Comfy-Org/desktop

Fork 0

mirror of https://github.com/Comfy-Org/desktop.git synced 2026-11-04 23:47:43 +00:00

[GH-ISSUE #1252] [Feature Request] Native MLX (Apple Machine Learning Extensions) Backend Support #352

New issue

Closed

opened 2026-10-31 22:11:55 +00:00 by bsakaguchi · 2 comments

bsakaguchi commented

2026-10-31 22:11:55 +00:00

Owner

Originally created by @karl-cheng on GitHub (Aug 16, 2025).
Original GitHub issue: https://github.com/Comfy-Org/desktop/issues/1252

Prerequisites

I have searched the existing issues and checked the recent builds/commits
I have read the documentation and this feature is not already available
This is a single feature request (not multiple features combined)

Problem Statement

Currently, MLX (Apple's Machine Learning Extensions) can only be used in ComfyUI via community-made custom nodes. This requires extra steps for installation, limits integration with core workflows, and prevents direct use of MLX-optimized models alongside other model formats.

Proposed Solution

Please add native, first-class support for MLX as an official backend within ComfyUI/Comfy Desktop, similar to the support for PyTorch or CUDA. This would allow users on Apple Silicon (M1/M2/M3) to load, run, and manage MLX and Core ML models directly—without relying on external custom nodes or manual model conversion.

Key benefits:

Seamless integration with Comfy Desktop's standard model loader and workflow nodes.
Improved performance and resource management leveraging Apple Silicon hardware.
No need for separate (or redundant) custom node installation/maintenance for MLX support.
Broader range of out-of-the-box model format compatibility for Mac users.

Potential workflow

Open Comfy Desktop on an Apple Silicon Mac (M1/M2/M3)
Go to Settings/Preferences and find a new "Backend" or "Compute" section
Select MLX as the backend option alongside existing PyTorch/CUDA options
Download or load MLX-compatible models directly through the standard model loader
Create workflows using standard ComfyUI nodes that automatically leverage MLX acceleration
Models would seamlessly integrate with existing workflows without custom node requirements

Impact Level

High (Major improvement)

Alternatives Considered

Current alternatives involve installing and maintaining custom MLX nodes, which may lag behind core updates, and manual model conversion and management, leading to duplicated effort and a fragmented user experience.

Additional information

MLX is Apple's new, officially supported, PyTorch-like ML framework, optimized for macOS and iOS hardware. Native support would make Comfy Desktop more accessible for the growing Apple developer and creative community.

See: https://github.com/apple/mlx for reference.

┆Issue is synchronized with this Notion page by Unito

Originally created by @karl-cheng on GitHub (Aug 16, 2025). Original GitHub issue: https://github.com/Comfy-Org/desktop/issues/1252 ### Prerequisites - [x] I have searched the existing issues and checked the recent builds/commits - [x] I have read the documentation and this feature is not already available - [x] This is a single feature request (not multiple features combined) ### Problem Statement Currently, MLX (Apple's Machine Learning Extensions) can only be used in ComfyUI via community-made custom nodes. This requires extra steps for installation, limits integration with core workflows, and prevents direct use of MLX-optimized models alongside other model formats. ### Proposed Solution Please add native, first-class support for MLX as an official backend within ComfyUI/Comfy Desktop, similar to the support for PyTorch or CUDA. This would allow users on Apple Silicon (M1/M2/M3) to load, run, and manage MLX and Core ML models directly—without relying on external custom nodes or manual model conversion. Key benefits: - Seamless integration with Comfy Desktop's standard model loader and workflow nodes. - Improved performance and resource management leveraging Apple Silicon hardware. - No need for separate (or redundant) custom node installation/maintenance for MLX support. - Broader range of out-of-the-box model format compatibility for Mac users. ### Potential workflow 1. Open Comfy Desktop on an Apple Silicon Mac (M1/M2/M3) 2. Go to Settings/Preferences and find a new "Backend" or "Compute" section 3. Select MLX as the backend option alongside existing PyTorch/CUDA options 4. Download or load MLX-compatible models directly through the standard model loader 5. Create workflows using standard ComfyUI nodes that automatically leverage MLX acceleration 6. Models would seamlessly integrate with existing workflows without custom node requirements ### Impact Level High (Major improvement) ### Alternatives Considered Current alternatives involve installing and maintaining custom MLX nodes, which may lag behind core updates, and manual model conversion and management, leading to duplicated effort and a fragmented user experience. ### Additional information MLX is Apple's new, officially supported, PyTorch-like ML framework, optimized for macOS and iOS hardware. Native support would make Comfy Desktop more accessible for the growing Apple developer and creative community. See: https://github.com/apple/mlx for reference. ┆Issue is synchronized with this [Notion page](https://www.notion.so/Issue-1252-Feature-Request-Native-MLX-Apple-Machine-Learning-Extensions-Backend-Support-2516d73d365081ab950efddb9db77e77) by [Unito](https://www.unito.io)

bsakaguchi

2026-10-31 22:11:55 +00:00

closed this issue
added the
enhancement
label

bsakaguchi commented

2026-10-31 22:11:55 +00:00

Author

Owner

@lucasye2008 commented on GitHub (Aug 16, 2025):

Add 1

I love ComfyUI and also is a MacBook Pro M4max user. I always faced many of issues or limits while using the customs nodes which not supported MPS/MLX and hope we could have resolutions.

@lucasye2008 commented on GitHub (Aug 16, 2025): Add 1 I love ComfyUI and also is a MacBook Pro M4max user. I always faced many of issues or limits while using the customs nodes which not supported MPS/MLX and hope we could have resolutions.

bsakaguchi commented

2026-10-31 22:11:55 +00:00

Author

Owner

@dosubot[bot] commented on GitHub (Nov 15, 2025):

Hi, @karl-cheng. I'm Dosu, and I'm helping the desktop team manage their backlog and am marking this issue as stale.

Issue Summary:

You requested native support for Apple's MLX as an official backend in ComfyUI/Comfy Desktop to enhance performance and compatibility on Apple Silicon Macs.
This would replace the need for custom nodes and allow MLX to be selected alongside PyTorch and CUDA.
Another user expressed support, noting issues with current custom nodes on MacBook Pro M4max.
The issue remains unresolved with no recent updates or progress.

Next Steps:

Please let me know if this issue is still relevant with the latest version of the desktop app by commenting here.
If I don’t hear back within 7 days, the issue will be automatically closed.

Thanks for your understanding and contribution!

@dosubot[bot] commented on GitHub (Nov 15, 2025): Hi, @karl-cheng. I'm [Dosu](https://dosu.dev), and I'm helping the desktop team manage their backlog and am marking this issue as stale. **Issue Summary:** - You requested native support for Apple's MLX as an official backend in ComfyUI/Comfy Desktop to enhance performance and compatibility on Apple Silicon Macs. - This would replace the need for custom nodes and allow MLX to be selected alongside PyTorch and CUDA. - Another user expressed support, noting issues with current custom nodes on MacBook Pro M4max. - The issue remains unresolved with no recent updates or progress. **Next Steps:** - Please let me know if this issue is still relevant with the latest version of the desktop app by commenting here. - If I don’t hear back within 7 days, the issue will be automatically closed. Thanks for your understanding and contribution!

bsakaguchi referenced this issue

2026-10-31 22:17:08 +00:00

[PR #352] [MERGED] Add Github issue templates #813

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

Comfy-Org/desktop#352

No description provided.

Rows
Columns