USER GUIDE

Storyteller end to end

Every image on this page is a live screenshot of the app, taken on a real project: “The Final Ridge”, a documentary about the 1972 Andes plane crash. Walk through all eight steps and you come out with a finished video file.

Mode picker screen: Storyteller, Storyboard, Dubbing, Editor
This is what you see when the app opens. The guide focuses on Storyteller — the most complete mode.

What Story Machine is

Story Machine is not a button that spits out a video. It is a pipeline: every stage is its own page, the AI does the heavy lifting and then stops so you can review, edit and re-run — before moving on.

🧠

AI writes

Finds the story, splits it into scenes, writes the image prompts and the motion prompts.

🎨

AI draws & animates

G-Labs Automation paints the images and turns them into clips.

🎙

AI narrates

Voice Studio runs offline and reads the narration on the exact timecodes.

How to read this guide: if you already have an .srt file or a script, you can skip step 02 (Story) and go straight from step 01 to step 03.

Requirements & connections

Story Machine conducts; the other two apps do the heavy work. All three run on your own machine.

1
An LLM CLI — Claude CLI, Antigravity, Codex, or the 9Router gateway. This is the brain that writes the story, splits the scenes and writes the prompts.
2
G-Labs Automation running (webhook 127.0.0.1:8765) Used to paint the images and build the videos. Leave it off and those two steps stay locked.
3
G-Labs Voice Studio running (webhook 127.0.0.1:8766) — optional Only needed if you want narration. Skip it for a silent video.
Config panel: choosing the LLM model, image model, thumbnail, voice and video model
The Config panel — one place for all of it: LLM model, image model, thumbnail, voice, video model and where the output is saved.
While the webhook is not connected, the image and video model dropdowns are greyed out and read “— Connect webhook —”. That is deliberate: it stops you picking a model and only discovering it is broken once the run fails.
STEP 01

Setup

Pick your input source — this is the decision that shapes the whole project:

  • Video from a topic — you type a topic, the AI researches it and writes everything. Runs through all eight steps.
  • Video from subtitles — load an existing .srt file; the subtitle timing becomes the video timing.
  • Video from a script — paste your own narration, and the app splits it into sentences and estimates the durations.
The Setup page with three input modes, aspect ratio and video title
Three mode tiles across the top. Below them: aspect ratio, video title, and where you load the .srt file.

Scroll down to the Visuals section to choose the narrative angle and the visual style — there are 46+ styles built in, from cinematic and documentary to animation, stick figures and layered paper cut-outs.

The Graphics block: narrative angle plus the visual-style grid with a sample image for every preset
The Graphics block: pick the narrative angle (with characters / without characters / animal world), then pick a look from the style grid — every tile is a real preset, and the highlighted one is active.
STEP 02

Story

Only appears if you chose Video from a topic. The AI comes back with four true stories, each with a summary, a reason it works, and source references so you can check it.

Four candidate story cards found by the AI, each with a summary and a reason it works
Four candidates for the topic “The 1972 Andes plane crash”. The highlighted card is the one that was picked.
None of the four work for you? Narrow the topic and search again. The more specific the topic, the more concrete detail comes back in the story for you to build images from.
STEP 03

Scene planning

This is the step that turns words into a visual blueprint. The script is cut into timestamped beats, each beat is tagged by type — narration, cutaway, or atmosphere — and each one gets its own image prompt.

The Prompt page: timestamped beats, narration lines and image prompts
Each block is one scene: the timecode, the narration line, and the image prompt generated from it. All editable in place.
  • The pacing slider at the top sets how long each scene runs — longer scenes mean fewer images, shorter scenes mean a faster cut.
  • Regenerate a single scene when its prompt is not what you wanted, instead of re-running the whole file.
  • Prompts are written in English because image models understand English better — the content still follows your narration line for line.
STEP 04

Images

Sends the prompts over to G-Labs Automation in bulk. Images appear in the grid as they come back.

The generated image grid for The Final Ridge
The image grid of the sample project. The Quick select bar takes the first 10 scenes, the last 5, a random set or every Nth one — then you regenerate just that group.

Work in bulk

Select several scenes, or drag a marquee over them the way you would over files, then regenerate exactly that group. There is a Generate selected (N) button.

Import your own images

Already have artwork of your own? Bulk import and match by filename or by order — with a preview table before anything runs.

STEP 05

Videos

Every image becomes a short clip with motion. The key point: the video prompt describes only the action and the camera move, never the content already in the image — because the image is the first frame.

The Videos page: scene list with a motion prompt for each clip
A motion prompt per scene, with a “Ready” status. Preview the clip right there in the grid.
One camera move per scene. Cramming two motions into a single prompt is the fastest way to end up with a warped clip.
STEP 06

Voiceover

Hands off to Voice Studio, which runs offline. The app reads one sentence at a time on the subtitle timecodes, so the narration always lands with the picture.

The Voice page: sentence list with a regenerate button on each line
One line per sentence. Play it back, and hit the button on the right to re-read any sentence that stumbles.
STEP 07

SEO

Generates the title, description, tags and chapters for YouTube, plus a thumbnail drawn specifically for the film.

The SEO page: thumbnail and YouTube metadata
Thumbnail on top, metadata below. A Copy button beside each field so you can paste straight into YouTube.
STEP 08

Render studio

The final step assembles everything: clips, voiceover, background music, subtitles — on a timeline with waveforms.

The render studio: preview pane and timeline
Preview pane in the middle, timeline underneath, three groups of settings on the left.

Audio settings

Set the voiceover and background music levels, and most importantly the ducking — the music drops itself under the narration and comes back up when the line ends.

The audio settings panel: voiceover level, background music, ducking amount and fade in/out timing
Sliders for the voiceover level, the background music, how far the music ducks, and the transition time.

Subtitle settings

Turn on burn-in subtitles and set the type styling. Subtitles are written into the picture, so they show up wherever the video is played.

The subtitle settings panel in the render studio
Toggle the subtitle display and adjust the styling before you render.
When everything is in place, hit Render Video (or Render Slideshow if you only have stills). Renders run through a queue — leave it going and get on with something else.

The other three modes

🎬

Storyboard

Builds a cinematic storyboard from a script: extract the characters, locations and props → paint reference images → break it into frames → draw every frame consistently.

🎙

Dubbing

Translates and dubs an existing film: pull or transcribe the subtitles → translate with AI → new voiceover → burn in subtitles, mix the audio, render.

✂️

Editor

Cuts any video on a timeline with waveforms. Snaps clips to subtitle lines, to detected silences, or to a BPM beat grid.

Tips & troubleshooting

The image/video model dropdown is greyed out and will not open

G-Labs Automation is not running, or the webhook address is wrong. Start that app, check the webhook at 127.0.0.1:8765, then hit Refresh in the Config panel.

The Voice step says Voice Studio is not configured

Open G-Labs Voice Studio, go to the Webhook tab, start the server (127.0.0.1:8766) and paste the API key into Story Machine. Making a silent video? Hit Skip.

A few scenes came out broken or off-content

Do not re-run the whole project. Select exactly those scenes in the grid and hit Generate selected. If the prompt is the real problem, go back to the Prompt step and fix it first.

The video stutters or characters come out deformed

Usually a motion prompt carrying too much at once. Cut it back to one action and one camera move, for example “slow push in, the animal lifts its head”.

The background music drowns out the narration

Open Audio settings in the render studio, raise the ducking amount and lower the background music level.

The project is huge and slow to reopen

Use Save project / Open project in the sidebar. Each session lives in its own folder under output/; reopen that session and you carry on where you left off.