Accomplice makes AI design fun and accessible. Here’s how beginners can get started.
Accomplice is a design suite that makes the benefits of artificial intelligence accessible to users of all types.
AI-powered design is a quickly emerging field, and whether you’re a content creator, marketer, or designer, it’s probably either already become a part of your daily workflow, or you are currently looking into adding it to your toolbox.
Why is AI design worth a second look from content creators and designers? Here are some benefits:
But, first, before you get started you might wonder: how exactly does AI-powered design work?
AI-powered design is anything that uses artificial intelligence to create unique images or designs.
Today, most AI-powered design tools combine two different kinds of AI programs or algorithms: 1) an AI that knows how to make images and 2) an AI that knows what things look like.
Currently, the most powerful image creation algorithm is called "diffusion". Put very simply, diffusion is a mathematical process for removing noise from an image.
Now, with diffusion we can create an image using AI, but to effectively create an image you have to know what the image should look like, right?
That's where CLIP comes in. CLIP is an open source AI algorithm from OpenAI that knows what things should look like.
So, now we have 1) our AI that creates images – diffusion – and 2) our AI that knows what images should look like – CLIP. When we combine them together CLIP will use its image identification skills to iteratively guide the diffusion denoising process toward an image that closely matches a text prompt.
So, if you were to ask the AI for “red apple on a green background”, it will combine its knowledge of the world with its knowledge of creating images to iteratively make your image.
After each iterative step, the diffusion model will make some progress and then ask CLIP if the image is good enough yet. After a couple hundred back and forths of this – all done in just a few minutes – you will end up with a 100% unique image of a “red apple on a green background”.
Above we covered how AI-powered design works in general. But how does Accomplice specifically work and what makes it special?
Remember when we talked about how a large, general diffusion algorithm works? We called them “diffusion algorithms” then, but they're actually more commonly called “diffusion models”.
And a general diffusion model is trained, or taught, on millions and millions of images and through that training learns how to make images of all kinds of different styles: photorealistic, watercolor, acrylic, pixelated, etc.
Popular applications like DALL-E 2 and Midjourney have just one large, general model that you can use to create images.
Accomplice is the only AI-powered design tool that supports dozens of different models.
Some of Accomplice’s models, like Disco Diffusion, are general and can create a little bit of everything. Some, like Portrait Diffusion, Pixel Art Diffusion 4k or Watercolor Diffusion are specialized. And some, like Butterflies AI or Landscapes AI, are super specialized and do an amazing job at just one thing!
Accomplice’s support of multiple models gives you a lot of choice when you're creating your images.
And, Accomplice is the only AI-powered design tool that lets you train your own models to create exactly the kinds of images you're looking for (we'll touch on that a bit in the advanced section below and go much more in depth about training your own custom models in future articles).
So, to quickly recap, a model is an AI program that has been taught how to make images. Sometimes in a lot of different styles. Sometimes in only one.
Accomplice supports the largest selection of image models in the world. And when creating an image on Accomplice, the first step is always to select the model.
When you want to create an image, hit the blue “Create” button. You’ll be prompted to choose a model. You can choose from “Your active models”, which are the models you have already activated and used in the past (sorted by most recently used):
Or you can choose “Browse models” to select a new type of model. Models on the browse screen are categorized by type, tagged, and searchable. Some of the models are free. Some are premium models – created either by the Accomplice team or by Accomplice users – that cost 10 credits.
A text prompt is a sentence that tells Accomplice what you want it to make. The more detailed you can make the prompt the better.
For example, in the above image, the text prompt was “hiker at the top of a mountain #pixelart” using the Pixel Art Diffusion 4K model. Not bad. Looks like he's a cute little videogame character with a big backpack on his back.
But what if we weren't necessarily looking for a cute videogame character and wanted something a bit more realistic?
If you want to give Accomplice an ever better idea of what you want, you can use a text prompt paired with an image prompt (sometimes called an “init image”).
The following is the same text prompt “hiker at the top of a mountain #pixelart” but this time it was paired with an image prompt quickly grabbed from Accomplice’s Unsplash integration.
This image is exactly what I was looking for! And Accomplice has done a great job of taking the initial image, pixelating it, and then embellishing it with extra mountains – and even what looks like a group of people on top of the mountain across the way!
If you want to get just the right image, image prompts are a great shortcut, and Accomplice’s Unsplash integration makes it really easy to find just what you're looking for.
You can greatly influence what Accomplice makes simply by adding what style you want at the end of the text prompt – like "watercolor", "4k", or "photorealistic" – or what artist you want Accomplice to emulate – like "by James Gurney" or "by Megan Duncanson”.
Once you have an image you are happy with, you can Upscale it. Upscaling your image greatly increases the image’s quality and size to 4x its original size.
Once you upscale an image you can choose to Share it and Download it. Sharing it will give your created image a shareable URL and add it to the Accomplice Community for other creators to see. Downloading it will let you download the high quality PNG to your computer or device.
If you want to run your prompt again, you can hit the "Rerun" button. This will copy your prompt and take you back to the new prompt screen, where you can tweak the text prompt or image prompt and any other settings you want to change before running the prompt again.
Accomplice comes with almost 50 models, but Pro and above subscribers can create and train their own models.
Models like Landscapes AI and Anime Skies were created by Accomplice users. They simply uploaded 20-100 images to teach Accomplice what kind of images they were looking for.
From a content creation perspective, you could create a model that is personalized to your own brand, for example. You could upload graphic assets and describe them so that Accomplice attributes a certain style to your desired images.
Learn more about no-code custom model training here.
There you have it!
Accomplice exists to help every team and individual find an AI-powered image workflow that works for them.
And the great thing is it’s easy to get started. Get 10 free credits when you sign up and 10 free credits every month following.