Give It A Try

There are two ways to try the Cosmopolitan model:

• Through Distillery on Discord
• Run locally on your machine by downloading this image model

In this example, I will walk you through the process of generating images on Discord. Head over to the free-generations channel.

Hit the “/” key, and you’ll see a couple of options:

• /command-infos — information on Distillery commands
• /serve — create four upscaled images using paid models
• /serve-free — create four upscaled images using free models
• /suggest — create an elaborate prompt from simple text using Distillery’s LLMs
• /upscale — upscale an image using the paid model
• /upscale-free — upscale an image using the free model
• /zoomout — zoom out an image

Prompt: an image of an astronaut riding a horse on the moon

The image generation took about a minute to complete. What do you think of the results? I think they look pretty good; the level of detail is exceptional.

They are insanely detailed and are on par with Midjourney.

LoRA

LoRA, or Low Rank Adaptation, is an optional and easy way of adding a pre-trained style to your generated images.

There are dozens of curated LoRAs available that are integrated into the Distillery’s default models. For best results, it is important to use the LoRAs with their corresponding activation words in the prompt.

Here are some examples:

Prompt: /serve-free an astronaut riding a bike on Mars — lora realism — image

Prompt: /serve-free a photo of Jennifer Lawrence having sitting in a Parisian coffee shop, analog style — lora analog — cfg 5 — seed 2000 — neg deformed, low quality — ar 16:9

Currently, users have access to 90+ different styles. Users can blend up to five different LoRAs into one prompt, which can lead to positively unexpected creations. I love the ability to combine so many styles and influences into a single image.

See the complete list of supported LoRAs here.

Merge styles

A unique feature of Distillery is the ability to merge styles and use image input as the base of the generation. Below is a demonstration of how to merge an image of a warrior with a background of Ipanema Beach using the control and adapt functions of Distillery.

That is awesome.

Here’s another example of the practical use case of Distillery. The final output that’s expected to be produced is a blend of realism and magic, symbolizing open windows to infinite creative opportunities. To get the results, an image of an ‘open window’ is used as a base image and a second image for stylistic input.

It takes some time to learn all the features, models, styles, and their use cases in Distillery, but once you get familiar with them, that is when it gets the most interesting because you can use everything together and start blending the styles and images to make novel discoveries.

Pricing

Distillery is a freemium product on Discord with three-tier service:

• Free—this gives you 10 free generations per day. Each free request provides you with four high-resolution images (1024x1024 pixels).
• Early User—this costs $9.99 per month. You get 300 premium generations per month, with no daily usage caps. Each request provides you with 4 images (as opposed to the 2 provided in the free tier).
• Founders’ Pass — for $24.99, you get 1,000 premium requests per month (3.33x more than the Early User). Founders’ Pass members will receive lifetime access to all future public tiers and related offerings

Note that this is still an alpha version of the model and could still improve in the future.

We must stress that this is still our MVP. We have huge plans for Distillery in the very near future, and we greatly look forward to discussing them in upcoming posts.

Final Thoughts

Overall, Distillery stands out as a remarkable open-source image model that rivals the quality of Midjourney while offering the most control over the generation process. The open-source nature of the image model also makes it an attractive choice for those looking into integrating it into their own products.

Distillery is still in development, so we can expect improvements to happen in the next iterations. This is one to watch closely in the months ahead!