Skip to content

Write the scene. Shoot it.

Vishot stages a scene from text in Unreal® Engine or in an editor right in your browser and shoots a film from it on your graphics card. Everything behind this text was shot by the models on one NVIDIA GeForce RTX 5090.

Create accountWatch the films

Early access. The account is free, Studio opens with a sample scene, we send the Unreal plugin by invitation, cloud generation is on paid plans from $25 a month.

LocalThe plugin and local models are free and run on your PC. The cloud only where you turned it on.

On screenCoastal Roadroad movie

On screenLast Fleetspace opera

On screenIron Frontmilitary sci-fi

On screenDragon Passfantasy

On screenThe Chargehistorical epic

On screenGround Controlscience fiction

On screenCoastal Roadroad movie

Shot with Vishot

Eight short films from script to edit on one NVIDIA GeForce RTX 5090. Here are 5-second excerpts, more excerpts and the numbers of each film are in Demo.

DemoAll cases

These eight films were shot by the model pipeline without Unreal: frames come straight from the model. Last Unit, the first film staged in Unreal, is shown below in “The staging and the frame”.

Coastal Road: Aerial shot at golden hour: a serpentine road in green cliffs above a turquoise sea, a small red convertible driving along it.
Coastal RoadMaya and Leo drive along the coast in an old convertible and decide where to turn before sunset.
Last Fleet: A fleet of massive grey ships in formation above a banded gas giant.
Last FleetCaptain Nadia, in the command chair, leads twelve ships against forty.

Six more films

What was shot in each and how

All eight were made by one pipeline: frames by Krea 2, motion by LTX 2.5 or MiniMax H3, voice by Seed-VC, upscale by SeedVR2.

How it works

You describe the scene in words. Vishot splits it into shots, stages it in Unreal or in Studio if you want, and the models shoot every shot on your graphics card or, where you choose, through the portal. Below is the whole path of one film, Coastal Road.

You give

  • Place and time“A Mediterranean coast, golden hour”
  • CharactersA few words about their looks, or a photo
  • LinesWho says what, and how
  • Staging, if you wantA scene in Unreal Engine or in Vishot Studio

You get

  • An MP4 film, 1080pEdited, with a shared grade and grain
  • Character voicesEach one sounds the same in all of their shots
  • .srt subtitlesA separate file, timed to the lines
  • Every shot separatelyThe key frame and the clip, to take into your own edit

  1. You write the scene

    Like a screenplay or in plain words: where and when, who is in the frame, who says what. That is enough to start.

    Ext. Coastal road. Late afternoonA red convertible winds along the coast road above the sea. Maya, mid thirties, white linen shirt, freckles. Leo, a little older, short beard, at the wheel.

    Maya(playfully)Okay, honest answer. Left to the mountains, or right to the sea?

    Leo(warmly)I picked the last three. Your turn to surprise me.

  2. Vishot splits it into shots

    The director model on your PC makes the storyboard: which shot, who is in the frame, how the camera moves, which line. MiniMax H3 shoots the shots with lines, LTX 2.5 the shots without words.

    11 shots5 with lines6 silent

    Coastal Road storyboard
    No.ShotIn the frameModel
    01AerialThe coast road above the sea, a red convertibleLTX 2.5
    02TrackingThe convertible among the pinesLTX 2.5
    03MediumMaya: Left to the mountains, or right to the sea?MiniMax H3
    04MediumLeo: Your turn to surprise me.MiniMax H3
    05AerialThe fork: the mountains or the seaLTX 2.5
    06Close-upMaya: The sea. Always the sea.MiniMax H3
    07TrackingDriving into the harbourLTX 2.5
    08WideThe empty convertible on the pierLTX 2.5
    09Two-shotLeo: Same time next year?MiniMax H3
    10Close-upMaya: Same road. Different turn.MiniMax H3
    11AerialVillage lights at duskLTX 2.5

  3. Staging, when precision matters

    When it matters who stands where and where the camera is, Vishot stages the scene in Unreal Engine or in Vishot Studio: mannequin actors on marks, cameras, light. The model then shoots from this staging. Coastal Road was shot without it, straight from the text.

    Compare the staging with the finished frame

    Staging in Unreal Engine: a workshop with a workbench, a lamp and a spotlight, the actors' marks on the floorUnreal Engine
    Vishot Studio: a room, two mannequin characters, the view through a cameraVishot Studio

  4. A key frame for every shot

    Krea 2 paints the first frame of every shot: the light, the lens, the same faces and clothes in all the shots. Frames are a separate step, so you can see the film before any video is computed.

    18 s per frame3 min 20 s for all 11

    1. 1. The serpentine road above the sea, from a drone
    2. 2. The convertible among the pines
    3. 3. Maya and Leo in the car, medium shot
    4. 4. Leo smiles at the wheel
    5. 5. The convertible at the fork in the road, from a drone
    6. 6. Maya and Leo laugh in the convertible, the sun behind them
    7. 7. Driving into the harbour
    8. 8. The convertible on the pier at sunset
    9. 9. Leo and Maya on the bonnet at sunset
    10. 10. Maya and Leo smile at each other in the warm sun
    11. 11. Village lights at dusk, from a drone

  5. The frame comes alive and speaks

    MiniMax H3 moves the frame and speaks the line, and the lips match the speech. LTX 2.5 shoots the shots without words. Seed-VC carries each character's voice, so Maya sounds the same through the whole film.

    Shot 0673 s for this clip

    Maya laughs in the convertible and says: The sea. Always the sea.MiniMax H3 and Maya's voice

  6. The film

    SeedVR2 brings every shot up to 1080p, then come a shared grade, grain, subtitles and the edit. You get an MP4 and a separate .srt subtitle file.

    55 s of film29 min of generationone RTX 5090

    All eight films in Demo

    The same shot in the film: 1080p, grade and subtitleIn the film

The staging and the frame

On the left, the scene in Unreal Engine: a workshop, mannequins on their marks, the camera and the light. On the right, the frame the model shot from this staging for the film Last Unit.

Staging in Unreal Engine: a dark workshop, a mannequin on its mark in a spotlight beam, a workbench with tools on the leftThe model's frame: Juno the mechanic in goggles and the robot K-7 in the workshop, a workbench with a lamp on the leftUnreal Engine 5.8Model frame
Last Unit, wide shot. The workbench on the left and a mannequin in the beam of light; in the film K-7 stands where the mannequin was.Drag the edge or use the arrow keys
  1. Unreal viewport: a workshop with yellow and black striped columns, a workbench, a floodlight on a stand
    1. SetThe plugin builds the workshop, the marks, the cameras and the light
  2. Draft render from Unreal: mannequins on their marks in the workshop
    2. Draft renderMovie Render Queue, mannequins on marks
  3. The end of Last Unit: the robots' eyes light up one by one
    3. FilmThe robots switch on one by one

The frame does not match pixel for pixel yet. Unreal sets the shot, the angle and who stands where; the model paints the frame anew from the character references. Exact overlay by depth and pose from Unreal is still in the works.

Without Unreal. You can build the same staging in Vishot Studio right in your browser. See what it looks like

A film set in your browser

No Unreal? Vishot Studio stages the scene right on the site: a room, people on marks, cameras and lines. The models get the staging itself, shot by shot, not a description in words.

  1. A room from ready-made pieces

    Start from a template: a room, a village street or a forest clearing. Or build your own from walls with doors and windows, a table, a bench, chairs and a bed. A prop dropped on the table lands on top, and four walls in a row close the room.

    • + Add
    • Ready-made
    • Grid
    • Ruler
    • Group
    The Add menu in Studio: shapes and ready-made props, from walls with doors and windows to a table, a bench, a bed, stairs and a lantern. Behind the menu, a room with a table and two characters.
    1 / 5The Add menuShapes and ready-made props

  2. People on marks

    A character stands, sits or lies down and takes height and direction from the furniture. The head follows whoever you point it at, and in the shots of a scene everyone looks at the speaker. A photo dropped on a character becomes its reference.

    • Standing
    • Sits on
    • Lies on
    • Head looks at
    The character inspector in Studio: the pose is Sitting and the head looks at Character 2. On the left, in the room, the first character sits at the table and the second stands next to it.
    2 / 5Character inspectorBody, head, gaze

  3. Cameras, as on a set

    Find the angle and press “Camera here”: the camera goes exactly there, with a lens for a 16:9 frame. You can look through it and steer it with the mouse, the wheel and WASD. The plan from above shows the marks and the cone of every camera.

    • Camera here
    • Look through
    • Plan from above
    Top view in Studio: a room with a bed and a table, the labels Character 1 and Character 2, and on the left the Wide camera with its view cone.
    3 / 5PlanMarks, labels, camera cone

  4. Lines become shots

    For every line: who speaks, with what emotion, in which shot, who they look at and over whose shoulder. Studio places the camera by the rules, “Frame” shows the shot through its eyes, and “To camera” turns it into a camera you can adjust.

    • Wide
    • Two shot
    • Medium
    • Over the shoulder
    • Close-up
    • Frame
    • To camera
    The scene table in Studio: three lines, “Good evening.”, “You too.”, “Sit down.”, with the columns Speaker, Emotion, Shot, On, Over and Camera and the buttons Frame and To camera.
    4 / 5Scene tableLines, shots, cameras

  5. A package for the model

    Studio renders every shot in three layers: colour, depth and mask. With photo references the package goes into the same pipeline that shoots Unreal scenes, in a small free app on your PC that installs with the plugin. “First frames” cheaply show what you will get.

    • First frames
    • Animate
    • Generate film
    • Local
    The shot package from Studio: three shots of the scene, each in three layers. Colour, depth in shades of grey and a mask where the characters are filled with blue and purple.
    5 / 5Shot packageColour, depth, mask
Open the sample scene

Free: the sample scene, one of your own and export to .vishot. Generation on your PC with the plugin. The editor is made for a computer. On a phone you can look through a scene and fix the lines.

A generated video cannot be reshot. A scene can.

A generator invents the frame from scratch every time, so in the next shot the heroine is already another person. In a scene she is one asset: you change the camera, not the actress.

A video generator compared with a scene in Unreal
What we compareVideo generatorScene in Unreal Engine
The heroine in another shotAnother face, another scarf, another heightThe same asset: face, costume, proportions
Changing one shotRegenerate the clip and hopeMove the camera, only that shot is rebuilt
Camera and lightDescribed in words in a promptSet by hand: marks, lens, key light
What you getAn MP4 fileA sequence in Sequencer with actors, cameras and voices
Where it runsOn someone else's server, your script goes thereOn your PC

What runs at home and what goes online

Unreal scenes, voices and model files stay on your PC. Vishot Studio scenes are stored in your account. The cloud appears only where you turned it on yourself, and such a role is always marked.

Your PC

  • Breaking the scene into shotsQwen3 14B or 8B, through LM Studio or OllamaLocal
  • Staging and renderingUnreal Engine: Sequencer, Movie Render Queue. Or Vishot Studio and a small free app on your PC that installs with the pluginLocal
  • Frames and videoKrea 2, LTX 2.5, SeedVR2 in ComfyUILocal
  • VoicesPiper for drafts, Seed-VC for one voice per characterLocal

Local models come under their own licenses. MiniMax H3 is offered through the portal (cloud); its open weights may not be used in the US, EU, UK and South Korea. Krea 2 and LTX 2.5 limit free commercial use by company revenue; check their licenses before commercial work. Model licenses in the Terms

Internet

  • Video in the cloudMiniMax H3, Kling, Veo, Seedance through the portal: on a paid plan or with your own keys, if your card can't cope or you chose soCloud
  • Final voicesElevenLabs, only if you switch it onCloud
  • AccountEmail, devices, plan, Vishot Studio scenes and their photo referencesCloud

The dashed line in the middle is the edge of your computer. Everything on the left works offline.

Which graphics card you need

The main limit is not speed but video memory. Staging and generation share the same card, so they run in turns. Pick your card to see what fits.

Tested: everything local

1. Staging

Qwen3 14B breaks down the scene, the rest of the memory goes to Unreal or Studio.

Qwen3 14B 10.5 GBUnreal or Studio: 21.5 GB

2. Generation

Before Unreal renders, the generation models are unloaded, so they do not get in each other's way.

The largest model we measured: 19.5 GBLocal, tested

Why in turns

Unreal and a generation model don't fit together even in 32 GB. A small free Vishot app on your PC, installed with the plugin, unloads the models before each render and generates after it.

A small card is not a dead end

If generation doesn't fit, these steps go to the cloud through the portal: on a paid plan or with your own keys. Staging, script and voices stay on the PC.

Director model

On 24 GB and more we take Qwen3 14B, on 12–16 GB Qwen3 8B. On 8 GB it is better to give the director to the CPU or the cloud and leave the memory to Unreal.

Tested on an RTX 5090 with 32 GB. For other cards this is a calculation from the model sizes in our catalogue (Qwen3 14B Q4_K_M 10.5 GB, Qwen3 8B 6.5 GB) and from the 19.5 GB peak we measured on Qwen Image Edit. We are still measuring the other models and will refine the numbers after tests on these cards.

What you need

To stage scenes and shoot the film on your own PC. Without an NVIDIA card you can still work in Vishot Studio in the browser.

Compare plans
System
Windows 10 or 11. Mac and Linux are not supported yet.
Graphics card
One NVIDIA GeForce RTX. 16 GB of VRAM is a comfortable start, 24–32 GB leave room for both the model and the scene. The picker above shows what fits on your card.
The Unreal path
Unreal Engine 5.8.
Software
Python 3.11 or newer and LM Studio for the local language model.
Installation
A few terminal commands, step by step in the docs.

In development

Our own Vishot model

We are making our own generation model for Vishot. Once it passes our tests, its first frames, card requirements and terms will appear here.

Tell me when it is out
Project
Vishot Model
Scene
01
Take
0
Status
Tests

The first take is yours

The account is free. Open the sample scene in Vishot Studio, build your own and export it. Generation on your PC with the plugin, which we send by invitation.

Create a free accountCompare plans

Making a series, an animatic or previs? We run pilots with studios.

News about Vishot's own model

Only about it: first frames, GPU requirements and terms.