Skip to content

Free, open-source AI video generator for Mac. Run LTX-2.5 and MiniMax H3 (Hailuo) locally on Apple Silicon: text-to-video and image-to-video with sound, lip-sync, character LoRA training, songs from lyrics (YuE2) and image editing (Qwen-Image-Edit). One-click install via Pinokio. Offline: no cloud, no API key, no NVIDIA GPU.

2.4 KB
Raw Blame

One Shot

One continuous shot of 30 seconds to 2 minutes that never cuts. You write the whole shot once; Phosphene renders it in parts, each continuing from the last frame of the one before, and joins them into one file.

Making one {#make}

  1. Choose the model in Model, at the top of the column: LTX 2.5 renders 10 s parts and is the one for faces, voices and dialogue; Hailuo H3 renders 15 s parts and is the one for motion, landscapes and crowds.
  2. Write The shot — who, where, the time of day and the weather, the sound — once, for the whole shot. This is the only thing required.
  3. Pick a Length: 30 s, 45 s, 1 min, 1½ min or 2 min. Each chip shows how many beats and parts it makes.
  4. Pick a Quality. The estimate line says about how long it takes on this Mac.
  5. Press Generate (or [[sc:prompt.generate]] in the prompt box). The whole shot is one request in the queue.

The optional parts {#options}

  • Start frame — Choose an image and the shot opens on it.
  • Who is in it (LTX) — pick a trained character: Quality switches to that character's qualities and Hand off where the line ends turns on.
  • Beats — one row per 5 seconds, each with its time stamp. Write what happens in each, or let Phosphene fill them: Split my prompt into beats puts one sentence per beat, Write the beats for me has the planner write them. [[sc:oneshot.beats]] Pasting a list fills the rows.
  • Camera — one move for the whole shot, its direction and speed. Every part continues it.

The three switches {#switches}

Switch On Off
Lock the light the same light in every part (default) each beat as written
Redo a part that drifts a part that drifts is rendered once more, at most (default) keep the first pass
Hand off where the line ends for a talking character: the next part starts where the spoken line stops each part starts from the last frame

Under More: Seed, a Label for the queue, and No music (voice, sound and ambience only).

Watching it render {#status}

The status card updates every few seconds: Rendering, part 2 of 6, and when it is done, in the gallery on the right. For each finished part with speech it reports whether the voice is on the mouth — voice on the mouth, borderline or voice-over — and light drifted when a part's light moved.