Skip to content

Demo

Eight short films the Vishot pipeline made from text to finished film on one NVIDIA GeForce RTX 5090.

Here are clips of about 5 seconds from each film. The full films run from 45 to 61 seconds: they are on the studio's PC, we will put them on the portal later.

The films

Aerial shot at golden hour: a serpentine road in green cliffs above a turquoise sea, a small red convertible driving along it.
The red convertible with its top down drives along the coast road under pine trees.
Maya and Leo laugh in the convertible, the sea in a sunny haze behind them.
The sea. Always the sea.
The convertible at the end of a stone pier, the sea and the sunset ahead.
Dusk over the coast: village lights, the pier and the shoreline stretching into the distance.

Shot 01LTX 2.5Aerial shot at golden hour: a serpentine road in green cliffs above a turquoise sea, a small red convertible driving along it.

Road movie

Coastal Road

Maya and Leo drive a red convertible along a coastal road, choose the sea over the mountains and meet the sunset on a stone pier.

Film length
55 s
Shots
11
Generation of frames and video on one RTX 5090
28 min 41 s
Krea 2 frames: 3 min 20 s. LTX 2.5 and MiniMax H3 video: 25 min 21 s.

5 MiniMax H3, with a line6 LTX 2.5, motion without a line

  • Maya
  • Leo
The film's case Coastal Road
films
8
shots: 38 with lines in MiniMax H3, 49 without lines in LTX 2.5
87
of screen time in all
7 min 16 s
of frame and video generation on one RTX 5090
4 h 7 min

Under its licence the local MiniMax H3 weights cannot be used in the US, the EU, the UK or South Korea; there only the cloud MiniMax H3 through the portal is available. Model licenses in the Terms

How one shot is made

Coastal Road, shot 6. One line of Maya's goes through five steps on the same RTX 5090. Generation seconds come from the film's log.

  1. Script

    The shot description and the line. The pipeline takes everything else from this text: what is in the frame, who speaks and what exactly.

    EXT. COASTAL SERPENTINE. LATE AFTERNOONClose on Maya in profile in the passenger seat, looking at Leo and laughing. The sun behind her.MayaThe sea. Always the sea.
  2. Frame

    Krea 2 IntoRealism paints the starting frame of the shot: faces, light, composition.

    18 sframe generation

    Krea 2 starting frame: Maya and Leo in the convertible, the sea in a sunny haze behind them
  3. Motion, speech and lips

    MiniMax H3 brings the frame to life: Maya says the line in the voice the model gives, and her lips move in time.

    73 svideo generation

    Raw MiniMax H3 output: Maya says the line in the model's own voice
    The sea. Always the sea.

    MiniMax H3raw output1344×768

  4. One voice

    Seed-VC turns the line into the same voice Maya speaks with in her other shots. Every character in the film has one voice.

    After Seed-VC: the same line in Maya's voice from her other shots
    The sea. Always the sea.

    Seed-VCthe character's voice

  5. 1080p and grade

    SeedVR2 brings the shot up to 1080p, then ffmpeg lays on the film's grade and grain. Subtitles go as a separate .srt file.

    Final shot: SeedVR2 1080p and the film's grade
    The sea. Always the sea.

    SeedVR21080pgrade

Only the clips in this section have sound, and it turns on only with the button. All other clips on the page are muted.

These films were shot without Unreal

The model generates the frames of these eight films directly: nobody built a set or placed cameras and lights.

They show the second half of Vishot: from the frame to the finished film. The first half, staging the scene in Unreal with actors on marks and cameras, is on the home page.

Staging in Unreal on the home page

Try it on your own script

The account is free: Vishot Studio opens with a sample scene, and you can save and export one scene of your own. Cloud generation on paid plans from $25 a month.