# Auto Color
Source: https://docs.premierecopilot.com/features/auto-color
AI colorist grade of your timeline — exposure, white balance, contrast and a cinematic look, with shots of the same scene matched to each other.
Auto Color grades your timeline like a colorist: it looks at an actual frame from every source clip, measures it, corrects the exposure and white balance, sculpts the contrast, and applies a look on top — while keeping shots from the same scene consistent with each other so your cuts don't jump.
Works in **Premiere Pro** and **DaVinci Resolve**.
## Quick Start
Shooting Log? Apply your **Log → Rec.709 conversion** (LUT or color-space transform) **before** running. Auto Color grades on top of your base image.
Open the **Tools Menu** and select **Auto Color**.
Type your instruction in the **prompt box** — e.g. *"warm cinematic look, match all the clips"*. Optional, but this is where the creative direction lives.
Leave **Time Start** and **Time End** empty to grade the whole timeline.
Click **Start Auto Color**.
**Log / flat footage without a conversion will not grade well.** Auto Color reads the image it is given: if you hand it a washed-out Log frame, it corrects a washed-out Log frame. Put your LUT or color-space transform on first.
## What it actually does
Every source is fingerprinted, then shots that belong to the same scene, lighting setup or camera are grouped so they share one consistent look. Unrelated scenes are deliberately **not** matched — a daylight exterior and a tungsten interior stay separate.
Each source is then graded individually: correction first (exposure, white balance, contrast), creative look second — following its group's shared direction when it has one.
### What gets adjusted
The grade is a full **Basic Correction** pass, not a two-slider nudge:
| Control | What it's used for |
| ------------------------ | ------------------------------------------------------------------------------------------ |
| **Exposure** | Bring the image to a balanced brightness. |
| **Temperature / Tint** | Neutralise the measured color cast — or push it for the look. |
| **Contrast** | Add shape, especially on flat footage. |
| **Blacks / Whites** | Set the black and white points — this is what makes an image feel filmic instead of milky. |
| **Shadows / Highlights** | Open crushed shadows, recover blown highlights. |
| **Saturation** | Punch or mute, with skin tones protected. |
**One frame per source is analysed**, not one per clip. If the same source appears eight times in your edit, it's analysed once and the grade is applied to all eight instances — consistently, and for the cost of one.
### How it lands on your timeline
| Host | Where the grade goes |
| ------------------- | ------------------------------------------------------------------------------------------------------------------------------------ |
| **Premiere Pro** | On **adjustment layers** placed on an empty video track above your clips — fully non-destructive. Your original clips are untouched. |
| **DaVinci Resolve** | Applied per clip on the first node. |
In Premiere, hide or delete the adjustment-layer track to see the before/after instantly — and to throw the whole grade away in one move.
## Settings
| Setting | What it does |
| ------------------ | ----------------------------------------------------- |
| **Prompt box** | Free-text creative direction. Overrides the defaults. |
| **Time Start (s)** | Only grade clips at or after this time. |
| **Time End (s)** | Only grade clips up to this time. |
Unlike Auto Mix, the time range here scopes **both the analysis and the grade** — only the clips in the window are looked at. Grading a section on its own means it won't be matched against shots outside the window, so expect it to sit slightly differently from the rest.
## Writing a Good Prompt
Describe the **image you want to see**. Reference a mood, a film, a palette — not slider values.
### What works
* **A look:** *"Warm cinematic look, teal shadows, film-like contrast."* / *"Clean and neutral, corporate."* / *"Cold, desaturated, moody."*
* **A reference:** *"Nordic thriller palette."* / *"Sunny golden-hour vlog feel."* / *"Documentary — natural, no stylisation."*
* **Matching intent:** *"Make everything match, it's all one interview."* / *"Keep each location distinct, don't match across scenes."*
* **A known problem:** *"My footage is too warm under this lighting, neutralise it."* / *"The interior shots are underexposed."*
* **Restraint:** *"Correction only, no creative look."* / *"Subtle — don't push the saturation."*
* **Skin priority:** *"Talking head — skin tones are the priority, everything else is secondary."*
### What doesn't work
* **Slider values.** *"Set exposure to +0.4"* — say the image is dark instead.
* **Per-clip instructions.** *"Make the third shot cooler"* — grade the timeline, then adjust that clip by hand. It's already an editable Lumetri instance.
* **Timecodes in the prompt.** Use the Time Start / Time End fields.
* **Selective / secondary corrections.** No masks, no power windows, no "only change the sky" or "isolate the red jacket". This is a primary grade.
* **Split-toning by name.** *"Teal shadows, orange highlights"* is understood as a direction, but true wheel-based split-toning isn't part of what gets applied — expect the spirit, not the exact technique.
* **Fixing the shot.** Clipped highlights that hold no data, heavy noise, or a badly out-of-focus frame can't be recovered by a grade.
* **A LUT by name.** *"Apply Kodak 2383"* won't load a LUT. Describe the look and it will be approximated with the correction controls.
### Recommended pattern
> *"\[The look you want]. \[Matching intent]. \[Any known problem]. \[How far to push it]."*
**Example:**
> *"Warm, filmic look with deep blacks — golden-hour feel. All the interview shots are the same setup, match them. The B-roll outside is a bit cold, neutralise it. Keep it subtle, skin must stay natural."*
If the first pass overshoots, add one restraint sentence (*"much more subtle, correction-led"*) and re-run, rather than rewriting the whole brief.
## After the run
The grade is standard, editable color correction — nothing is baked in. You can retune any shot, copy a grade from one clip to another, or delete it entirely.
Review on your cuts, not clip by clip. A grade that looks great in isolation but jumps at a cut is the real failure mode — and it usually means two shots that should have been matched weren't. Say so in the prompt and re-run.
## Common Pitfalls
* **Log footage with no conversion applied** → the most common cause of a disappointing result. Convert first.
* **An already-graded timeline** → Auto Color adds its own correction on top of yours. Remove your existing grade first, or the two stack.
* **Nested sequences** → the clips inside can't be analysed as separate sources. Flatten first.
* **Expecting scene-to-scene uniformity** → grouping is deliberately conservative. Different lighting setups stay different unless you explicitly ask to match everything.
* **A very large timeline** costs proportionally more: the price scales with the number of **unique sources**, not the number of clips or the duration.
* **Frames that can't be exported** (footage offline, media on a blocked or synced folder) are skipped — check your media is online before running.
**Recommended workflow:** convert Log → one full pass with a short look-and-matching prompt → review at the cuts → re-run with a restraint or matching correction if needed → hand-tune the two or three shots that still stand out.
# Auto Mix
Source: https://docs.premierecopilot.com/features/auto-mix
AI sound-engineer mix of your whole timeline — levels, EQ, compression, de-noise and music ducking under the voice.
Auto Mix listens to every audio clip in your sequence and mixes it like a sound engineer would: it sets each source to a proper loudness target, cleans up the voice, leaves the music alone but puts it at the right level, and rides the music down under every line of dialogue.
Everything is applied as **real, editable effects** on your clips — nothing is baked in.
Auto Mix is a **Premiere Pro** feature. It is not available in DaVinci Resolve.
## Quick Start
Open the **Tools Menu** and select **Auto Mix**.
Type your instruction in the **prompt box** at the bottom of the panel — e.g. *"clean podcast mix, keep the music well under the dialogue"*. Optional, but this is where most of the control lives.
Leave **Time Start** and **Time End** empty to mix the whole timeline.
Click **Start Auto Mix**. Count roughly **one minute per hour of timeline**.
## What it actually does
Auto Mix works in two passes, and knowing the difference explains most of the results.
Each source clip is identified as **speech**, **music** or **SFX**, then given its own treatment: level, high-pass, compression, de-noise, de-ess, limiter — only what that clip actually needs.
Wherever a music or SFX bed overlaps speech, a smooth **volume automation** is written on the bed: it dips before the line starts, holds under it, and comes back up after.
### Level targets
Levels are set relative to a delivery standard, with the **voice as the anchor** and everything else placed below it:
| Delivery | Speech anchor |
| -------------------------------- | ------------- |
| **Podcast / voice** (default) | -16 LUFS |
| **YouTube / streaming / social** | -14 LUFS |
| **Broadcast / TV / EBU R128** | -23 LUFS |
Music sits clearly below the voice and is ducked further under it; SFX land in between.
Just name your platform in the prompt — *"master for YouTube"*, *"broadcast delivery"* — and the whole mix re-anchors around it. That single word does more than any manual gain pass.
### Processing philosophy
* **Voice gets the work.** High-pass to remove rumble, compression only when the delivery is genuinely uneven, de-noise only when there's a real noise floor, de-essing only when it's actually sibilant.
* **Music is left alone.** A music bed is already produced and mastered, so it gets a **level and nothing else** — no EQ, no compression. Only its volume moves.
* **SFX stay minimal.** Level, and a limiter if a peak demands it.
Nothing is applied "just in case". If your voice track is clean and even, you'll get a gain change and a high-pass, and that's it — that's the correct mix, not a missing one.
## Settings
| Setting | What it does |
| ------------------ | -------------------------------------------------- |
| **Prompt box** | Free-text instruction. Overrides the defaults. |
| **Time Start (s)** | Apply the mix only to clips at or after this time. |
| **Time End (s)** | Apply the mix only to clips up to this time. |
**The time range limits where the mix is applied, not what is analysed.** The analysis always reads the whole timeline, so a range still gets levels consistent with the rest of your sequence. This is deliberate — mixing a 30-second window in isolation would make it jump against its neighbours.
Use the range to fix one problematic section after a first full pass, rather than re-mixing everything.
## Writing a Good Prompt
The prompt is optional, but it's how you set intent. Describe the **result you want to hear**, not the parameters.
### What works
* **The delivery target:** *"Master for YouTube."* / *"Broadcast delivery, EBU R128."*
* **The music/voice balance:** *"Keep the music very low under the dialogue, it should barely be there."* / *"The music can stay present between the lines."*
* **The overall character:** *"Clean, neutral podcast mix."* / *"Punchy and loud, energetic edit."* / *"Natural and dynamic, don't squash it."*
* **A known problem with the recording:** *"There's air conditioning in the background, clean it up."* / *"I'm sibilant, tame the S sounds."* / *"My delivery goes from whisper to shout, even it out."*
* **What to protect:** *"Don't touch the music, only fix the voice."* / *"Don't over-compress, keep the dynamics."*
* **Ducking behaviour:** *"Duck the music hard under every line."* / *"Only light ducking, this is a music-led video."*
### What doesn't work
* **Per-clip instructions.** *"Lower the third clip by 3 dB"* is a local retouch — do it by hand, or ask the Copilot for that single clip. Auto Mix is a whole-timeline pass.
* **Timecodes.** *"Boost the intro at 0:15"* — use the Time Start / Time End fields instead.
* **Creative sound design.** No reverb, delay, pitch, telephone effect, or vocal character processing. This is a mix, not a design pass.
* **Fixing the recording itself.** De-noise reduces a noise floor; it doesn't remove a passing motorbike, a clipped peak or a room with bad acoustics.
* **Renaming or re-classifying a clip.** If a clip is misread as music, say so in words (*"the file called 'ambience' is dialogue"*) rather than expecting a setting.
* **Asking for a specific dB number per effect.** Say what you want to hear; the target does the rest.
### Recommended pattern
> *"\[Delivery target]. \[Voice/music balance]. \[Known problem to fix]. \[What to leave alone]."*
**Example:**
> *"Podcast mix for Spotify. Music clearly under the voice, ducked hard on every line. There's a constant hum in my mic — clean it. Don't over-compress, I want the delivery to breathe."*
If the first pass isn't right, change **one** thing in the prompt and re-run rather than rewriting the whole instruction — it's much easier to tell what moved.
## After the run
The mix arrives as standard Premiere audio effects on your clips — volume, high-pass, compressor, de-noise, de-esser, limiter — plus volume keyframes on the ducked beds.
That means you can:
* **Adjust anything by hand** in the Effect Controls panel.
* **Reshape a duck** by dragging its keyframes.
* **Remove a treatment** you disagree with, clip by clip.
* **Undo the whole thing** with Cmd/Ctrl+Z.
Listen on the monitoring you actually deliver on. A mix anchored at -16 LUFS is correct even if it feels quiet next to a loud reference track in another app.
## Common Pitfalls
* **All audio on one track** → clips still get treated individually, but a voice and a music bed sharing a track make classification harder. One source type per track gives the best result.
* **A music bed that never ducks** → it probably wasn't read as music. Say what it is in the prompt.
* **Nested sequences or multicam** → the audio inside them can't be analysed as separate sources. Flatten first.
* **Already-processed clips** → Auto Mix adds its own effects on top of yours. Remove your previous audio effects before running, or the two stack.
* **Expecting a repair tool** → de-noise handles a steady noise floor. Clipping, wind blasts and room reverb are recording problems it cannot undo.
* **Touching the timeline mid-run** → let it finish; it's reading and writing your sequence.
**Recommended workflow:** one source type per audio track → a two-sentence prompt naming your platform and the music/voice balance → full-timeline pass → listen → fix any single section with a Time Start / Time End re-run.
# AutoZoom
Source: https://docs.premierecopilot.com/features/auto-zoom
Automatically generate punch-ins and dynamic zooms on your timeline.
AutoZoom analyzes your sequence and places a series of zooms (punch-ins / outs) on your timeline automatically. It combines video cuts, speech, emotional peaks and AI context analysis to decide *where* and *how* to zoom.
## Quick Start
Open the **Tools Menu** (bottom left) and select **AutoZoom**.
Position the playhead anywhere inside the sequence you want to process.
Activate one or more **Triggers** (Cuts, Emotion, Speech, Random, Context) — see below.
Set the sliders (**Motion Camera**, **Zoom Rhythm**, **Zoom Fastness**, **Zoom %**) to match the feel you want.
Click **Start** — a backup of your sequence is created automatically before zooms are applied.
![AutoZoom interface]()
## Triggers
Each active trigger gets an equal share of the total number of zooms. Activating more triggers spreads zooms across more types of moments.
Places a zoom on every video cut in your sequence. **Free** — no transcription needed.
AI detects emotional peaks in speech (laughs, surprise, conviction). Requires transcription.
Places a zoom at the start of each spoken segment. Requires transcription.
Fills the longest gaps between zooms. **Free** — useful to keep rhythm steady.
AI reads the transcript and zooms on punchlines, topic changes, or anything you describe in the **text prompt**. Requires transcription.
Activating **Emotion**, **Speech** or **Context** triggers a **single shared transcription** — you are not charged multiple times.
## Intensity Sliders
| Slider | Range | What it does |
| ----------------- | ------- | --------------------------------------------------------------------- |
| **Motion Camera** | 0–4 | How much the camera moves during a zoom. At 0, zooms are static. |
| **Zoom Rhythm** | 1–5 | Density of zooms per minute (10/min at 1 → 30/min at 5). |
| **Zoom Fastness** | 1–5 | Duration of the zoom animation. Higher = snappier. |
| **Zoom (%)** | 100–300 | Maximum zoom amount. Each individual zoom uses 30–100% of this value. |
## Transition Style
Pick one or more transition types in **Box Settings**. At least one must stay active.
| Type | Feel |
| --------------- | --------------------------------------- |
| **Jump Cut** | Hard, instant snap to the zoomed frame. |
| **Ease In Out** | Smooth animated transition. |
| **Snap In Out** | Short, aggressive snap. |
You can enable **Sound In / Out** for each type to add a SFX (woosh / reverse) on every zoom.
## Interactive Preview
Click the refresh button to export the current Premiere frame as a preview.
**Drag** anywhere on the image to define the **center** of the zoom (where the camera will land).
**Drag the corners** of the green box to set how much the zoom crops in. The green box represents the area that will fill the screen after zoom.
## Writing a Good Text Prompt
The text prompt at the bottom of the extension is **only used if Context (AI) is active**. Otherwise it is ignored.
### What works
* **Target a type of information:** *"Zoom on every mention of numbers, statistics or results."*
* **Target a theme:** *"Zoom when I talk about personal experience or tell an anecdote."*
* **Target a subject:** *"Zoom whenever I mention a brand name or a product."*
* **Reinforce editorial beats:** *"Focus on punchlines and impactful sentence endings."*
* **Filter by energy:** *"Avoid calm moments — only zoom when I raise my voice."*
### What doesn't work
* Asking to change **intensity** or **speed** — these are controlled by sliders only.
* Asking for a specific **transition type** (jump / ease / snap) — chosen automatically.
* Asking for **spatial coordinates** — driven by the interactive preview only.
* Giving a **manual list of timecodes** — the AI picks moments freely.
### Recommended pattern
> *"Zoom on \[type of editorial moment] when \[narrative condition]. Avoid \[what you don't want]."*
**Example:**
> *"Zoom every time I announce a statistic or give an actionable tip. Avoid transitions and descriptive passages."*
## Common Pitfalls
* **No trigger active** → only cuts will be processed, which may produce zero zooms on a single-shot video.
* **No transition type active** → AutoZoom returns nothing. Keep at least one box type checked.
* **Transcription fails** → Emotion / Speech / Context are silently skipped. Only Cuts and Random will run.
For talking-head / podcast footage, start with **Speech + Context + Cuts**, Zoom Rhythm at 2–3, and a short text prompt describing the editorial intent.
# Claude Cut
Source: https://docs.premierecopilot.com/features/claude-cut
AI rough cut for talking-head footage: removes bad takes, repetitions and dead air, and lets you review every cut before applying.
Claude Cut turns a raw talking-head take into a clean rough cut. It transcribes your sequence, then an AI reads the whole transcript and decides which segments belong in the final video — dropping the mic checks, the retakes, the "wait, let me say that again", and the dead air between words.
You review everything as **editable text** before a single frame is touched. Nothing is applied until you click **Apply Cut**.
Works in **Premiere Pro** and **DaVinci Resolve**.
Claude Cut is built for **single-speaker footage**: YouTube monologues, founder explainers, video essays, course modules, interview answers. It is not a multicam or dialogue-scene tool.
## Quick Start
Open **Claude Cut** from the Tools Menu. Your sequence must be open, with at least one audio track carrying the voice.
Choose the AI model in the **Model** box (top left). Sonnet 5 is the default.
The green **Analyze** button runs the whole pipeline on its own: silence detection → transcription → AI analysis. Leave the app alone while the progress modal is up.
Segments the AI wants to remove are flagged **CUT** in red; the rest stay **KEEP** in green. Read them, flip whatever you disagree with.
Click **Apply Cut**. The cut clips are removed and the gaps closed, in one click.
While the progress modal is open, **keep Premiere Pro / DaVinci Resolve in the foreground and don't touch the timeline**. The pipeline is driving your sequence.
## Step 1 — Analyze
### Model
| Model | Best for |
| ------------------- | ---------------------------------------------------------------------------------------------------------- |
| **Sonnet 5** | The balanced default. Fast, reliable, good judgement on most footage. **Start here.** |
| **Claude Opus 4.8** | The most careful reader. Use it on dense, technical or high-stakes content where a wrong cut is expensive. |
| **Gemini 3.1 Pro** | Long-form. Useful on very long takes where the same idea comes back 20 minutes later. |
If a run over-cuts or misses obvious repetitions, **change the model and re-run** before hand-fixing dozens of segments. Different models make different judgement calls on the same take.
### Transcription precision
The small chevron button to the left of **Analyze** opens the engine menu:
Fast, reliable transcription. Included — no extra credits. The right choice in the vast majority of cases.
Re-syncs **every word** to the audio for sharper cut points. Slightly slower and costs extra credits (shown on the button).
Turn on **advanced word precision** when the cuts land a fraction of a second early or late — typically on fast speech, strong accents, or heavily overlapping delivery. On a calm, well-recorded take, standard is indistinguishable and cheaper.
### What the pipeline actually does
The single **Analyze** button chains three steps, shown live in the progress modal:
| Step | What happens |
| ---------------------- | ----------------------------------------------------------------------------------------- |
| **1. Detect silences** | Finds the dead air between words. Nothing is applied yet — the ranges are kept in memory. |
| **2. Transcribe** | Exports and transcribes your sequence audio. |
| **3. AI analysis** | The model reads the full transcript and marks what to remove. |
Both the silences and the AI's cuts are applied together, later, when you click **Apply Cut** — so the transcript you review always matches the timeline you're looking at.
The whole run is billed as **one combined charge**, and only if it completes. A failed or aborted run doesn't cost you.
## What the AI removes — and what it never touches
Understanding this is what makes the review step fast.
### It removes exactly two things
Mic checks, "is it rolling?", framing chatter, self-direction spoken out loud, asides to whoever is behind the camera, post-take reactions ("was that good?"). Almost always clustered at the **very start** and the **very end** of the take.
Retakes of the same line, verbatim repeats, and reformulation chains where the speaker searches for the right phrasing. **One version survives** — usually the most complete one — and the others go.
### It never does these
* **Rewrite, rephrase or reorder anything.** Your words stay in your order. The only lever is which segments survive.
* **Tighten for style.** A segment that is merely slow, long-winded or "could be punchier" is kept. That's an editing decision, not a cleanup one.
* **Cut a deliberate repetition.** Rhetorical triples ("It's not about speed. It's not about price. It's about trust.") and emphasis ("very, very fast") stay.
* **Drop new information**, even when it's clumsily phrased.
* **Break a seam.** Connector segments ("and that's why…", "but the thing is…") are kept when removing them would make the following line dangle.
The AI is deliberately **conservative**: keeping a slightly weak segment costs you almost nothing, while dropping a load-bearing one breaks the video. Expect it to leave in things you'd personally cut — that's the intended bias, and the review pass is where you tighten further.
## Step 2 — Review
This is not an optional step. Treat the AI output as a strong first pass, never as a final cut.
### Editing the transcript
| Action | Effect |
| ------------------------------- | ---------------------------------------------------------------------------------------- |
| **Click the KEEP / CUT chip** | Flips the whole segment. |
| **Click a word** | Toggles that single word — and moves the playhead there, so you can hear the exact spot. |
| **Double-click a word** | Toggles the whole clause around it (up to the nearest punctuation). |
| **Triple-click** | Toggles the entire segment. |
| **Click and drag across words** | Toggles the whole selection at once. |
| **Click a timecode** | Jumps the playhead to that segment. |
| **Reset button (top right)** | Brings everything back to KEEP. |
The **Timeline Metrics** box tracks the damage live: **Del.** is how much you're removing, **New Dur.** is what the sequence will last after applying.
Review in this order: (1) scan the **red** blocks at the start and end of the take — that's where pollution clusters and the AI is almost always right; (2) read the **green** lines immediately before and after each red block, out loud if you can — that's where a bad cut shows up; (3) only then hunt for extra fat to trim yourself.
## Step 3 — Apply Cut
**Apply Cut** disables the cut segments and immediately removes the disabled clips, closing the gaps — one click, clean timeline.
Listen back after applying. If a seam jumps, **undo** (Cmd/Ctrl+Z), flip that segment back to KEEP, and apply again. Don't re-run the whole analysis for a single bad seam.
## Steering the AI with a prompt
The **Analyze** button in the panel runs the default cleanup — there is no prompt box in the panel. To steer the AI, ask the **Copilot** in the chat instead:
> *"Clean up my sequence — remove the bad takes and repetitions, but keep every personal anecdote even if it's long."*
The Copilot runs the same engine, plus your instruction.
### What a steering prompt can do
* **Protect a category of content:** *"Keep every emotional anecdote and personal story, even the rambling ones."*
* **Protect a topic:** *"Never cut anything about pricing or the guarantee — I need those beats intact."*
* **Set the aggressiveness:** *"Be aggressive on the intro and the outro, conservative in the middle."*
* **Choose between duplicates:** *"When I say something twice, keep the first, more spontaneous version."* (The default keeps the most complete one.)
* **Handle a known habit:** *"I restart my sentences a lot — drop the false starts."*
* **Flag your workflow:** *"I always do three takes of each line back to back — keep the last one."*
### What a steering prompt cannot do
* **Rewrite or reorder.** No rephrasing, no moving a paragraph earlier, no adding a transition word to fix a seam.
* **Hit a target duration.** *"Cut it down to 5 minutes"* will not work — the AI removes what's removable, not a quota.
* **Cut by timecode.** *"Cut everything after 12:30"* is a timeline operation, not a cleanup one. Do it yourself, or ask the Copilot for a plain cut.
* **Judge delivery.** It reads the transcript, not the performance. It can't know which take had better energy, only which is more complete.
* **Cut for topic.** *"Remove the part about the competitor"* is an editorial decision outside its two rules — do it in the review pass, it's two clicks.
### Recommended pattern
> *"\[How aggressive]. Protect \[content you refuse to lose]. \[Any habit of yours worth knowing]."*
**Example:**
> *"Standard cleanup, slightly aggressive at the start. Protect anything where I give a number or a concrete example. I restart sentences often — keep the last attempt."*
Keep the prompt to **two or three sentences of intent**. Long, rule-heavy prompts don't produce a better cut — they mostly make the AI hesitant, and a hesitant pass keeps everything.
## Common Pitfalls
* **Several audio tracks with clips** → from the Copilot, the run stops and asks you to confirm first. Claude Cut is designed for a single voice track; on a multi-track timeline, check the result carefully.
* **Music or B-roll audio under the voice** → transcription quality drops and so does the cut. Mute those tracks before running.
* **Applying without reading** → the fastest way to a broken video. The AI over-cuts casual, vlog-style delivery and under-cuts subtle reformulations.
* **Very long takes** are processed in batches automatically: the analysis takes longer, the result is the same.
* **Nothing got cut** → that's a valid answer. On a clean, scripted take there may be nothing to remove.
* **Re-running Analyze** starts from scratch and discards your manual KEEP/CUT edits. Finish reviewing before you re-run.
**Recommended workflow:** Sonnet 5 + standard precision → Analyze → review start and end first → Apply Cut → listen back → fix seams with Cmd+Z and a toggle. Move to Opus 4.8 or advanced word precision only when the standard pass disappoints you on that specific footage.
# Jump Cut Tool
Source: https://docs.premierecopilot.com/features/jump-cut
Automatically remove silence from your footage.
Jump Cut removes silence from your footage automatically. It uses **voice detection** — not a simple volume threshold — so it keeps whispers, breaths at the end of a sentence and quiet delivery, while cutting real dead air. It is approximately **10x faster** than competitors like AutoPod.
Works in **Premiere Pro** and **DaVinci Resolve**.
## Quick Start
The panel is organised in three numbered steps: **Settings → Preview → Apply to timeline**.
Open the **Tools Menu** (bottom left) and select **Jump Cut**.
Put the playhead over a representative part of your sequence — one containing both speech and silence.
Click **Preview**. A 15-second sample is analysed and drawn as a waveform: everything shaded **red** will be removed.
Move the sliders and watch the red zones update **instantly** — no re-analysis needed. Press play to listen to the sample.
Click **Create Jump Cuts**. Your sequence is backed up automatically before anything is cut.
![Jump Cut Interface]()
## Step 1 — Settings
| Setting | Range | Default | What it does |
| ------------------------ | ----------- | ---------- | ----------------------------------------------------------------------- |
| **Voice Sensitivity** | 0.10 – 0.90 | **0.50** | How confident the detector must be that it hears a voice. |
| **Remove Silences Over** | 0 – 2 s | **0.20 s** | Silences shorter than this are left alone. |
| **Keep Segments Over** | 0 – 2 s | **0.20 s** | Speech chunks shorter than this are treated as noise and removed. |
| **Padding** | -1 – 2 s | **0.10 s** | Margin added before and after every kept segment. |
| **Split L/R Padding** | toggle | off | Splits Padding into independent **Padding Left** and **Padding Right**. |
**Voice Sensitivity replaces the old "Silence Cutoff" (dB) setting.** It is no longer a volume threshold: the tool detects *voice*, so background noise, room tone or music no longer count as "not silent". Higher value = stricter = cuts more.
### Voice Sensitivity
* **0.10 – 0.30** — very permissive. Keeps whispers, soft-spoken delivery, ASMR, quiet interviews. Use when the tool is eating the beginning of your words.
* **0.40 – 0.60** — balanced. Standard talking-head, podcast, tutorial. **Start here.**
* **0.70 – 0.90** — strict. Cuts aggressively; useful in noisy rooms, with air conditioning, street noise or a loud music bed where the detector is too generous.
If the cut sounds *clipped* — the first syllable is missing — lower the sensitivity before touching padding. If it sounds *sloppy*, with breaths and hesitations surviving, raise it.
### Remove Silences Over
The minimum length a silence must reach before it's worth cutting.
* **0.10 – 0.20 s** — very tight, machine-gun pacing. Great for social/vertical edits.
* **0.30 – 0.60 s** — natural rhythm. Removes the dead air but keeps the breathing room of speech.
* **1 s and above** — only removes the long pauses. Good for interviews where the silences carry meaning.
### Keep Segments Over
Anything shorter than this is not considered real speech: a lip smack, a click, a door, a cough between two sentences. Raise it if isolated noises are kept as tiny orphan clips; keep it low if you have short interjections ("yes", "no", "hm") that must survive.
### Padding
Padding gives each kept segment a bit of air, so cuts don't feel surgical.
* **Positive values (0.05 – 0.30 s)** — the usual case. Extends each segment on both sides.
* **0** — hard, immediate cuts. Maximum tightness, harsher listening.
* **Negative values (down to -1 s)** — shrinks the kept segments, cutting *into* the speech. Only for very aggressive stylistic edits — it will clip words.
Turn on **Split L/R Padding** to get the most natural result: a **small left padding** (0.05 s, so the attack of the word isn't chopped) and a **larger right padding** (0.15–0.25 s, so the end of the sentence can breathe).
### Presets
The preset bar at the top of Step 1 saves your whole slider configuration under a name and reloads it in one click. Build one per format — *Podcast*, *Reels*, *Interview* — instead of re-tuning every session.
## Step 2 — Preview
The preview analyses a **15-second sample** taken at the playhead and is **free** — it doesn't consume credits, so you can run it as often as you like.
Blue = kept. **Red overlay = removed.** The picture reflects your current sliders in real time.
Hit the green play button, or click anywhere on the waveform to jump there and audition the passage.
Moving a slider after a preview **does not** require a new analysis — the cuts are recomputed locally and instantly. Only click **Preview** again if you move the playhead to a different part of the sequence.
Preview the hardest part of your footage, not the easiest: the passage where you hesitate, where the room is noisy, or where you speak softly. Settings that survive there will work everywhere.
## Step 3 — Apply to timeline
Analyses the **whole sequence** and applies the cuts. A **backup of your sequence is created automatically** beforehand, so you can always go back.
Jump Cut disables the silent clips instead of deleting them, so you can review the result. Click **Delete Disabled Clips** to remove them and close the gaps.
Tick the checkbox on that same button to **auto-delete** at the end of every run.
**On DaVinci Resolve**, the *Delete Disabled Clips* control is hidden — and that's expected. Resolve rebuilds a new timeline containing only the kept segments, so there is nothing left to delete.
## From the Copilot
You can also just ask in the chat:
> *"Remove the silences from my sequence."*
The Copilot runs the same engine with sensible defaults. Use the panel when you want to see the waveform and dial the pacing in yourself.
## Common Pitfalls
* **Playhead over a silent or music-only area** → the preview sample contains no speech and looks entirely red. Move it over real dialogue.
* **Words getting clipped** → lower **Voice Sensitivity**, then raise **Padding Left**. In that order.
* **Breaths and hesitations surviving** → raise **Voice Sensitivity** and lower **Remove Silences Over**.
* **Tiny orphan clips left over** → raise **Keep Segments Over**.
* **Loud music bed under the voice** → voice detection struggles. Mute or lower the music track before running, or push the sensitivity up.
* **Negative padding** eats into speech by design — it is not a "tighter" setting, it's a destructive one.
**Recommended starting point** for a talking-head edit: Voice Sensitivity **0.5**, Remove Silences Over **0.3 s**, Keep Segments Over **0.2 s**, Split L/R Padding **on** with **0.05 s** left and **0.20 s** right.
# Multicam Podcast Editing
Source: https://docs.premierecopilot.com/features/multicam-editing
Automatically edit multicam podcasts with AI.
Access this feature by clicking the **Microphone icon** (bottom left) in the extension.
## Timeline Setup (Crucial)
**Timeline Rules:**
* **Audio Tracks:** You must have the **exact same number of audio tracks as speakers**. (e.g., 2 Speakers = 2 Audio Tracks). Deviating from this will cause bugs.
* **Sync:** Ensure all audio and video tracks are synchronized before starting.
![Timeline Setup Placeholder]()
## Configuration
Match this to the number of video tracks.
* Name your speakers (e.g., Click "Speaker 1" to rename).
* Assign speakers to specific tracks (Default: Speaker 1 on Track 1).
* Define speaker locations on tracks.
* **Frequency Control:** Set how often a track appears (High/Low frequency).
* **Wide Shots:** The algorithm balances wide shots based on their assigned frequency. If multiple wide shots exist, it weighs them according to your settings.
## Advanced Settings
* **Delay Cut:** Shifts all cuts by X milliseconds/frames.
* **Ignore Cut Less Than:** Ignores cuts that result in clips that are too short.
* **Cutting Method:**
* *Disable:* Deactivates unused clips.
* *Cut:* Removes unused clips (Classic method).
* *Audio:* Marks cuts on audio tracks as well.
## Execution
Click **"Start Editing"**. Initialization may take \~10 seconds. A backup of your sequence is automatically created.
# Vibe Kinetic
Source: https://docs.premierecopilot.com/features/vibe-kinetic
AI kinetic text: animated captions and hooks rendered from your own words, placed on your timeline.
Vibe Kinetic listens to your sequence, picks the sentences worth putting on screen, art-directs them (typography, colour, animation, word-level highlights) and renders each one as an **animated clip with a transparent background**, placed automatically on your timeline.
It is not a subtitle tool. It writes **short on-screen hooks** — the punchline, the number, the payoff — the way a motion designer would, not a word-for-word transcript.
Works in **Premiere Pro** and **DaVinci Resolve**.
## Quick Start
Open the **Tools Menu** and select **Vibe Kinetic**.
Click a card in **Style presets** — it fills the **Brief** box with a ready-made style description you can then edit.
Describe the look and the editorial intent in plain language. Leave it empty for the clean default look.
**Texts per minute** = how many captions appear. **Seconds on screen** = how long each one stays.
Click **Generate captions**. The run modal shows four steps: audio export, curation, render, placement.
You need a **saved project** and an **active sequence with audio**. The tool reads the whole sequence, not just the part around the playhead.
## What you get on the timeline
Each caption arrives as its own **video clip with alpha**, stacked above your footage and cut to the exact in/out point the AI chose.
These are rendered clips, **not** Essential Graphics / Text layers. You can move, trim, delete or re-order them freely, but you **cannot retype the text** inside Premiere or Resolve. To change wording or style, adjust the brief and run again.
Rendered files are written to the **download folder** before import, so you keep a copy of every clip.
## Main Settings
| Setting | Range | What it does |
| --------------------- | --------- | -------------------------------------------------------------------------------- |
| **Brief** | free text | Describes the look *and* the editorial intent. The single most powerful control. |
| **Style presets** | 6 cards | Click to load a proven brief into the box. Fully editable afterwards. |
| **Texts per minute** | 2–15 | How dense the captions are. Default **8**. |
| **Seconds on screen** | 2–10 | Target display time per caption. Default **2**. |
| **Download folder** | folder | Where the rendered clips are stored. Defaults to your project folder. |
### Texts per minute
This is a **target density**, not a quota. The AI reasons in *ideas*, so it will only place a caption where there is something worth highlighting — a quiet or repetitive passage will get fewer.
Sparse, editorial. Long-form, interviews, documentary, corporate.
Balanced. YouTube talking-head, tutorials, podcasts clips. **Start here.**
Dense, high-energy. Reels, TikTok, Shorts, fast-cut promos.
Above 15 a caption would live less than four seconds on screen and nobody can read it — that's why the slider stops there.
### Seconds on screen
This is the **target** duration. A caption is automatically shortened when the next one arrives sooner, and captions that would end up too short to be readable are dropped rather than flashed.
* **2–3 s** — punchy, rhythmic, matches fast dense edits. Best default.
* **4–6 s** — the viewer has time to read a longer line. Good for stats, quotes, definitions.
* **7–10 s** — near-permanent lower third. Only useful at very low density, otherwise captions collide and get trimmed anyway.
Density and duration interact. High density + long duration = the duration wins on paper but gets clipped in practice. If you want long-lived captions, lower the density.
## Advanced Settings
Advanced is **off by default**, and that is a real mode — not just hidden fields.
The AI art-directs everything: it derives the palette and the mood from what is actually being said, varies placement per caption, and uses a font auto-detected on your machine.
Your palette, your placement and your font pool are respected. The AI still composes hierarchy, highlights and animation — but inside your constraints.
### Color palette
Four slots, plus five ready-made palettes.
| Slot | Typical role |
| ------------- | -------------------------------------------------------- |
| **Primary** | The body of the text — the colour most words will be. |
| **Secondary** | The accent: highlighted keywords, the payoff word. |
| **Third** | Secondary accent / a second line in a pairing. |
| **Fourth** | Support tone — muted lead-ins, decorations, panel tints. |
Fill **Primary + Secondary** and leave the rest empty if you only care about "white text, one brand-coloured keyword". Filling all four does not make the result busier — it just gives the art director more room.
Legibility is protected automatically: a colour that would vanish on its own background gets corrected. If your exact brand hex comes back slightly adjusted, that is why — pair it with a panel/box in the brief to keep it untouched.
### Placement
| Value | Behaviour | Use it when |
| ---------- | ---------------------------------------------- | ------------------------------------------------------------- |
| **Auto** | The AI varies the position caption by caption. | You want the composition to breathe and follow the frame. |
| **Top** | Every caption locked to the upper zone. | Subject's face or a lower-third graphic occupies the bottom. |
| **Center** | Every caption locked to the middle. | Full-screen statement style, B-roll montage, no talking head. |
| **Bottom** | Every caption locked to the lower zone. | Classic. Safest for readability. |
For vertical social exports, **Center** or **Top** is usually safer: platform UI (captions, username, buttons) covers the bottom of the frame.
### Fonts
The dropdown lists the fonts **installed on your machine**, and the exact font file is sent to the renderer so what you see is what you get. You can add **up to 3**.
* **1 font** — the safest, most coherent result.
* **2 fonts** — enables *pairing*: a small lead-in line in one face and a big payoff line in another. This is what makes a caption look designed rather than typed.
* **3 fonts** — extra material for the art director to choose from; useful when you're not sure which one fits your footage.
Typography stays **consistent across the whole video**: whatever the AI settles on for the first captions is applied to all of them, so the look never drifts halfway through. Multiple fonts create contrast *inside* a caption (line 1 vs line 2), not a different font every five seconds.
Pick a **heavy display or condensed** face for the payoff line and a **regular sans** for the lead-in. Pairing two similar sans faces produces contrast so weak it reads as a mistake.
### Download folder
Defaults to your **project folder** (marked `project`). Change it if your project lives on a slow network drive — rendering writes one file per caption, and a fast local disk keeps import snappy. **Reset** puts it back to the project folder.
## Writing a Good Brief
The brief is read as **art direction**, so describe what you want to *see*, not what you want the tool to *do*.
### What works
* **The container:** *"Put every caption on a solid white rounded card with generous padding and a soft shadow."*
* **The typography:** *"Bold white uppercase condensed text, tight tracking, sentence case for the lead-in."*
* **The composition:** *"Exactly two lines: a small lowercase lead-in, then a payoff about twice the size, lines almost touching."*
* **The motion:** *"Words rise one by one with a gentle ease-out, no scale change, no rotation."*
* **The restraint:** *"No outline, no glow, no tilt, nothing else on screen."*
* **The editorial intent:** *"Only highlight actionable advice and numbers. Skip pleasantries and transitions."*
* **The tone of the words:** *"Rewrite what I say into short punchy hooks, never full sentences."*
### What doesn't work
* **A manual list of timecodes or exact sentences** — the AI chooses the moments itself from what is said. If you already know the exact texts and timings, use the **Copilot** instead (see below).
* **Asking for more or fewer captions** — that's the *Texts per minute* slider.
* **Asking for a longer display time** — that's the *Seconds on screen* slider.
* **Asking for a font you didn't select** — only fonts installed and picked in Advanced are available.
* **Asking for images, logos, icons or video inside a caption** — the engine renders type, panels and effects, nothing else.
* **Vague vibes alone** (*"make it cool"*) — you'll get the default look. Name a colour, a weight, a shape, a motion.
### Recommended pattern
> *"\[Container]. \[Typography and case]. \[Composition / number of lines]. \[Motion and rhythm]. \[What to avoid]. \[What to highlight editorially]."*
**Example:**
> *"No box. Bold white uppercase condensed text in the lower third with a very discreet halo. One line only, never stacked. Words rise gently one by one, no scale, no rotation. Highlight only numbers and product names in a saturated accent. Nothing else on screen."*
Start from a **style preset**, run once, then edit two or three words of the brief and run again. Iterating on a working brief beats writing a perfect one from scratch.
## Style Presets
Clicking a preset only **fills the brief box** — nothing is locked. Edit it, merge two of them, or delete half a sentence.
| Preset | Look |
| ------------------ | ----------------------------------------------------------------------------------------------------- |
| **Institutional** | White rounded card, dark text, one line, soft shadow. Corporate, clean, very legible. |
| **Retro Button** | Sky-blue card, hard offset shadow, heavy black uppercase. Pops in as one block. |
| **Cinematic** | No box, white condensed uppercase in the lower third, discreet halo, slow restrained motion. |
| **Apple Message** | Blue iMessage-style bubble, white conversational text, springy pop-in. |
| **Grid + Pairing** | Paper card hugging the text, two lines, two fonts, a headline plus a handwritten accent. |
| **White Pairing** | No box, pure white with a soft bloom, small lead-in over a large uppercase payoff, word-by-word rise. |
## Using Vibe Kinetic from the Copilot
You can also ask the **Copilot** in chat to place animated text — it uses the same engine, but skips transcription and curation because *you* supply the words.
> *"Add kinetic text: 'ONLY 3 STEPS' at 4s, 'STEP ONE' at 9s, 'STEP TWO' at 15s — Apple style, white on two lines, 3 seconds each."*
What you can specify in the request:
| You provide | Notes |
| ------------------------------- | ---------------------------------------------------------------------------------------- |
| **The texts and their timings** | Required. Up to 200 items. |
| **A style brief** | Same language as the panel brief. |
| **Colours** | Given as a small palette. |
| **Placement** | `auto`, `top`, `center` or `bottom`. |
| **A font** | Free-form name — it is matched to a font available on your machine or to a built-in one. |
| **A target duration** | 2–10 s, applied to any text you left open-ended. |
Ask for a font by name and you'll get the closest real match, never a missing-font error. If nothing close is installed, a sane built-in face is used instead.
## Common Pitfalls
* **No sequence audio** → nothing to analyse, no captions. The panel path needs a spoken track.
* **Project not saved** → the download folder falls back to your temp directory; rendered clips still import, but they're stored outside your project.
* **Too dense + too long** → captions overlap and get trimmed or dropped. Lower the density before raising the duration.
* **All four palette slots set to near-identical colours** → highlights become invisible. Keep one clearly contrasting accent.
* **Advanced ON with no font selected** → the Generate button stays disabled. Pick at least one font.
* **Very long sequences** take a while: one clip is rendered per caption. A one-hour interview at 8/min is roughly 500 clips.
**Recommended starting point** for a talking-head social edit: pick **White Pairing** or **Cinematic**, density **8**, duration **2 s**, Advanced **ON** with one condensed display font plus one regular sans, placement **Center** for vertical / **Bottom** for horizontal.
# Create an Account
Source: https://docs.premierecopilot.com/getting-started/create-account
Get started with Première Copilot by creating your account and accessing the dashboard.
## Creating an Account
To begin using Première Copilot, you must first create a user account.
Go to the [Première Copilot website](https://premierecopilot.com).
Click on the **"Download for free"** button located at the top right of the page.
Fill out the form with your first name, last name, and other required details.
Click **"Sign up"**.
**Verify your email:** Go to your inbox and validate your account by clicking the confirmation link.
**Note:** Upon signing up, you will receive a promo code to purchase the "Podcast" plugin at a discounted rate.
## Accessing the Dashboard
Once verified, return to the website and click **"Sign in"** to access your dashboard. From here, you can:
Download the latest version of the application.
Schedule a demo to learn more features.
Join our community on Discord.
Access your promo code for referrals.
![Dashboard Placeholder]()
# Setup Fal.ai API Key
Source: https://docs.premierecopilot.com/getting-started/fal-api-key
Generate and configure your Fal.ai API key to enable AI image and video generation features.
## Why Fal.ai?
PremiereGPT uses **Fal.ai** to provide state-of-the-art AI generation capabilities, including high-quality image and video generation (using models like FLUX, Kling, or Luma). To use these features, you need to connect your own Fal.ai account.
## Step 1: Create a Fal.ai Account
To get started, you need to register on the official Fal.ai platform.
Go to the [Fal.ai official website](https://fal.ai/).
Click on the **"Login"** or **"Get Started"** button in the top right corner.
Fal.ai uses OAuth for secure access. You can sign up using:
* **GitHub** (Recommended for developers)
* **Google**
Select your preferred provider to create your account instantly.
**Note:** New accounts often come with complimentary credits to test the models. Check your [Billing Dashboard](https://fal.ai/dashboard/billing) to see your balance.
## Step 2: Generate your API Key
Once your account is created, you can generate the secret key required by PremiereGPT.
Navigate to the [Fal.ai Dashboard](https://fal.ai/dashboard).
Click on the **"API Keys"** tab in the left sidebar or go directly to [fal.ai/dashboard/keys](https://fal.ai/dashboard/keys).
Click the **"Create Key"** button.
Give your key a name to remember where it is used (e.g., `PremiereGPT-Production`).
**Important:** Your API Key will only be displayed **once**.
* Click the copy icon next to the key.
* Save it in a secure location (like a password manager).
* If you lose it, you will have to delete it and create a new one.
**Security Warning:** Never share your API key or commit it to a public repository. Treat it like a password.
## Step 3: Configure PremiereGPT
Now that you have your key, you need to add it to PremiereGPT.
You can paste your API key in two places:
* **User Settings:** Open the **User Settings** panel in PremiereGPT and look for the **Fal.ai API Key** field.
* **API Modal:** Paste it directly into the configuration modal that appears when you first use an AI generation tool.
Find the dedicated field and paste your secret key. Note that while many keys start with `fal_`, this is not always the case.
Click **Save**. You can now start generating AI content directly in your timeline!
## Official Resources
For more detailed information, consult the official Fal.ai documentation:
* [Fal.ai Authentication Guide](https://fal.ai/docs/documentation/setting-up/authentication)
* [Managing API Keys](https://fal.ai/docs/platform-apis/v1/keys/create)
* [How to add Credits / Billing](https://fal.ai/dashboard/billing)
**Pro-Tip:** Make sure you copied the entire string from the dashboard. You can verify your key is working by attempting to generate a simple image or searching for a model in the library.
# Installation & Setup
Source: https://docs.premierecopilot.com/getting-started/installation
Download, install and initialize Première Copilot on your system.
## Downloading the App
Log in to your **Première Copilot Account**.
Click the **"Download App"** button.
Select the correct installer for your operating system:
* **Windows**
* **Mac Intel** (Older models)
* **Mac Silicon** (M1, M2, M3, M4 chips) — *Please ensure you select the correct architecture*.
## Installing the Software
**Unzip** the downloaded file.
**Run the installer**. This will install Première GPT in the correct directory.
**Restart Adobe Premiere Pro**. It is also recommended, though not mandatory, to restart your computer.
**Security Disclaimer:** You may receive a virus alert from Windows or Mac security. This is a false positive caused by the software being new and not yet fully whitelisted by manufacturers. The software is safe. Please authorize the application in your security settings to proceed.
## Initializing the Extension
Open Adobe Premiere Pro.
Go to **Window > Extensions > Première GPT Beta**.
**Log in** using the same credentials created on the website.
**First Launch:** If this is your first time starting the extension, allow 1–2 minutes for the backend engine to initialize. Subsequent launches will be much faster.
![Extension Menu Placeholder]()
# Introduction
Source: https://docs.premierecopilot.com/getting-started/introduction
Welcome to PremiereGPT documentation
# Welcome to PremiereGPT
This is the official documentation for PremiereGPT.
## Getting Started
Check out the [Create an Account](/getting-started/create-account) guide to get started.
# Affiliate Program
Source: https://docs.premierecopilot.com/support/affiliate
Earn 33% commission on sales generated through your referrals.
Earn **33% commission** on sales generated through your referrals.
## Eligibility
To become an affiliate, you must:
* Have an active account.
* Have purchased at least one product (Subscription or Podcast plugin).
## How it Works
Locate your **custom promo code** in your dashboard.
Share this code and the link with your contacts.
If a contact makes a purchase using your code (or on your recommendation), send an email to support containing the **email addresses of the contacts who created an account**.
Verification will be performed on our end to confirm the referral.
## Payouts
* **Commission Rate:** 33% of the amount paid.
* **Recurring:** Applicable monthly if the user subscribes.
* **Payment Methods:** Payouts are made via PayPal or Stripe refund.
# Support & Community
Source: https://docs.premierecopilot.com/support/community
Get help and join the Premiere Copilot community.
If you require assistance, we have several channels available to help you.
Accessible via the Dashboard or the "Join Discord" button.
Click "Book a demo" in your account to schedule a session.
Contact us directly at `admin@premieregpt.com` for specific inquiries.
# Copilot doesn't start
Source: https://docs.premierecopilot.com/support/copilot-doesnt-start
Listing of known bugs and best practices to follow if the extension doesn't launch.
If you are having trouble launching Premiere Copilot, here are the solutions to the most common problems.
## 1. Using a VPN
If you are using a **VPN**, it can sometimes block the local connection between the extension and its computation engine.
Disable your VPN and try restarting the extension.
## 2. First Launch (Waiting)
During the very first launch, the extension must initialize the local AI engine. This can take some time depending on your machine's performance.
It is sometimes necessary to wait between **2 and 5 minutes** during the first launch for everything to be operational.
## 3. Debugging Session
If the previous steps don't work, you might need to manually check what is blocking the engine.
Go to the [Debugging](/support/debug) section of this documentation.
Follow the instructions to launch the API manually and identify any error messages.
***
If the problem persists, feel free to join us on [Discord](https://discord.gg/WyBNCbZhBD) or contact us via email.
# Debugging
Source: https://docs.premierecopilot.com/support/debug
Troubleshoot and fix issues with the Première Copilot API.
If the extension is not responding, launch the API manually to identify the issue.
### 1. Copy the API Path
Choose the path corresponding to your installation:
**Mac:**
```bash theme={null}
/Library/Application Support/Adobe/CEP/extensions/PremiereGPTaescripts/js/libs/PremiereCopilotAPI/
```
**Windows:**
```bash theme={null}
C:\Program Files (x86)\Common Files\Adobe\CEP\extensions\PremiereGPTaescripts\js\libs\PremiereCopilotAPI\
```
**Mac:**
```bash theme={null}
/Library/Application Support/Adobe/CEP/extensions/PremiereGPTBeta/js/libs/PremiereCopilotAPI/
```
**Windows:**
```bash theme={null}
C:\Program Files (x86)\Common Files\Adobe\CEP\extensions\PremiereGPTBeta\js\libs\PremiereCopilotAPI\
```
**Mac:**
```bash theme={null}
/Library/Application Support/Blackmagic Design/DaVinci Resolve/Workflow Integration Plugins/DavinciGPT/js/libs/PremiereCopilotAPI/
```
**Windows:**
```bash theme={null}
C:\ProgramData\Blackmagic Design\DaVinci Resolve\Support\Workflow Integration Plugins\DavinciGPT\js\libs\PremiereCopilotAPI\
```
### 2. Access the Folder
* **Mac:** Open **Finder**, press `Cmd + Shift + G`, paste the path, and hit `Enter`.
* **Windows:** Paste the path into the **File Explorer** address bar and hit `Enter`.
### 3. Run and Report
Double-click **PremiereCopilot** (Mac) or **PremiereCopilot.exe** (Windows). A terminal window will open.
**If any error appears:**
* Take a screenshot of the terminal window.
* Send it **immediately** to **[admin@premierecopilot.com](mailto:admin@premierecopilot.com)** or in Discord.
If the window closes too fast to see the error, record a quick video of your screen to capture the message.
# Uninstalling Première Copilot
Source: https://docs.premierecopilot.com/support/uninstall
Completely remove the extension and all of its files from Windows or macOS.
Première Copilot installs in **two places only**:
1. The **extension folder** (the CEP extension + the local API engine), installed system-wide.
2. A **data folder** in your Documents (cache, transcriptions, generated files, presets).
Deleting both removes the software completely. Follow the steps for your operating system.
**Close Adobe Premiere Pro (and DaVinci Resolve) before you start.** Files that are in use cannot be deleted, and Premiere Pro must not be running when the extension folder is removed.
## Windows
### 1. Delete the extension folder
All extension files live under the Adobe CEP extensions folder in **Common Files**. Paste this path in the **File Explorer** address bar:
```bash theme={null}
C:\Program Files (x86)\Common Files\Adobe\CEP\extensions\
```
Delete the folder(s) below that exist on your machine — depending on which installer you used, only one of them will be present:
| Version | Folder to delete |
| --------------- | ----------------------------------------------------------------------------------------------------- |
| Beta / standard | `PremiereGPTBeta` |
| aescripts | `PremiereGPTaescripts` |
| aescripts GenAI | `PremiereGPTaescripts_genai` |
| Older versions | `PremiereCopilot` |
| DaVinci Resolve | `C:\ProgramData\Blackmagic Design\DaVinci Resolve\Support\Workflow Integration Plugins\davinciClaude` |
Deleting these folders requires **administrator rights**. Windows will ask you to confirm — click **Continue**.
The **aescripts** version also registers a classic uninstaller. You can remove it from **Settings > Apps > Installed apps > PremiereGPTaescripts** instead of deleting the folder by hand.
### 2. Delete the data folder
This folder holds the cache, thumbnails, transcriptions, generated audio and presets. Paste this in the File Explorer address bar (replace nothing — `%USERPROFILE%` resolves automatically):
```bash theme={null}
%USERPROFILE%\Documents\Adobe\Premiere Pro\Premiere Copilot
```
Delete the whole **Premiere Copilot** folder. It contains the following subfolders created at install time:
`audio_sync` · `image_generation` · `matplotlib_cache` · `music_analysis` · `rush_db` · `script` · `seq_preset` · `sfx` · `temp` · `thumbnails` · `transcription_analysis`
If you saved anything of your own in these folders (custom sequence presets in `seq_preset`, generated files in `image_generation` or `sfx`), back it up before deleting.
### 3. (Optional) Remove the debug registry key
The installer enables Adobe's unsigned-extension mode. It is shared by **all** CEP extensions, so only remove it if you use no other third-party Premiere Pro extension.
Open **Registry Editor** (`regedit`) and delete the `PlayerDebugMode` value in:
```bash theme={null}
HKEY_CURRENT_USER\Software\Adobe\CSXS.11
HKEY_CURRENT_USER\Software\Adobe\CSXS.12
```
Removing `PlayerDebugMode` will break any other unsigned extension you have installed. If in doubt, leave it as it is — it has no effect once Première Copilot is gone.
## macOS
### 1. Delete the extension folder
Open **Finder**, press `Cmd + Shift + G`, and paste:
```bash theme={null}
/Library/Application Support/Adobe/CEP/extensions/
```
Delete the folder(s) below that exist on your machine:
| Version | Folder to delete |
| --------------- | -------------------------------------------------------------------------------------------------------- |
| Beta / standard | `PremiereGPTBeta` |
| aescripts | `PremiereGPTaescripts` |
| aescripts GenAI | `PremiereGPTaescripts_genai` |
| Older versions | `PremiereCopilot` |
| DaVinci Resolve | `/Library/Application Support/Blackmagic Design/DaVinci Resolve/Workflow Integration Plugins/DavinciGPT` |
macOS will ask for your **administrator password** — this folder is outside your user account.
If you prefer the Terminal, this removes every possible variant in one go:
```bash theme={null}
sudo rm -rf "/Library/Application Support/Adobe/CEP/extensions/PremiereGPTBeta" "/Library/Application Support/Adobe/CEP/extensions/PremiereGPTaescripts" "/Library/Application Support/Adobe/CEP/extensions/PremiereGPTaescripts_genai" "/Library/Application Support/Adobe/CEP/extensions/PremiereCopilot"
```
### 2. Delete the data folder
Press `Cmd + Shift + G` in Finder and paste:
```bash theme={null}
~/Documents/Adobe/Premiere Pro/Premiere Copilot
```
Move the whole **Premiere Copilot** folder to the Trash. It contains the subfolders created at install time:
`audio_sync` · `image_generation` · `matplotlib_cache` · `music_analysis` · `rush_db` · `script` · `seq_preset` · `sfx` · `temp` · `thumbnails` · `transcription_analysis`
Or, in the Terminal:
```bash theme={null}
rm -rf ~/Documents/Adobe/"Premiere Pro"/"Premiere Copilot"
```
Back up anything of your own stored in these folders (custom presets in `seq_preset`, generated files in `image_generation` or `sfx`) before deleting.
### 3. (Optional) Forget the installer receipt
macOS keeps a record of the `.pkg` installation. It takes no meaningful space, but you can clear it:
```bash theme={null}
sudo pkgutil --forget com.trial17.premierecopilot
```
For the GenAI and DaVinci builds, the identifiers are `com.trial17.premierecopilot-genai` and `com.trial17.davinciclaude`.
## Verify the uninstall
Reopen Adobe Premiere Pro.
Go to **Window > Extensions**. **Première GPT Beta** should no longer be listed.
Both the CEP extension folder and `Documents/Adobe/Premiere Pro/Premiere Copilot` should be gone.
Uninstalling does **not** cancel your subscription. Manage or cancel it from your [account dashboard](https://www.premierecopilot.com/account).
***
Reinstalling later? Just run the installer again — it recreates both folders automatically. Note that the installer already deletes any previous version before installing, so you do **not** need to uninstall manually before an update.