MochiDiffusion

Who remembers the release of Mac OS X Snow Leopard? 🙋

At WWDC 2009, Snow Leopard was advertised with having under the hood improvements rather than adding more features. It was famously marketed as having "zero new features"

Just like Snow Leopard, this new release of Mochi Diffusion includes many under the hood improvements that improves performance and stability. It should also help simplify adding new features down the line

The successor of Leopard was called Snow Leopard to denote it as a refinement. Therefore I found it appropriate to call this release...

🥁🥁🥁

Mochi Ice Cream

Also both snow and ice cream are somewhat similar, who knew? ❄️🍦

It also proudly has "zero new features." Well, there is one new feature...

A nice animation has been added when removing images from the Gallery 😎

Also "zero new features" doesn't mean much didn't change. Actually there were a lot of changes...

Those numbers represent the lines of code changed...whew

As usual, special thanks to the following for supporting me and making this app possible 🎉
raisingfightingspirit, serovar, orange-wedge, RuralRob, vacekj, julien-c, Quick-Eyed-Sky, BirdSesame, Da-mi-en, monks1975, angeenes, & anonymous donors

• Fixed Quick Look not displaying the correct image if the first image was selected and removed (@CarterLombardi)
• Fixed Quick Look displaying the last removed image when all images were removed from the Gallery (@CarterLombardi)
• Fixed import process for images with empty Exclude from Image description (@vzsg)
• Fixed short descriptions in Info panel causing text to be centered rather than left aligned (@vzsg)
• Added token counter to Include in Image & Exclude from Image text inputs (@CarterLombardi)
• Increased text input size of Include in Image & Exclude from Image to show more lines of text
• Organized Settings window using tabs

Special thanks to the following for supporting me and making this app possible 🎉
raisingfightingspirit, serovar, orange-wedge, RuralRob, vacekj, julien-c, Quick-Eyed-Sky, BirdSesame, Da-mi-en, monks1975, angeenes, & anonymous donors

• Added ability to import images to Gallery (File > Import Image...)
  • Image must be generated by Mochi Diffusion 2.2 or later
• Fixed potential crash if image is removed while being converted to high resolution (@vzsg)
• Improved model selector by sorting by name (@vzsg)
• Improved translation of singular & plural words

Let's D. I. S. C. O. rd

Special thanks to the following for supporting me and making this app possible 🎉
raisingfightingspirit, serovar, orange-wedge, RuralRob, vacekj, julien-c, Quick-Eyed-Sky, BirdSesame, Da-mi-en, monks1975, angeenes, & anonymous donors

• Added Russian translation (Regulus)
• Added message if prompt description is too long
• Changed max step setting back to 50
• Postponed changing minimum macOS version to Ventura 13.2

Link to Discord, this is - Yoda

Special thanks to the following for supporting me and making this app possible 🎉
raisingfightingspirit, serovar, orange-wedge, RuralRob, julien-c, Quick-Eyed-Sky, BirdSesame, Da-mi-en, monks1975, angeenes, & anonymous donors

• Added Spanish translation (k-latte)
• Changed max step setting to 40
• Next release will change minimum macOS version to Ventura 13.2

There's a Discord server?

Special thanks to the following for supporting me and making this app possible 🎉
raisingfightingspirit, serovar, orange-wedge, RuralRob, julien-c, Quick-Eyed-Sky, BirdSesame, Da-mi-en, monks1975, angeenes, & anonymous donors

• Fixed prompt file name when saving all images

Tip: File > Save All to save all generated images.

Special thanks to the following for supporting me and making this app possible 🎉
raisingfightingspirit, serovar, orange-wedge, RuralRob, julien-c, Quick-Eyed-Sky, BirdSesame, Da-mi-en, angeenes, & anonymous donors

• Added ability to search image by seed
• Changed step & guidance scale control style
• Changed saved image metadata to be similar to webui images
• Changed default step setting to 12 (sweet spot)
• Changed max step setting to 50
• Converting image to high resolution replaces existing image
• Cleanup temp images that were created by Quick Look on app close
  • Previous temp images were already being deleted on reboot
• Fixed app briefly freezing when converting image to high resolution

Do you have a model that was converted and wish to upload? Join our community on Hugging Face or create a Pull Request to get started!

Tip: File > Save All to save all generated images.

Special thanks to the following for supporting me and making this app possible 🎉
raisingfightingspirit, serovar, orange-wedge, julien-c, Quick-Eyed-Sky, BirdSesame, Da-mi-en, eidermar, & anonymous donors

We've reached 10 languages!

• Added Traditional Chinese translation (JacobLinCool)
• Added ability to select Compute Unit option when changing model
• Added option to Filter Inappropriate Images under Settings (model must have safety checker module to work)
• Fixed duplicated model options on refresh

Do you have a model that was converted and wish to upload? Join our community on Hugging Face or create a Pull Request to get started!

Special thanks to the following for supporting me and making this app possible 🎉
raisingfightingspirit, serovar, orange-wedge, julien-c, Quick-Eyed-Sky, BirdSesame, Da-mi-en, eidermar, & anonymous donors

• Changed model folder to be any path and to no longer enforce sub folder structure
• Added Swedish Translation (teodorzacke)
• Added keyboard shortcuts to the Gallery (see under Image menu)
• Added option to save all images (File > Save All...)
• Added color shadow to Inspector image
• Improved auto sizing Settings window
• Improved Quick Look code (@azyu)

You will need to set the model folder again after updating.

Do you have a model that was converted and wish to upload? Join our community on Hugging Face or create a Pull Request to get started!

Special thanks to the following for supporting me and making this app possible 🎉
@raisingfightingspirit, serovar, orange-wedge, julien-c, Quick-Eyed-Sky, BirdSesame, Da-mi-en, eidermar, Draxredd, & anonymous donors

• Changed Batches option to Number of Images for simplicity
• Updated Settings layout for different languages
• Updated Generate button to Stop Generation as well
• Updated Download Model menu link
• Added button to set random seed

Do you have a model that was converted and wish to upload? Join our community on Hugging Face or create a Pull Request to get started!

Special thanks to the following for supporting me and making this app possible 🎉
serovar, orange-wedge, julien-c, Quick-Eyed-Sky, BirdSesame, Da-mi-en, eidermar, Draxredd, & anonymous donors

• New Translations
  • Brazilian Portuguese (thiagomsoares)
  • French (Draxredd)
  • Italian (Zabriskije)
• Fixed converting non-square images to high resolution
• Clicking Apply button in Settings closes the window

Can you speak another language? Visit the project page on Crowdin!

Special thanks to the following for supporting me and making this app possible 🎉
serovar, orange-wedge, julien-c, Quick-Eyed-Sky, BirdSesame, Da-mi-en, eidermar, Draxredd, & anonymous donors

I've figured out how to convert the models to export 512x768 sized images. I will add these new models to the existing list of models I've converted here under the folder original. Note that it only supports running with CPU & GPU compute unit. See the wiki for details on the updated steps to create these new models.

• New Translations
  • Simplified Chinese (@Jerry23011)
  • Finnish (tuhoojabotti)
  • German (@eidermar)
  • Japanese (@atatakun)
  • Korean (@godly-devotion)
• Added Simplified Chinese & Korean translations for README
• Added app version info to EXIF data
• Fixed generation progress not displaying for some users
• Changed search to be case insensitive
• Changed minimum Gallery columns from 3 to 1
• Include up to 70 characters of prompt text in default image filename

Can you speak another language? Visit the project page on Crowdin!

Special thanks to the following for supporting me and making this app possible 🎉
serovar, orange-wedge, julien-c, Quick-Eyed-Sky, BirdSesame, Da-mi-en, eidermar, Draxredd, kayzen

• Added support for Intel Macs (universal binary)
  • CPU & GPU compute unit will be used
  • High performance CPU & GPU is required
• Updated Gallery UI
  • Images are displayed in a grid
  • Added ability to view images in Quick Look (double click on image)
  • Added ability to search generated images by prompt
  • Added total generated image count
• Update Inspector UI
  • Moved to sidebar to allow easier at a glance view
• Updated names of buttons & labels
• Added menu option to download pre-converted models (Help > Download Models)
• Added menu option to support this project (Help > Support Me)
• Added support for localization

Special thanks to the following for supporting me and making this app possible 🎉
serovar, orange-wedge, julien-c, Quick-Eyed-Sky, BirdSesame, Da-mi-en, eidermar, Draxredd, kayzen

Mochi Diffusion is now properly code signed and notarized thanks to your donations. Thank You ❤️

• Image generation progress was moved to the toolbar
  • View both step and batch progress
  • Cancel generation
• Updates can now be downloaded and installed directly from the app

Tip: DPM-Solver++ Scheduler works very well with only 10-25 steps

Special thanks to the following for supporting me and making this app possible 🎉
serovar, orange-wedge, julien-c, Quick-Eyed-Sky, BirdSesame, Da-mi-en, eidermar, Draxredd

I've converted a few models for Mochi Diffusion here.

Read the explanation for Compute Unit and Models to understand the difference between split_einsum and original (tl;dr download the split_einsum version to use Neural Engine).

Mochi Diffusion is now properly code signed and notarized thanks to your donations. Thank You ❤️

• Added option to convert all generated images to high resolution (will use more memory)
• Changed progress status to show batch progress instead
• Changed filename to include image index at the end (helps avoid name conflicts with images in the same batch when saving)

Tip: DPM-Solver++ Scheduler works very well with only 10-25 steps

Special thanks to the following for supporting me 🎉
serovar, orange-wedge, julien-c, Quick-Eyed-Sky, BirdSesame, Da-mi-en

I've converted a few models for Mochi Diffusion here.

Read the explanation for Compute Unit and Models to understand the difference between split_einsum and original (tl;dr download the split_einsum version to use Neural Engine).

When trying to open the app for the first time, Gatekeeper will prevent you from doing so because the app is not code signed. In order to bypass this warning, you need to right-click on the app and select "Open". You will have to do this twice in order to get the option to open the app.

• Added built-in ability to convert generated images to high resolution (using RealESRGAN)
• Added Copy to Prompt option to Gallery image right click menu
• Added button to open working directory in Finder under Settings
• Improved Info popup
  • Each section is clearly separated with headers
  • Added button to selectively copy an option to the sidebar
• Changed maximum step count to 100 (there are diminishing returns over this)
• Changed scheduler option location to Settings

Tip: DPM-Solver++ Scheduler works very well with only 10-25 steps

Special thanks to serovar, julien-c, Da-mi-en for supporting me 🎉

I've converted a few models for Mochi Diffusion here.

Read the explanation for Compute Unit and Models to understand the difference between split_einsum and original (tl;dr download the split_einsum version to use Neural Engine).

When trying to open the app for the first time, Gatekeeper will prevent you from doing so because the app is not code signed. In order to bypass this warning, you need to right-click on the app and select "Open". You will have to do this twice in order to get the option to open the app.

• Added ability to stop image generation
• Added border around the currently selected image in the Gallery
• Changed help text for Reduce Memory Usage option
• Fixed minor bugs

Tip: DPM-Solver++ Scheduler works very well with only 10-25 steps

Special thanks to @serovar @julien-c @Da-mi-en for supporting me 🎉

I've converted a few models for Mochi Diffusion here.

Read the explanation for Compute Unit and Models to understand the difference between split_einsum and original (tl;dr download the split_einsum version to use Neural Engine).

When trying to open the app for the first time, Gatekeeper will prevent you from doing so because the app is not code signed. In order to bypass this warning, you need to right-click on the app and select "Open". You will have to do this twice in order to get the option to open the app.

• View the gallery while images are being generated
• Reorganized some options in the sidebar
• Added Number of Batches option
  • Each batch increments the seed by 1 making it easier to regenerate the same image
• Added Images per Batch option (previously called Number of Images)

Note: [Number of Batches] x [Images per Batch] = [Total Number of Images Generated]

I've converted a few models for Mochi Diffusion here.

Read the explanation for Compute Unit and Models to understand the difference between split_einsum and original (tl;dr download the split_einsum version to use Neural Engine).

I am also looking for help to subsidize for an Apple Developer Program membership which will allow me to properly sign and notarize my apps (I'll be able to finally take down the Gatekeeper banner message).

When trying to open the app for the first time, Gatekeeper will prevent you from doing so because the app is not code signed. In order to bypass this warning, you need to right-click on the app and select "Open". You will have to do this twice in order to get the option to open the app.

I've converted a few models for Mochi Diffusion here.
Read the Compute Unit and Models explanation about the difference between split_einsum and original model version (tl;dr download the split_einsum version to use Neural Engine).

• Added remove image option to toolbar
• Changed toolbar item names to be consistent
• Changed compute unit description label to recommend Neural Engine option for most cases

When trying to open the app for the first time, Gatekeeper will prevent you from doing so because the app is not code signed. In order to bypass this warning, you need to right-click on the app and select "Open". You will have to do this twice in order to get the option to open the app.

I've converted a few models for Mochi Diffusion here.
Read the Compute Unit and Models explanation about the difference between split_einsum and original model version (tl;dr download the split_einsum version to use Neural Engine).

• Minor improvements to Gallery UI
• Added save & remove options to gallery image right-click menu