Create AI art locally: Apps for stable diffusion and other deep learning models

It seems that AI art is currently on everyone's lips - and eyes. In addition to completely confused compositions, there are sometimes very respectable images from the individual programs that can create images locally with "Stable Diffusion" and other deep learning models. With regard to the Apple Mac, these apps are primarily made for models with "Apple Silicon" chips; you won't find Intel versions. And the operating system should also be up to date, since the programs are mostly Monterey oder sogar Ventura assume. The following is about Mochi Diffusion, Diffusers, AI Photo and Invoke AI.

In the following I will introduce you to four programs for creating AI art locally on the Apple Mac. I tried three of them myself. The fourth should also be good, but I failed with the installation.
In the following I will introduce you to four programs for creating AI art locally on the Apple Mac. I tried three of them myself. The fourth should also be good, but I failed with the installation.

Preview: The setup I use to try out apps

Apps for creating AI art on the Apple Mac are primarily designed for M chips, i.e. the M1 Pro and M1 Max, truly M2 chip as well as on the new ones M2 Pro and M2 Max. In addition, you should always have the latest operating system installed to ensure compatibility with the programs. Personally, I'm currently working with this setup, which is well suited for testing the AI ​​apps mentioned:

  • Device: 16 inch MacBook Pro from 2021
  • System on Chip: Apple M1 Pro
  • CPU cores: 10 / GPU cores: 16
  • RAM: 16GB shared memory
  • Operating system: macOS Ventura 13.2

Mochi Diffusion – My current favorite

I've made my first steps in the field of AI art in the last few days. I looked at various apps for macOS. The one I think is the best so far is called Mochi Diffusion. It's not the easiest when it comes to installing and integrating deep learning models - but you can get it done. You can then enter image descriptions, specify things that should not occur, set the number of work steps and AI intensity and more. Also, not only one image can be output, but up to 100 attempts per prompt to satisfactorily implement the entered instructions.

Mochi Diffusion is currently my favorite due to the customization options and ease of implementation of various deep learning models. However, I've only just begun to look into the subject.
Mochi Diffusion is currently my favorite due to the customization options and ease of implementation of various deep learning models. However, I've only just begun to look into the subject.

In addition to stable diffusion, deep learning models such as Fruity Mix, Anything, Redshift Diffusion and the like can also be used. Some are better for landscapes, some better for people, still others for animals or objects, and others for imitating art styles. Here you either have to read in or try everything out and make a note of which models and which settings are best for your own purposes. From an image of a D&D character for the character sheet to a fictional artwork in the style of a specific person, anything is possible. Sci-Fi or a crocodile with a top hat can also be created.

Mochi Diffusion Download

You can find the free download of Mochi Diffusion with this link on GitHub. The source code of the open source software is also available there. In order to feed the software with models, from which the necessary resources for image creation are then drawn, it is best to visit this website. There are models designed for Apple “Core ML” technology, currently 38 pieces. There is a small description and example pictures for each. For coolly designed comic images that should look like they were drawn with ink, e.g. B. Made Inkpunk Diffusion. The models are each several gigabytes in size. After downloading and unzipping, pull them in Macintosh HD/Users/[name]/Documents/MochiDiffusion/models.

As of now, there are 38 models that you can download for different AI styles and implement them in Mochi Diffusion. However, each model takes up several gigabytes of storage space on your hard drive.
As of now, there are 38 models that you can download for different AI styles and implement them in Mochi Diffusion. However, each model takes up several gigabytes of storage space on your hard drive.

Diffusers - Simpler, but also quite good

Although I'm mostly into Mochi Diffusion, I have other apps installed for creating AI art as well. Diffusers is one of them. It comes with various "Stable Diffusion" models and can also be fed with prompts (image descriptions) and negative prompts (content that should not appear in the image). Up to 150 work steps can be defined, but no AI intensity. All in all, it's a similar app, but with fewer customization options. However, it is useful for first steps, especially since it can also be downloaded and used free of charge.

Diffusers is also a good app for creating AI art on macOS. Several "Stable Diffusion" versions are already included. Definitely suitable for the first steps and beyond.
Diffusers is also a good app for creating AI art on macOS. Several "Stable Diffusion" versions are already included. Definitely suitable for the first steps and beyond.

Diffuser's download

Diffusers doesn't need a trip to GitHub and other websites — and neither do they Port-Inputs. Because this app is offered in the Mac App Store. With this link you come directly to the offer. There you will also find the first example images, including the "Labrador in the style of Vermeer" example, which is displayed when you first open the app. It's a bit reminiscent of one of Vermeer's most famous paintings, Girl with a Pearl Earring. 

AI Photo - Image sorting as a great strength

While I've had good experiences with Mochi Diffusion when creating landscapes, fantasy characters, animals, vehicles and sci-fi scenes, the AI ​​Photo app convinced me with a completely different thing: sorting the created images. In addition to an "All images" overview, there are various "Collections" that you can expand with your own folders as you wish. From cats to architecture to textures, everything can be organized and quickly found here. There is also a search function and you can mark favorites. So if you work with large amounts of AI images, you should give AI Photo a try.

The strength of AI Photo, in addition to creating images in a wide variety of styles, lies in the sorting of the various images. These are saved in the app and are not lost when it is closed (which is the case with Mochi Diffusion, for example, if you do not save the images manually).
The strength of AI Photo, in addition to creating images in a wide variety of styles, lies in the sorting of the various images. These are saved in the app and are not lost when it is closed (which is the case with Mochi Diffusion, for example, if you do not save the images manually).

AI Photo Download

AI Photo can also be easily downloaded from the official Mac App Store – Right this way. So you don't have to search for the right version on GitHub or use the terminal for any scripts. And useful pictures come out with this program as well. Of course, you also have to play around a bit here and refine the specifications from time to time in order to get exactly the picture you want. But what you have to keep in general and what you have to describe in detail, you learn over time. A bit of patience is required.

Invoke AI – The most complicated install of them all

I value the developer community and love GitHub, the open source software provided elsewhere, and the free app offerings beyond that. But you can also make the installation unnecessarily complicated. Sometimes the (ease of) installation and use of the developed application does not seem to be as important as the feasibility of the project. At least that's how I feel about Invoke AI. I tried the GitHub download (to be found here) with the instructions from this page to be implemented via the terminal. But it didn't work because of some Python errors or something. 

??

I have to admit that I didn't linger long here with the error solution. After a second attempt didn't work, I postponed the test of Invoke AI indefinitely. However, if you were able to successfully install the app and noticed special advantages (especially compared to the other apps presented here), then please leave a comment. I'm really excited to see what the program can do and whether it was able to convince you. Personally, at the moment I just feel reinforced in my wish that developers think about the normal user and make the installation of software as easy as possible.

Did you like the article and did the instructions on the blog help you? Then I would be happy if you the blog via a Steady Membership would support.

8 thoughts on “Creating AI Art Locally: Apps for Stable Diffusion and Other Deep Learning Models”

  1. Beatrice Willius

    I downloaded an app like this a few months ago. But I don't remember which ones. Was way too slow for my Air. The result resembled unattractive lumps. I've only tried very simple things like "siamese cat".

    Programming is just more difficult if other people are supposed to use the program as well. These apps all look very immature.

    If the python3 command is missing then the whole python of https://www.python.org/downloads/ . And after that, the paths will probably need to be adjusted. I don't like that at all.

    I also checked out a discord server. But somehow I couldn't do anything with it.

    1. Hello Beatrix,

      Yes, most of the time you have to make the entries a little more detailed and try out certain terms and limitations. The image examples for this article were perhaps chosen unappealingly. However, the apps can output quite respectable things. I'll add a few more pictures to make it clearer. Or make a new post about prompts.

      I would call the apps “simple” rather than “immature”. If you want to take the first steps in AI art, then they are quite useful - at least that's my observation so far.

      Yes, I didn't want to experiment with Python and such. In the terminal, which had to be used for the installation, it was also shown that I would need Xcode. But why should I assemble the app myself (or have it assembled using a script) when you could also offer it ready for download? It doesn't really make sense to me at all.

      Discord is another topic. Doesn't really have anything to do with AI, except maybe in terms of bots. Anyway, I like Discord.

      Best regards
      John

      1. Beatrice Willius

        Building an app is work. With Xojo, I push a button. In Python, it's much more complicated. Then the program has to be signed and notarized, which is more work.

        I signed up for some AI thing (midjourney???). Absolutely terrible onboarding process. And then I thought, I can put my prompts in there somewhere. But I didn't see through it at all and didn't feel like working through it.

        1. Oh right, Midjourney uses Discord. Now I understand the connection. I tried it out months ago, but then preferred to wait for a web application or app for local use. And luckily there are quite a few of them. I had already repressed the whole Discord trouble because I mainly use Discord to communicate with other people or with certain communities - depending on the server.

          1. Beatrice Willius

            My mochi test was even faster than the app I tried a few months ago. The thing crashed hard after a few minutes. Scrap metal!

          2. Interesting. With the setup mentioned in the article I have not had any problems or crashes of the app.

    2. So I'm a big fan of Midjourney. What you get there for pictures is just crazy. I was already using Midjourney 6 2 months ago and was so excited about it, but the current version 4 has blown my socks off. But the operation via Discord is just a thing in itself. But if I can get along with it, everyone else should be able to do it too. 😊

  2. I once tried diffusers on my Intel imac. It runs, calculates for a long time, but only ejects images from the junk box. In other words, not the slightest reference to the terms in the prompt.

Post a comment

Your e-mail address will not be published. Required fields are marked with * marked

In the Sir Apfelot Blog you will find advice, instructions and reviews on Apple products such as the iPhone, iPad, Apple Watch, AirPods, iMac, Mac Pro, Mac Mini and Mac Studio.