Create on a Mac: Local Image, Video and Music Generation Without ComfyUI

On Windows and Linux, the Create tab of Locally Uncensored runs on ComfyUI, which the app installs for you. On a Mac there is no ComfyUI. Since 3.0.6 Create makes pictures and video on the Mac itself, on Apple's MLX, and chat runs on local models with Metal. Version 3.0.7 adds character training, Talking Character, Motion Control, Edit with up to ten images, Music, Enhance Image, Remove Background, masks in Edit, a HiRes pass, LoRAs and a size for video, Extend Video and a prompt enhancer. This guide shows what runs locally on a Mac and what has really been run.

What Is New on the Mac in 3.0.7

Images

The Image tab runs the Mac image model you install under Models. On our M5 Pro with 48 GB, DreamShaper XL Turbo made an image in 33 seconds and Qwen-Image 2.1 Turbo in 38. In 3.0.7 an image with a LoRA from the Mac's LoRA folder also ran with the real engine.

With a model that edits, such as Qwen-Image 2.1, HiRes fix runs as a second pass. The finished picture goes through a second pass at the larger size, with the HiRes denoise and steps. The latent upscale method is not offered on a Mac.

Edit with up to ten images

With Qwen-Image 2.1 the reference strip under the source takes up to nine more images, so the model reads ten in all. From four extra images on, the hint asks you to say what each one brings, for example "the scarf from image 3". With more than four images, every image is scaled smaller so that together they take no more memory than four did. At a 1024 canvas, ten images run at 640 pixels each. On a Mac the result keeps the size you set.

Each picture takes memory of its own, and ten at full size left a 48 GB Mac only a third free before the first step. So the strip on a Mac offers only as many images as its memory takes at the size you set. Images already in the strip past that stay there and are marked not sent, and a run past the limit is refused with the number that fits. On our Mac, ten images at 640 pixels each ran with the real engine in 78 seconds.

Edit with a mask

With Qwen-Image 2.1, Edit on a Mac offers Paint mask, Edit strength and Mask edge feather, and a result offers Edit with mask. In our run with the engine, a hard mask at full strength changed only the masked area in 189 seconds.

Remove Background and Enhance

Remove Background uses the subject detection built into macOS, so it needs no model and no Python. A picture without a clear subject gets a sentence instead of an empty image.

Enhance Image is a local tab once SeedVR2 3B (7.28 GB) is installed from the Model Manager. It offers 2x or 4x, and the longest side is held to 4096 pixels. Without the model the tab stays locked. At 4x a 1024 picture needs about 22 GB, so a smaller Mac is told to pick 2x or a smaller picture instead of freezing. A local Enhance result in the gallery can list steps and CFG that the enhance run did not use.

The Qwen prompt enhancer

Two enhancer packs of about 6 GB each install from the Model Manager, one for new pictures and one for edits. With one installed, a new picture or an edit on Qwen-Image 2.1 can have its prompt rewritten before the run, and the edit enhancer also looks at the picture. Both ran with the real engine on our Mac.

Video

LTX-2.3 Distilled 4-bit made a clip of two seconds at 768 by 512 with sound in 64 seconds on the M5 Pro.

Where the model's entry allows it, the LTX and Wan video models take a stack of LoRAs and a width and height on the grid of the engine, and Wan takes 1 to 60 steps. Extend Video is a local tab while a video model that takes a start image is installed, and the Shots slider reaches the LTX-2.5 packs. The large LTX models now render at the size Create shows; before, they always ran at 512 x 512, so a landscape clip came out square.

What ran on our Mac: LTX-2.3 video at a chosen size, and Extend Video. LoRAs on a video run and multishot have not run with real models on a Mac yet. The large LTX models (64 GB and more) have not run on a real Mac, so their new size handling is checked in automated tests only.

Talking Character

A portrait and a voice recording become a clip whose lips follow the voice. The clip is as long as the recording, up to 10 seconds. It runs on LTX-2.3 4-bit and needs one more download of 11.3 GB, which the tab offers. It needs a Mac with 24 GB or more, and it is slow: a 4 second clip took about 10 minutes on our M5 Pro, roughly 150 seconds per second of clip.

Motion Control

A motion clip and a start image become a video that follows the motion. LU turns the motion clip into edges first, then runs LTX-2.3 4-bit with the IC-LoRA Union Control, about 650 MB, offered on the tab. A 2 second clip at 512 x 512 took 103 seconds on our M5 Pro.

Both tabs ran with the real engine on our Mac. The downloads they offer on their tabs have run in automated tests only. When you switch from Cloud to Local, Talking Character and Character Studio now stay open instead of jumping to Video or Image.

Music

The Music tab in Create is a local tab on a Mac once the music engine is installed from the Music rail of the Model Manager, about 10 GB. It runs ACE-Step 1.5. A run takes a style, lyrics, a length from 10 seconds to 4 minutes and a seed, and puts the track in the gallery. Decoding uses less memory than before: a 30 second track used to leave 1.4 GB of a 48 GB Mac free, and now leaves over 9 GB. For music on Windows and Linux, see local AI music generation.

Characters

New in 3.0.7: Character Studio's Train tab trains a Z Image character on the Mac, from 4 to 100 photos, with the Mac's image engine. A short text field under each photo takes what changes from photo to photo, such as clothes, so the character is not tied to one outfit. The first run downloads a training adapter of 340 MB. On an M5 Pro one step takes about 7 seconds, so the Standard setting with 1200 steps runs a little over two hours. Use frames for a character hands its photos to the same training.

The finished character lands in the Mac's LoRA folder. Character Studio's Use tab lists the characters in that folder. A pick brings the Z Image model the character was trained for, and Image, the video tabs and the chat tools take the character. A video starts from a picture of the character made on the Mac.

Training on the Mac has run with 10 photos and 100 steps on our Mac with 48 GB. The Standard and Thorough settings, Macs with less memory and stopping a training in the middle have not run yet. The Character Studio chapter of the handbook covers the rest.

Memory and Long Runs

All our Mac runs were on 48 GB. Nothing has run on a Mac with less memory yet, and we have no run of the Mac build on macOS 14 or 15. The tiers by memory are in What a Mac can actually run locally.

In the Model Manager, a model that does not fit the memory of your Mac says so on its tile. In Create, a line tells you when the picked model is too big or when a video run on a Mac would take long, and you can close that line. Both lines also point to LU Cloud. They have run in automated tests only.

Limits

• Beta, Apple Silicon only, macOS 14 or newer. Intel Macs are not supported
• Every real Mac run we name was on one M5 Pro with 48 GB. Nothing has run on a Mac with less memory
• The Mac parts of 3.0.7 have run in automated tests and have not been clicked through in a built app yet
• LoRAs on video and multishot have not run with real models on a Mac. The large LTX models of 64 GB and more have not run on a real Mac
• Training on a Mac has run with 10 photos and 100 steps only. The text field per photo has run in automated tests only
• The downloads that Talking Character and Motion Control offer on their tabs have run in automated tests only
• Computer control is in the app on a Mac but not fully tested yet
• No ComfyUI on a Mac, so ComfyUI workflows and custom nodes do not apply

The Mac page has the install steps.

Locally Uncensored is AGPL-3.0 licensed and built by PurpleDoubleD. Corrections welcome on GitHub Discussions or in the Discord.

In the handbook

• Create, every lane of the Create tab, local and hosted
• Install, the installers for Windows, Linux and the Mac
• Settings and troubleshooting, when a run stops with a message