Video Bridge docs

Introduction

Video Bridge is an AI co-pilot for Premiere Pro. Tell it what you want — it cuts silences, tags B-roll, transcribes interviews, and acts on your timeline.

What it helps you do

• Smart Cut — auto-detect silences and filler words
• B-roll classifier — AI tags every clip
• Captions — word-level, 10 animated styles
• Chat your edits — "cut everything before 0:30" → executes
• Bin organise — bulk rename and auto-bin by tag

Installation

Signed .zxp extension for Premiere Pro 2024+.

Quick install

1. Download. Get VideoBridge.zxp from your order receipt.
2. Close Premiere. Installer needs Premiere closed.
3. Run installer. Double-click — ZXPInstaller signs and registers.
4. Open Premiere. Window → Extensions → Video Bridge.

System requirements

Setup & API keys

Video Bridge is BYOK — you pay providers directly.

Supported providers

Adding a key

1. Open Settings. Gear icon.
2. Paste your key. Each provider has its own field.
3. Pick defaults. Sonnet 4 for chat, Whisper-large-v3 for transcripts, Gemini Flash for B-roll.

Your first cut

Two minutes from panel-open to a cleaned-up timeline.

1. Drop a clip on the timeline.
2. Switch to Smart Cut tab. Default thresholds work for most footage.
3. Click Preview. Cuts appear as markers — scrub to verify.
4. Click Apply. Ripple deletes silences. Fully undoable.
5. Iterate. Raise threshold for noisier audio.

Writing effective prompts

When chatting with the timeline, specificity wins.

Be specific about edit operations

tighten this up

remove silences over 0.5s
between 00:00:30 and 00:02:00

Transcript-anchored prompts

# Find moments by what was said
"Find moment where I say 'the secret is'"
"Cut every section about pricing"
"Add a marker on every laugh"

Bin queries

# Query via B-roll tags
"Show all outdoor clips under 5s"
"Find clips tagged interview + closeup"

Common workflows

Three recipes — copy them, adapt them.

Recipe 1 — Podcast in 10 minutes

1. Smart Cut. Pull silences + fillers (saves 15–20%).
2. Caption. Cloud Whisper → "Podcast-Bold".
3. Chat. "Cut sections under 8 seconds".
4. Organise. Auto-bin by interview / B-roll / titles.

Recipe 2 — Tag a 4-hour shoot

1. Select all unclassified clips. 200+ at once is fine.
2. B-roll → Classify. Gemini Flash, ~$0.20 for whole shoot.
3. Search. "outdoor + person + warm" → 12 instant results.

Recipe 3 — Caption a 60-min interview

1. Smart Cut first. Cuts Whisper bill by 10–20%.
2. Cloud Whisper. Pick "Minimal-Mono".
3. Adjust. Essential Graphics → bulk restyle.

Best practices

Pick the right model

Smart Cut before transcribe

Removing silences first reduces Whisper costs by 10–20%. Always cut first.

Custom B-roll prompts

Default schema is generic. For wedding/vlog/doc — write a project-specific prompt in Settings → B-roll.

Troubleshooting

Smart Cut produces too many/few cuts

• Raise silence threshold for noisy recordings
• Increase padding (80ms → 150ms)
• Disable filler removal to keep pauses

"API key invalid"

• Re-paste — trailing space breaks it
• Confirm OpenAI key has Whisper access
• Switch to Local Whisper as fallback

B-roll returns weird tags

• Write a custom prompt for your genre
• Increase frames-sampled-per-clip

Panel crashes Premiere on open

CEP cache issue. Delete ~/Library/Caches/CSXS/ (macOS) or %APPDATA%\Adobe\CEP\extensions\cache\ (Win).

Smart Cut

Auto-detect and ripple-delete silences, fillers, bad takes.

Workflow

1. Select clip(s). Multi-select supported.
2. Configure thresholds. Silence (-40dB), padding (80ms), fillers.
3. Preview cuts. Markers appear — scrub to verify.
4. Apply. Ripple-deletes. Undoable.

Filler word list

Settings → Smart Cut. Add language-specific fillers (Hindi: matlab, haan) or your own tics.

B-roll classifier

AI scans your bin and tags every clip — searchable instantly.

How it works

1. Select clips (or "all unclassified")
2. Choose vision model (Gemini Flash recommended)
3. Click Classify — samples frames, asks AI to tag
4. Tags appear as metadata, searchable

Default schema

type:      "interview" | "broll" | "screen"
location:  "indoor" | "outdoor" | "studio"
subjects:  ["person", "laptop"]
mood:      "warm" | "tense" | "energetic"

Custom prompts

Settings → B-roll → Custom prompt. Paste your own schema for project-specific tagging.

Caption panel

Word-level captions with animated styles, built on Whisper.

Workflow

1. Select footage clip. Any audio-bearing video.
2. Choose transcription. Local (offline, ~1.5GB model) or Cloud.
3. Pick a style. 10 presets.
4. Generate. Renders as Essential Graphics on V2.

Languages

Whisper supports 99 languages. Force language for code-mixed audio.

AI to Premiere

Chat with your timeline. Plain English → Premiere operations.

Example prompts

# Edit ops
"Cut everything before 00:00:30"
"Move clip 3 to V2"
"Add a 0.5s dip-to-black between cuts 4 & 5"

# Transcript-based
"Find moment where I say 'the secret is'"
"Cut every section about pricing"

# Bin queries
"Find outdoor B-roll under 5s"

Bin organise

Rename, recolor, re-bin clips in bulk.

Smart rename

AI suggests names from B-roll tags: MVI_4382.MOV → interview_sagar_closeup_01.mov.

Auto-bin

Create bins by tag: Interviews / B-roll / Screen / Titles. Clips moved, not duplicated.

Shortcuts

Cost & usage

Privacy & data

What's sent to providers

• Audio file (Whisper transcription)
• Sampled 720p JPEG frames (B-roll)
• Transcript text (chat)

What stays local

• Video files (never uploaded)
• API keys (OS keychain)
• Chat history, classification cache

Local Whisper

Settings → Captions → Local. First run downloads ~1.5 GB. Then fully offline.

FAQ

Changelog

What's new, version by version.

Roadmap

What's coming next.