Story Mode Guide
Take a story from a blank page to a finished video without leaving the app
🎬 What is Story Mode? v3.0.0
Story mode is a pipeline that turns an idea into a scene table with prompts, voices and sound effects — all inside AutoFlowCut. You give it a title (or a script you already wrote), and it walks through research, synopsis, script, scene split, audio and prompts, handing each step's output to the next one.
You have a subject, a title or a finished script — but no scenes, no prompts and no audio yet. Story mode builds all of that for you.
You already have a storyboard table or subtitles plus an audio package built outside the app. See the File Import Guide.
💡 Tip: The two paths meet at the same place. Whichever way your scenes arrive, generation and export work exactly the same afterwards.
🧭 The Stepper at a Glance
The stepper sits at the top of the Story view. Every step is a chip you can click to go back and look at what it produced. A coloured dot on the chip shows its status: Pending, Running, Done or Error.
0 Setup → ① Research → ② Synopsis → ③ Script → ④ Scene split → ⑤ Audio → ⑥ Prompts
(optional) (gate) [auto] [auto] [auto]
▶ Run all
| Step | Hands to the next step |
|---|---|
| 0 Setup | Genre, model, output language, script length, split unit, review settings, title — and optionally a script you paste or import |
| ① Research | A structural analysis of reference YouTube videos, with fact-checked claims |
| ② Synopsis | A synopsis and a confirmed cast of characters |
| ③ Script | The full script text |
| ④ Scene split | Scenes, segments per speaker (with an emotion) and sound-effect cues |
| ⑤ Audio | A narration track per speaker plus the generated sound effects |
| ⑥ Prompts | An image prompt and a video prompt for every scene |
💡 Note: Research and Synopsis are gates, not generation steps — they are only clickable when they apply to your project. If you import a script that already has scenes fixed to imported images, the steps that would rewrite them stay locked.
0️⃣ Setup
The entry tab. Everything you choose here travels with the project — reopen it later and the form comes back exactly as you left it.
Options
| Field | What you get |
|---|---|
| Story type | The genre offered depends on the output language. Korean projects can pick Yadam (Korean historical tales); English projects can pick Dark history. Both offer Bespoke, the general-purpose default. |
| Generation AI | The model that writes for you — Claude or Codex. Models that support it also show a Reasoning level selector next to the dropdown. |
| Output language | Korean or English. This is the language of the story, not of the app UI. |
| Script length | A number plus a unit. Minutes work in both languages; you can also give a character count (Korean) or a word count (English). Up to 60 minutes. |
| Scene split unit | By scene with a target length range (5–10 seconds by default), or By sentence. See Scene split. |
| Review | An auto-review toggle and a round count (1–5) for the script, the scenes and the prompts. |
| Title | Optional. If you leave it blank and a script exists, AutoFlowCut generates a title for you before splitting scenes. |
Two ways in
Start from a title
Type a title, press ✨ Start, and the pipeline moves to the Synopsis gate, where the AI drafts a synopsis and a cast for you to approve.
Bring your own script
Paste the script into the text box, or drop a .txt / .md file onto it (📁 Choose file works too). Press ✨ Start: the script is saved as-is, then the Synopsis gate pulls the characters back out of it.
💡 Note: Once a script exists, the Setup button becomes ↻ Restart with changes — it only lights up if you actually changed something. It restarts the pipeline from the script, so use it deliberately.
1️⃣ Research optional
Look up what already works on YouTube before you commit to a subject. Research runs before the Synopsis and is available while your project has no script yet — it searches by keyword, so you don't even need a title.
The flow
Search
Enter a keyword, choose how many results (10 / 20 / 30) and an upload period (all time, last week, last 30 days). You can also paste a YouTube URL to add a video by hand.
Pick your references
Results come back as cards with thumbnail, channel, view count and upload date. Tick the ones you want; double-click a card to open its details — subscribers, publish date, viral score and an inline player.
Analyse
Analyze all runs the three stages back to back: fetch transcripts → analyze structure → fact-check. You can also run Fetch transcripts, Analyze structure and Fact-check one at a time, and stop a long run with ⏹ Stop.
Confirm or skip
Use this research saves it with the project and moves on to the Synopsis. Skip goes straight to the Synopsis without it.
Breaks the reference videos into story beats with a summary each, lists the themes they share, and extracts their key claims.
Each claim gets a verdict — Supported, Refuted or Unverified — with source links. Only the ones you tick are carried into the project; by default that is the supported ones.
⚠️ Requires yt-dlp. Research fetches search results and subtitles through yt-dlp. If it isn't installed, the panel shows the exact install command for your OS.
📜 For reference and reconstruction only. Copying or redistributing someone else's subtitles verbatim can infringe copyright. The app repeats this notice at the top of the Research panel.
2️⃣ Synopsis gate
The Synopsis is the one place where a human signs off before the machine runs away with your story. Nothing downstream — no scene split, no audio, no prompts — starts until you confirm the cast here.
What it does
- From a title: the AI writes the synopsis, streaming it into an editable text box, and proposes a cast.
- From your own script: it works backwards — the characters and a synopsis are pulled out of the script you brought.
- If you confirmed a Research step, a Include research context checkbox appears. It is the only switch: research is never injected behind your back.
- The synopsis text is fully editable, with a live line and character count.
The character cards
Each row is one character, and every field is yours to edit. Add or remove rows freely.
| Field | Used for |
|---|---|
| Name | Identifies the speaker in the scene table — and becomes the @mention handle later |
| Gender | Male / Female / Unknown. Pre-filters the voice picker and warns you about a mismatched voice |
| Ethnicity | Free text. Folded into the character's visual description in the image prompt |
| Age | Free text |
| Role | Free text |
| Prompt | The character's appearance. This is the same text the reference card is generated from — see Characters & @mention |
If the script has no characters at all — a pure narration piece, for instance — the panel says so and you can confirm an empty cast and move on.
Review and immersion score
Press Review and the AI reads its own synopsis, scores it for immersion out of 100, and rewrites it. Set how many rounds it gets (1–5). The score badge under the editor shows the first score and the last one, so you can see whether the rewrite actually helped — and if a round makes it worse, that revision is discarded and the log tells you so.
✅ Confirm: from a title, the button reads Generate script from this synopsis and moves straight on to Script. From your own script, it reads Confirm characters — your script is left untouched, and only the cast is committed.
3️⃣ Script
The script is the single source of truth for everything after it. It streams in live as the model writes, then drops into an editor you can rewrite by hand at any time.
Regenerates the whole script from the current title and settings. If the new run fails, the old script stays.
Keeps what is already written and carries on from where it stopped — useful when a long script comes up short.
Same review-and-revise loop as the synopsis, with its own immersion score and round count. Tick Auto review in Setup to have it run on every generation.
Saves your edits and runs the scene split in one action. If the title is still empty, it is generated first.
💡 Note: While generation runs, a stopwatch shows the elapsed time and ⏹ Stop cancels it. A high reasoning level can take a while before the first words appear — the clock tells you it is still working.
4️⃣ Scene Split
Cuts the script into scenes, and each scene into segments. A segment is either a line of narration or dialogue — with a speaker and an emotion — or a sound-effect cue.
Scenes are cut to a target length you set — a minimum and a maximum in seconds (5–10 by default).
One scene per sentence, always cutting on a change of speaker. Very short fragments are merged, and anything over 10 seconds is split.
The scene table
Three columns: the scene number, the speaker, and the segment itself. Character lines carry their emotion underneath — normal, happy, sad or angry. Narration does not: the narrator is never given an emotion. Sound-effect rows are marked SFX and show the effect description instead of a line.
The emotion is not decoration. It is what the TTS engine is told to perform.
🔁 Not happy with the cut? The split unit and the length range sit right above the table. Change them and press ↻ Re-split scenes — you can do this as often as you like.
5️⃣ Audio
Every speaker gets their own voice, and every sound-effect cue gets a sound. The result is a separate narration track per speaker, which is what makes a multi-character story sound like a cast rather than one person reading aloud.
Casting the voices
The speaker list sits at the top of the panel: each row shows the character's name, a gender badge and their appearance. Click the 🎙 button to open the voice picker.
- Any engine, per speaker. Filter by Typecast, Gemini or ElevenLabs — or leave it on All. Different characters can use different engines in the same story.
- Search and preview. Search by name or trait, filter by gender, and hit play on any card to hear the voice before you commit. ElevenLabs' shared library is searched remotely, so type at least two characters to reach beyond the first page.
- Gender tags. Every card is tagged male, female or unknown. If a voice is untagged or the tag is wrong, right-click the card to set it yourself.
- Mismatch warning. If a character's gender and the chosen voice's gender disagree, a ⚠ appears on the button. It doesn't block you — it just makes sure the mistake was on purpose.
- Default voice. Leave a speaker on Default voice and the app falls back to its own default for that engine.
Emotion — characters only
The emotion the scene split assigned to a character's line is passed to the TTS engine, so the delivery matches the moment. The narrator is deliberately excluded: an emoting narrator sounds unhinged, and a steady one carries the story.
Sound effects
Sound effects are pulled out of the script during the scene split and placed on the scenes they belong to — they show up as SFX rows with their description. They are generated with ElevenLabs, and you can audition them one at a time from the row.
Working row by row
| Action | What it does |
|---|---|
| ▶ Test | Synthesises just that one segment with the voice you picked and plays it back — no need to run the whole batch to hear whether the casting works |
| ▶ / ⏹ | Plays back audio that has already been generated |
| ↻ | Forces that one segment to be generated again |
Each row also shows its own status — pending, running, done or error — updating live as the batch runs.
🎧 When it's done, a timeline appears above the table with every voice track and the sound effects laid out against the subtitles — the same timeline you get from an imported audio package. ↻ Regenerate audio only re-synthesises the segments whose engine or voice actually changed.
6️⃣ Prompts
The last step writes an image prompt and a video prompt for every scene, using the scene's summary and segment text, the visual description of your confirmed characters, and the style. The table shows them side by side, one row per scene.
Prompts get the same treatment as the script: a Review button with a round count, plus an auto-review toggle in Setup. ↻ Regenerate prompts rewrites them all.
💡 Note: These prompts land in the normal scene table, so you can keep editing them anywhere else in AutoFlowCut — in the app, or through the MCP server from Claude Code.
⚡ Auto & Run All
Tick auto on the steps you trust, press ▶ Run all, and the pipeline carries itself to the end.
How it behaves
- The auto checkbox lives inside the chip itself, on the three steps that can run unattended: Scene split, Audio and Prompts. Setup and Script are not on the list — a script needs a human.
- Run all runs the auto-ticked steps in order and skips the rest. It needs a finished script, and it stops the moment a step errors.
- Audio is off by default. TTS costs money on every run, so you have to opt in on purpose.
- Confirm the cast in the Synopsis gate first: while it is unconfirmed, Run all stays disabled.
💡 A good rhythm: leave Scene split and Prompts on auto, keep Audio off until you have listened to a few test lines in the voice picker, then tick it and let Run all finish the job.
🎭 Characters & @mention
Characters do not stay inside the script. Once you confirm the cast in the Synopsis gate, they register themselves as reference cards in the References tab — the Prompt field you edited on each card is exactly the text the reference image is generated from.
Attaching a character to a scene
Type @ followed by the character's name inside a scene prompt, and it becomes a chip bound to that reference image:
@scholar sits alone in a lamplit study, rain against the window
A matched mention becomes a cyan chip; an unmatched one stays as plain text with a red wavy underline — so you always know whether the reference is really attached. This is why the character name matters: it is the handle.
📖 The full @mention spec — matching rules, chips, reference images — lives in the File Import Guide. Story mode simply gives you the cast for free.
🚀 Generate & Export
When the pipeline reaches the end, Story mode hands over a finished project: scenes, prompts, characters as references, and audio. From here on, nothing is Story-specific.
Generate images and video
Run the scenes through generation exactly as you would for an imported project — via Google Flow login or with your own Gemini / Veo API key. Review the output in the results panel: Timeline, Results or Grid.
Fix what needs fixing
Edit a prompt and regenerate a single scene, re-record one line of audio, or re-split the scenes. Each step stays reachable from its chip in the stepper.
Export to your editor
Export to CapCut, Adobe Premiere Pro (.prproj) or Vrew (.vrew). Your narration tracks and sound effects come along as separate audio tracks.
📖 Next: the Export to CapCut Guide covers where the project folder goes and how to open it.