LTX and Hailuo on Mac: what can yours run?
You can start making local AI video on a 16 GB Apple Silicon Mac. See where LTX-2.5, Hailuo H3, songs and character training fit, with disk sizes and real render times.
guide · hardware · ltx-2.5 · hailuo-h3 · apple-silicon
Before you shop for a bigger Mac, check the one you have. An Apple Silicon Mac with 16 GB of memory can already make AI video with sound in Phosphene. More memory opens up more modes and larger frames. It also makes room for Hailuo H3, starting at 36 GB.
Open Apple menu → About This Mac → Memory. Find your number below. These are the limits for Phosphene 4.19.
The app checks for you
Phosphene reads your unified memory when it starts and shows the modes that fit. On a smaller Mac, modes that need more memory stay hidden. Some also need extra model files; the app tells you which ones and how big the download is.
LTX-2.5 on Mac: start here
LTX-2.5, from Lightricks, is the default video engine. It makes the picture and sound together. At 16 GB, you can start with a text prompt or a picture.
| Memory | What you can do |
|---|---|
| 16–47 GB | Text to video and picture to video at smaller sizes. Keyframes, Extend and High are hidden. |
| 48–79 GB | All modes, including trained characters with Q8 weights. Keyframes and Extend are capped at 768 px on the long side. |
| 80–119 GB | Most modes at full size. Keyframes and Extend go up to 1024 px. |
| 120 GB + | No size limits. |
For reference, a standard 1280×704 render peaks around 22 GiB of resident memory; the Q8 high-quality path peaks around 38 GiB. The app adjusts what’s available to your hardware.
The base LTX model files take about 27.5 GB of disk. Q8 weights for trained characters add 30 GB, and the High add-on is another 29.5 GB.
Hailuo H3 on Mac: bring in the dialogue
MiniMax’s Hailuo H3 makes video, spoken lines and sound from one prompt. It also handles keyframes, extended shots, lip-sync to your audio and sound for silent clips.
- 36 GB and up: use the compact Q8 engine. The installer builds it for you.
- 60 GB and up: the full bf16 engine is available too.
- Disk: about 75 GB, installed with one click inside Phosphene.
On an M4 Max, a five-second shot at Draft with the Fast adapter takes about five minutes including Upscale & Face Fix, delivered at 1280×768. High with Fast takes about 16.5 minutes for 2048×1152. More detail takes more time.
Music Studio: songs from 24 GB
YuE2 makes songs with vocals from your lyrics and a style. You’ll need at least 24 GB of memory and about 11 GB of disk for the model. The engine peaks at about 11 GB of memory. A three-minute song at Final takes about three minutes to render on an M4 Max with 64 GB.
Put yourself in the cast
Full character training needs 64 GB for the rank-32, 512 px recipe. Smaller Macs use a lighter recipe. After training, Phosphene measures how much the file changed the model, so you can check that before committing to a render.
Training uses LTX-2.3 weights. The app offers to download them when you open the Train tab.
Make room on disk too
Memory and disk space are separate requirements. Set aside about 40 GB of disk for the first install, then add 75 GB for Hailuo H3 and 11 GB for Music Studio. Extra LTX weight packs need their own room, as listed above.
Downloads can resume if the connection drops. Once your models are installed, you can render offline.
The Macs this won’t run on
Intel Macs aren’t supported. Neither is Windows. Phosphene uses MLX on Apple Silicon. Linux on M1/M2 is available as a community-maintained port with best-effort support.
If your Mac fits, install Phosphene and see what it can make. You don’t need the biggest machine to have a first go.
Your Mac's turn. Get Phosphene free and make your first scene. Open source. Offline once the models are downloaded.