> For the complete documentation index, see [llms.txt](https://learn.firecut.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://learn.firecut.ai/features/script-guidance.md). # Script guidance ### ❓ What is it? Script guidance turns a raw, unedited recording into a clean first cut that follows a written script. If you record talking-head or "self-tape" style videos β€” reading from a script, doing several takes, stumbling over lines, and chatting between takes β€” you usually end up with a long clip full of retries and off-script moments. Script guidance reads your finished script, finds the parts of your footage that actually match it (in order), and builds a **new timeline that contains only those parts**. You give FireCut two things: * **One source video clip** (that already contains audio) * **Your script**, as a plain-text `.txt` file …and FireCut produces a new sequence with the off-script takes, slip-ups and retries removed. > Script guidance is in beta and the transcription is experimental, so always review the result before relying on it. {% embed url="" %} ### πŸ€” How does it work? The feature works in four stages. Stages 1–2 are automatic; stage 3 is where you review and fine-tune; stage 4 builds the cut. #### 1️⃣ Transcribe your clip Pick a video clip from your project tree (bins expand and collapse; only a single clip can be selected, not a bin), choose the **Audio language**, and click **Transcribe**. * FireCut transcribes the **whole clip** using a transcription model (Scribe) chosen specifically because it handles raw footage well β€” repeated takes, false starts, and filler don't throw it off. * Transcripts are **saved and reused**. If you've already transcribed a clip, the button reads **Load transcript** and the transcript opens instantly instead of being re-generated. Clips that already have a transcript show a captions icon in the tree. This means you can iterate on your script and settings as much as you like without ever re-transcribing. * Maximum clip length is **30 minutes**. #### 2️⃣ Map your script to the transcript Provide your script in the **Script** panel (left). If it's empty, click **"Missing script β†’ Click here to open a script"** to import a `.txt` file (up to 30,000 characters). FireCut then matches your script against the transcript and **color-codes the matched sections** across both panels β€” Script on the left, Video transcript on the right. The matcher is deliberately tolerant so it survives real-world recordings: * **Small misspeaks and transcription errors** still match (fuzzy matching at the word and phrase level). * **Numbers and their words are treated as equal** β€” `2` matches "two", `21` matches "twenty-one". * **Common symbols match their spoken form** β€” "and" ↔ `&`, "percent" ↔ `%`, "dollars" ↔ `$`, "at" ↔ `@`, "plus" ↔ `+`, and so on. * **Off-script content is ignored.** Anything you said that isn't in the script β€” retries, "okay, let me do that again", ums β€” simply never gets claimed by a section, so it's left out of the final cut. Script lines you never actually said are skipped too. * **Repeated lines** resolve to one occurrence. FireCut scans top-to-bottom and keeps the first good take of a repeated line (and, with *Align to sentences* on, snaps it to clean sentence edges). #### 3️⃣ Review and edit the mapping This panel is the heart of the feature β€” see **Reviewing & editing the mapping** below. Matched sections are color-coded; hovering a section highlights and scrolls to its counterpart in the other panel. #### 4️⃣ Create the timeline Click **Create timeline**. FireCut: * Creates a **brand-new sequence** named **"FireCut Script - {your clip name}"** (in a *Script* sub-bin) built from your source clip. **Your original sequence is left completely untouched.** * Keeps only the matched sections and **removes everything else** β€” the unmatched parts are cut out and the gaps closed (a ripple delete, the same mechanism as silence removal), not just muted or disabled. * If your matched sections already cover the whole clip, it tells you *"Script timeline created (no cuts required)"* and still creates the sequence. The result is your original clip reduced to just the parts that correspond to your script, in order β€” a starting point you can refine further in Premiere. ### ✏️ Reviewing & editing the mapping Before you create the timeline, you have full manual control over what gets kept. Everything below lives on the results screen. * **Edit the script** β€” the Script panel is directly editable. As you type, FireCut re-maps automatically. Use **Clear** to wipe the script and start over. * **Adjust a selection** β€” hover over a colored section in the transcript and drag the chevron handles on its edges to grow or shrink what's included. * **Add a section** β€” drag-select across un-highlighted transcript text to **force-keep** a portion that wasn't matched automatically (useful for an ad-lib you want to keep). * **Delete a section** β€” right-click a section and choose **Delete section**. * **Re-run mapping** β€” recompute the automatic mapping from scratch. > **Heads up β€” manual edits "freeze" the mapping.** The moment you adjust, add, or delete a transcript selection, FireCut stops auto-remapping so it won't overwrite your work. The status changes from **"βœ“ Mapped"** to a yellow **"✎ Edited"**, and a **Re-run mapping** button appears. While frozen, editing the script text saves it but no longer remaps. Clicking **Re-run mapping** recomputes from the script and **discards your manual transcript edits** (you'll be asked to confirm: *"Re-running the mapping will discard your manual transcript edits. Continue?"*). ### βš™οΈ Settings #### Transcription * **Source clip** β€” the single video clip to transcribe. Expand bins to find it; only clips (not bins) can be selected. One clip at a time. * **Audio language** β€” set this to the language of your audio. This is an **experimental** setting and currently supports **left-to-right languages only** (right-to-left languages such as Arabic and Hebrew aren't available yet). #### Mapping controls Open these with the sliders icon next to **Mapping**. Changing any of them re-maps live (after a short pause), and each has a reset (β†Ί) button. * **Minimum match length** *(default 5; range 1–20)* β€” how many words in a row must agree before a chunk counts as a match. **Higher** β†’ fewer, more confident matches. **Lower** β†’ more (and shorter) matches, but more false positives. * **Fuzzy matching word factor** *(default 50%; range 0–75%)* β€” how different two single words can be while still counting as the same word. **Higher** β†’ more forgiving of mis-transcriptions. **Lower** β†’ stricter. * **Fuzzy matching phrase factor** *(default 20%; range 0–75%)* β€” when a single word fails to match, FireCut looks ahead and compares short phrases instead; this controls how different those phrases can be. **Higher** β†’ bridges bigger gaps across errors. * **Fuzzy matching phrase length** *(default 50; range 10–250)* β€” how many characters of look-ahead the phrase comparison uses. * **Align to sentences** *(default on)* β€” snaps section boundaries to sentence endings where possible. Especially helpful around repeated lines. #### Timeline controls Open these with the sliders icon next to **Timeline**. These control the *cut*, and are applied when you click **Create timeline** (they don't change the on-screen mapping). * **Section edge padding** *(default 0.6s; range 0.1–5s)* β€” extra time kept before and after each matched section so cuts have breathing room. **Higher** β†’ looser cuts with more around each section. **Lower** β†’ tighter cuts. * **Remove pauses longer than** *(default 1.0s; range 0.1–5s)* β€” trims long silent gaps *inside* a kept section. **Lower** β†’ tighter (removes shorter pauses too). **Higher** β†’ keeps longer in-section pauses. > Together, **Section edge padding** and **Remove pauses longer than** act as your "cut tightness" controls. ### πŸ’‘ Tips for best outcomes * **Use clean audio.** Mute background music and sound effects, and keep speech clear β€” the cleaner the transcript, the better the matches. * **Match your script to what you actually said.** The closer the script wording is to your spoken words, the more reliably it maps. If a take drifted from the script, either loosen the fuzzy settings or tweak the script/transcript text. * **Keep your script in spoken order.** FireCut scans top-to-bottom and matches the first good occurrence within range, so large re-orderings can confuse it. * **Iterate freely.** The transcript is saved after the first run, so adjusting your script and the mapping sliders never re-transcribes. Tweak, watch the highlights update, then create the timeline. * **Fine-tune by hand when needed.** Drag the section handles or force-keep a selection rather than fighting the sliders for one stubborn section. * **Treat the output as a first cut.** It's a fast starting point; do your final trims in Premiere. ### πŸ™‹β€β™‚οΈ Common questions, issues, watch-outs * **"A good take or line is missing from the cut"** β†’ Loosen the matching: lower **Minimum match length**, or raise **Fuzzy matching word factor** / **phrase factor**. Also check that your script wording matches what you actually said. As a last resort, drag-select that part of the transcript to **force-keep** it. * **"It kept the wrong take of a line I repeated"** β†’ Drag the section's handles to move the selection, or **Delete section** and force-keep the take you want. Keeping **Align to sentences** on usually helps repeated lines snap to clean boundaries. * **"The cuts feel too tight / words get clipped"** β†’ Raise **Section edge padding**. * **"Long pauses are left in the cut"** β†’ Lower **Remove pauses longer than**. * **"I edited the transcript and now editing the script does nothing"** β†’ Manual transcript edits **freeze** the mapping so they aren't overwritten. Click **Re-run mapping** to return to automatic mapping β€” note this discards your manual transcript edits. * **"My language isn't in the list"** β†’ Script guidance currently supports left-to-right languages only. Right-to-left languages (Arabic, Hebrew, etc.) aren't supported yet. * **"My clip is too long" / "I have multiple source clips"** β†’ Clips are limited to **30 minutes**, and the feature works on **one clip at a time** β€” multiple or joined source media aren't supported yet. * **"Will this change my original sequence?"** β†’ No. Create timeline builds a **separate new sequence** ("FireCut Script - …"). Your original is untouched. The unmatched parts in the new sequence are removed (ripple-deleted), not just disabled. * **"The transcript has mistakes"** β†’ Because audio is transcribed automatically (and non-English is experimental), some errors are expected β€” fuzzy matching is designed to tolerate them. If the transcript is badly off, make sure the audio is clear and free of long silences; if it persists, share your audio with us at .