Getting started

1. Install the app

1. Check your Mac. In the Apple menu, open “About This Mac” and confirm the chip is “Apple M…” and macOS is 15 (Sequoia) or later. Intel Macs and macOS 14 or earlier will not work.
2. Download the DMG. Open the latest release page and, under Assets, click H3cApp-0.5.0.dmg (about 19.1 MB; the number changes with the version). The files labeled “Source code” are for developers, not for installing.
3. Open the DMG. Double-click the downloaded DMG file; a window with the H3cApp icon opens.
4. Copy the app to Applications. Drag the H3cApp icon onto “Applications”. If an older version is there, replace it.
5. Launch the app. Double-click H3cApp in the Applications folder. H3cApp is notarized by Apple. If a warning appears at launch, see Troubleshooting.

2. Get the AI model

H3cApp uses MiniMax H3, a video generation model released by MiniMax. The model is not part of the app, so you download it once at the start.

If no model is registered when H3cApp starts, a window titled “Download the MiniMax-H3 Model” opens.

To download

1. Choose where to save it. “Save to” starts at ~/Library/Application Support/H3cApp/MiniMax-H3 on the internal disk. To use another disk, pick a folder with “Change…” (a MiniMax-H3 folder is created inside it). The model is read during generation, so disk speed affects generation time; an external disk is not necessarily as fast as the internal SSD.
2. Decide which models you need. At first, we recommend leaving “Also include Ref2VA (reference images and videos; roughly doubles the size)” off. You can add the reference images/videos/audio feature later.
3. Click “Check Size…”. After it checks the Hugging Face file list, the real total appears, for example “134.2GB in total (81 files)”. Make sure the destination has enough free space.
4. Click “Start Download”. Progress, file count and speed are shown. If you “Cancel” or the connection drops, it resumes where it left off next time. If it fails, click “Resume”.
5. When “Download complete.” appears, click “Done”. The model is registered and starts loading.

If you already have the model

If you already have a MiniMax-H3 folder containing FL2VA (and optionally Ref2VA), choose it with “Add Folder…” in the download window to register it without downloading.

Downloading later, or adding Ref2VA

“Later” closes the download window. To download later, use “Download and Add…” in “Model Manager” (the stacked-squares icon at the top right of the window).

Use the same window to add Ref2VA for reference images, videos and audio. Make the destination the same MiniMax-H3 folder as your current model (no change needed if you first downloaded to the default location; otherwise use “Change…” and choose the folder that contains MiniMax-H3), turn on “Also include Ref2VA” and download. FL2VA files already there are skipped, so only Ref2VA is fetched. When it finishes, the model list gains a second entry for the same folder. The int8 cache is kept per entry, so keep the original entry selected to avoid rebuilding the cache.

Is it ready?

Check the status icon at the top right of the window.

Click the icon to see the model in use and any load error.

3. Check the compute mode

The compute mode decides how the model is loaded and computed. Open “Advanced Settings” at the top right of the panel (or press ⌘,) and look at “Compute mode”. To begin, the mode the app picked is fine.

To use the reference images/videos/audio feature (Ref2VA) with “Fast (int8 cache)”, build a separate Ref2VA cache too (another 19.3 GB or so). How this differs from the speed setting and “Fast mode (experimental)” is explained in the three speed settings.

4. Make and save your first video

Start with one small 5-second video to make sure everything works.

1. Make sure the status icon shows “Ready”.
2. In the switch at the top left of the panel, choose “From text” (selected by default).
3. Copy the example prompt below and paste it into the prompt field.
   Example prompt (a small orange cat playing with a ball of yarn on a wooden floor in soft afternoon light; fixed camera, gentle movement; only quiet room sound, no speech or music) A small orange cat plays with a ball of yarn on a wooden floor in soft afternoon light. Static camera, gentle movement. Soft room ambience, no speech, no music.

   You can also select the text and press ⌘C.

   This is just an example input; it won't necessarily give the same picture or sound.

4. Below the prompt field, set “Shape” to “Square”, “Size” to “Small (256×256)”, “Length” to “5 s”, and the video count to “×1”. Size starts at “Medium (512×512)”, so change it to Small.
5. Leave “Advanced Settings” at the top right of the panel as it is.
6. Click the up-arrow (↑) button at the bottom right, or press Enter in the prompt field, to start (Shift+Enter adds a line).
7. While it runs, the preview shows progress, elapsed time and “About … left (done around …)”. At first it says “Estimating the time left…”. On the author's M5 with 24 GB, this setting took 1 min 44 s.
8. When it finishes, the video appears in the preview. Play it with the play button (▶) and check the sound too.
9. Save it, in one of two ways:
   • “Export Video…” (⌘S): saves an MP4 wherever you like; it starts in Downloads.
   • “Save as Project…” in the project menu (folder icon at the top left): name it, and the current prompt, settings and this video are saved together in ~/Movies/H3cApp/<name>.