How to make AI video on a Mac (2026)
Make AI video on a Mac with Phosphene, Draw Things, ComfyUI or LTX Desktop. Compare memory needs, licences and cloud options.

guide · mac · ai-video · comparison
Use a local app, build a workflow in ComfyUI, or use a cloud service. Local apps run models on your Mac; cloud tools run them on a server. We make Phosphene, and this guide explains where our app fits and where alternatives are ahead.
What your Mac needs
Memory is the gate. Phosphene can run LTX-2.5 on Apple Silicon Macs with 16 GB of memory. Its songs and Hailuo H3 also start at 16 GB. Will my Mac run it? lists what each memory size can make. Other apps have different requirements.
Leave room for models. Phosphene downloads 27.5 GB before its first Start, then about 10.7 GB in the background. H3 adds about 75 GB of downloads.
Expect minutes per clip. For a five-second LTX-2.5 Balanced clip, Phosphene’s fleet medians (at 480p) were 10m40s on a 16 GB M4 and 4m55s on a 16 GB M5, measured 2026-08-10 to 2026-10-22. These anonymous usage counts include model loading. See render times by chip and the Mac memory guide before choosing settings.
Phosphene: local AI video on Mac
Phosphene is a free, open-source MIT app by Mr Bizarro. It uses Apple’s MLX to make video with sound, songs and images locally.
Its video engines are LTX-2.5 and Hailuo H3. Both generate sound with the picture; H3 can produce dialogue you can hear. There are also keyframes, extend, lip-sync to your audio, sound for silent clips, local training for your own original characters and styles from 15 to 500 images, and a timeline editor.
Install through Pinokio, a free launcher. Phosphene opens as a local web panel. A free native Mac app is in closed alpha, with no public date.
MIT covers the app. The models have separate licences, and H3’s licence excludes the EU, UK, South Korea and US. Anonymous usage counts are on by default and can be disabled in Settings.
Good fit if: you want local picture, sound and music tools together.
Where it falls short: large downloads, minutes per clip, no node graph, no cloud fallback and no iPhone or iPad version. There is no public signed app or App Store install yet.
Draw Things video
Does Draw Things generate video? Yes. Draw Things lists Wan 2.2, LTX-2.3 and MiniMax H3, with audio confirmed for LTX-2/2.3.
It is a free App Store app with in-app purchases for iPhone, iPad, Mac and Apple Vision. Its App Store description does not list LTX-2.5.
The free Community tier includes some Cloud Compute. Draw Things+ costs $8.99/month and increases cloud limits; Boost packs are also available.
Its public engine repository is GPL-3.0. The README says the app lives in a private repository that syncs with the public one.
Good fit if: you want an App Store install, mobile use or optional cloud compute.
Where it is ahead of Phosphene: a signed app, iPhone and iPad support, cloud compute, a much broader selection of image models and LoRA training, and a long track record.
ComfyUI Mac video
Can ComfyUI make video on a Mac? Yes. ComfyUI Desktop supports Apple Silicon Macs on macOS 13+, though video on MPS can still hit model-specific bugs.
ComfyUI builds workflows from nodes. Its broad model and custom-node ecosystem includes day-0 MiniMax H3 support (per Comfy’s blog), and an MIT-licensed MLX custom node supports LTX-2.3 talking video.
The Mac route uses PyTorch’s MPS backend. An open issue reports corrupt Wan 2.2 5B frames on an M4 Max with 48 GB. Check the issue tracker for your chosen model.
Install ComfyUI Desktop for macOS. ComfyUI is GPL-3.0; optional Comfy Cloud starts at $20/month for Standard, with five free runs for new users.
Good fit if: you want to control and connect individual generation steps.
Where it is ahead of Phosphene: node-level control, ecosystem breadth, early model support and cloud GPUs when your Mac is too small.
LTX Desktop Mac requirements
LTX Desktop comes from Lightricks, the company behind LTX. It is free and open source under Apache-2.0, with optional paid cloud API features.
Full local generation needs Apple Silicon, at least 48 GB of memory and macOS 26+; 32 GB Macs are limited to 540p and five seconds. Under 24 GB and on Intel Macs, it is API-only. Local generation on macOS 14 and 15 is listed as “Coming soon”.
It runs LTX-2.5 locally (its 1.2.7 release fixed an LTX 2.5 Fast memory freeze on Apple Silicon) and adds tools including retake, extend and IC-LoRAs; 1.3.1 shipped on 2026-10-19.
Good fit if: you want Lightricks’ own desktop workflow and meet its local requirements, or want its API route.
Where it is ahead of Phosphene: direct development by the model maker, retake, extend and IC-LoRA tools, and a cloud API option for smaller Macs.
Other open-source routes
These are worth considering if you are comfortable with more setup:
- mlx-video: an MIT-licensed Python package using MLX. It supports LTX-2/2.3 with audio and Wan 2.1/2.2.
- FreeVideo: a local H3 engine and ComfyUI plugin. Code is Apache-2.0; weights use the MiniMax H3 Community License. Its Mac DMG is a preview tested on an M5 with 24 GB and is not notarised.
- WanGP: installs through Pinokio and supports Wan 2.1/2.2, H3, LTX-2/2.3/2.5 and Hunyuan. Its README does not mention Mac support.
- Apple’s mlx-examples: includes Wan2.1 text-to-video and image-to-video examples. Apple does not offer a video-generation app.
Their appeal is direct access to packages, examples or plugin workflows. Check each project’s installation instructions and model terms.
The options at a glance
Model names below reflect the listed projects and sources. App licences and model licences are separate.
| App | Video models (as listed by each project) | Sound with the picture | Install | Price | Licence | Runs on |
|---|---|---|---|---|---|---|
| Phosphene | LTX-2.5, Hailuo H3 | Both engines | Pinokio | Free | MIT app; separate model terms | Apple Silicon Mac, 16 GB+ |
| Draw Things | Wan 2.2, LTX-2.3, MiniMax H3 | Confirmed for LTX-2/2.3 | App Store | Free with purchases; Plus $8.99/month | Public engine GPL-3.0; app repository private | Mac, iPhone, iPad, Apple Vision |
| ComfyUI Desktop | Many, through nodes (Wan 2.2, H3; LTX-2.3 via an MLX custom node) | Depends on the model and workflow | Desktop installer | Optional cloud from $20/month | GPL-3.0; node licences vary | Mac: Apple Silicon, macOS 13+ |
| LTX Desktop | LTX-2.5 | Not stated in the sources we read | Desktop app (GitHub releases) | Free; optional paid API | Apache-2.0 app | Mac: local requirements above; API route for smaller or Intel Macs |
Cloud tools
Google Veo 3.1 generates native audio and is available through products including Gemini and Flow.
Runway lists models including Gen-4.5 and Gen-4 Turbo. Its free tier includes 125 one-time credits; Standard costs $15/month and includes music, voice and sound effects. Luma lists Ray3.2, Veo 3.1, MiniMax H3 and Kling 3.0, with audio on some models; Plus costs $30/month.
Kling describes Kling 4.0 as “audio-visual”. Hailuo’s own app has H3. Sora has been discontinued: OpenAI ended its app and web service on 2026-04-26 and its API on 2026-09-24.
Cloud is useful on any Mac, including 8–16 GB machines, when you want to avoid 30–100 GB downloads or use larger hosted models, longer clips and higher resolutions. Rendering happens on someone else’s GPU and still takes minutes.
The trade-offs are credits or a subscription, prompts and uploads sent to a server, and the provider’s content rules.
How to choose
- 16 GB MacBook Air: on Apple Silicon, Phosphene’s LTX-2.5 is a local option.
- Want iPhone or iPad too: Draw Things covers those devices.
- Want node control: choose ComfyUI’s workflow approach.
- Want dialogue scenes: consider Phosphene’s H3 engine, which runs from 16 GB, where its licence applies.
- In the EU, UK, US or South Korea: Phosphene’s H3 licence does not cover your territory. Consider its LTX-2.5 engine instead.
- No disk space for models: cloud tools avoid the large local downloads.
Your first local clip in Phosphene
- Install through Pinokio. Follow the download instructions, search for Phosphene and choose Install.
- Wait for the downloads. Start appears after 27.5 GB; about 10.7 GB more downloads in the background after you start.
- Choose the video settings. Open Video, select LTX-2.5, choose Text, then Balanced and 5 s.
- Write the scene and sound. Describe the action, camera movement and what should be heard.
- Choose Generate. The first render loads and warms up the engine, so time the second for comparison with the benchmarks.
The documentation covers the controls. The gallery shows clips made with the app.
Sources
- Phosphene repository: https://github.com/mrbizarro/phosphene (read 2026-10-22)
- Phosphene benchmarks: https://phosphene.video/benchmarks/ (read 2026-10-22)
- Phosphene H3 licence note: https://phosphene.video/docs/hailuo-h3/ (read 2026-10-22)
- Draw Things App Store: https://apps.apple.com/us/app/draw-things-ai-generation/id6444050820 (read 2026-10-22)
- Draw Things site and pricing: https://drawthings.ai (read 2026-10-22)
- Draw Things public engine: https://github.com/drawthingsai/draw-things-community (read 2026-10-22)
- ComfyUI Desktop macOS: https://docs.comfy.org/installation/desktop/macos (read 2026-10-22)
- ComfyUI repository: https://github.com/Comfy-Org/ComfyUI (read 2026-10-22)
- ComfyUI MPS issue: https://github.com/Comfy-Org/ComfyUI/issues/16644 (read 2026-10-22)
- Comfy blog, MiniMax H3 day-0 support: https://blog.comfy.org/minimax-h3-day-0-support-in-comfyui/ (read 2026-10-22)
- ComfyUI LTX MLX custom node: https://github.com/noreff/ComfyUI-LTX-MLX-A2V (read 2026-10-22)
- Comfy Cloud pricing: https://www.comfy.org/cloud/pricing (read 2026-10-22)
- LTX Desktop: https://github.com/Lightricks/LTX-Desktop (read 2026-10-22)
- LTX Desktop releases: https://github.com/Lightricks/LTX-Desktop/releases (read 2026-10-22)
- mlx-video: https://github.com/Blaizzy/mlx-video (read 2026-10-22)
- Apple mlx-examples: https://github.com/ml-explore/mlx-examples (read 2026-10-22)
- FreeVideo: https://github.com/FlashML-org/FreeVideo (read 2026-10-22)
- WanGP: https://github.com/deepbeepmeep/Wan2GP (read 2026-10-22)
- Sora discontinuation: https://help.openai.com/en/articles/20001152-what-to-know-about-the-sora-discontinuation (read 2026-10-22)
- Google Veo: https://deepmind.google/models/veo/ (read 2026-10-22)
- Runway pricing: https://runway.com/pricing (read 2026-10-22)
- Luma pricing: https://lumalabs.ai/pricing (read 2026-10-22)
- Kling: https://kling.ai/ (read 2026-10-22)
- Hailuo: https://hailuoai.video (read 2026-10-22)