Transcript & AI
Transcribe sequences and analyze with AI
Ready
Transcribe Sequence
Exports sequence audio → ffmpeg → Whisper API · IN/OUT marks override auto-trim · Saved transcripts auto-restore on project re-open
Transcript
No transcript yet
Transcribe a sequence or import a file
Ask AI about this transcript
AI sees the loaded transcript as context. Conversation is not saved across panel restarts.
Silence Detection
AI Analysis
No analysis yet
Detect chapters, B-Roll, key points, motion graphics & CTAs from your transcript
B-Roll Studio
Generate AI videos for every B-Roll prompt in one click. Imports automatically into Premiere when ready.
No B-Rolls detected yet
Run AI Analysis on your transcript first — B-Roll suggestions will appear here for batch generation.
Web Capture
5s
Load ouvre la page dans Chrome (preview). Choisis le format (16:9 ou 9:16 vertical), la durée et le scroll, puis Enregistrer (vidéo MP4 auto-scroll) ou Capture (PNG pleine page) → Importer au playhead dans le bin
Web Captures. Ouvre un projet Premiere d'abord.
Media Finder
Images · B-Roll · YouTube · GIFs · Pinterest · SFX — ou colle un lien YouTube / TikTok / Instagram
Free web image search — no API key needed. A green ✓ transparent badge marks images with a real transparent background. ⤓ drops it at the playhead (on its own track, nothing overwritten) · ⬆4K upscales + removes watermarks via AI (needs a KIE key) · ✂BG removes the background with remove.bg (skipped automatically if already transparent). Files land in
~/Downloads/google-images/.
Transitions Gallery
Click a transition to apply it centered on the playhead
Animated Subtitles
Sous-titres animés mot à mot · export alpha
Load or transcribe a video first
Animated subtitles are built from the transcript's word-level timestamps. Re-transcribe if you don't have words yet.
Source script
With a loaded transcript: each shot inherits its real voiceover timing and the generated clips drop on the timeline at the exact spoken moment.
Pasted scripts (no timecodes): clips fall back to a contiguous strip.
Note: cuts between shots are hard cuts in v2 — keyframe-to-keyframe transitions (Opus phase 07) are not yet implemented.
Visual style
Brand logo
Optional
Drop a logo (PNG, JPG, WebP). Branded shots are generated image-to-video with your
logo as the opening frame — the real mark (exact colors & shape) appears, then the
shot animates from it. The intensity below sets how many shots are branded; the rest stay
text-to-video. Skip this for pure photo-real clips with no branding.
Drop logo or click to browse
PNG · JPG · WebP · max 5 MB
Generation settings
Runs the AI orchestration only — no video credits used yet. Review the shot list before generating.
⚡ One click: plans + generates EVERY shot as motion-design (a reference still → animated via image-to-video → auto-imported). Skips the review step and uses KIE credits immediately. The per-marker Motion Design Studio is unaffected.
Podcast Multi-Speaker
Map each audio track to a speaker, then let the plugin detect when each speaker is talking — handles overlapping speech (2-3 people at once).
Open a sequence with multiple audio tracks to begin.
Setup
Saved per sequence. Cut method drives how Apply Cuts treats idle clips; wide-shot frequency is a hint for future cadence-based wide-shot inserts (not used in Phase 2 — wide tracks are shown during overlapping speech).
Speakers
Each speaker maps to one video track (their camera). Use the Group/Wide role for the wide shot.
Audio Tracks
Assign each audio track to one speaker, or mark it Ignore (music, SFX, room mic).
Speech Detection
Solos each speaker's audio track in turn, runs silence detection, and computes per-speaker speech regions. Your mute states are restored automatically.
Aucun dossier connecté
Connectez les dossiers de votre bibliothèque (vidéos, sons, images) pour les parcourir, rechercher et prévisualiser ici — puis les déposer dans votre montage.
Project Organization
Confidentialité
Interface
API Providers
Pick a separate brain for analysis. Anthropic and KIE don't transcribe, so keep a Whisper provider above.
🎙️ Transcription — Whisper (Audio → Text)
Use a whisper deployment (api-version 2024-10-21+) — it returns word-level timestamps the subtitle & correction tools need.
gpt-4o-transcribe does not provide word timestamps, so don't use it here.
🤖 AI Analysis — Chat (GPT-5 / GPT-4o)
Paste either a
/chat/completions URL (deployment name in the URL) OR a /openai/responses URL — the plugin auto-detects which Azure API to use. Copy it verbatim from Azure AI Studio → Deployments.
The EXACT deployment name from your Azure resource — case-sensitive (e.g.
gpt-5, not GPT-5 or gpt-5-mini). For /chat/completions URLs this just drives reasoning-model detection; the deployment in the URL is what's actually called.
Paste the
/images/generations URL of your gpt-image deployment. Used as a fallback for image generation when KIE fails, and when you pick GPT-image as the reference-image model. Leave blank to keep using KIE only.
Dedicated key for the image deployment. Leave blank to reuse the Chat API Key above. Stored only in your local
~/.transcript-analyzer/config.json (mode 0600) — never in the plugin code.
Smart Fallback to Azure
When your primary analysis provider (Anthropic, KIE, OpenAI, Gemini) hits a transient error — rate limit, server overload, timeout — the call is automatically retried on Azure so your analysis still completes.
Auth errors (bad API key) never fall back — those surface as errors so you can fix them. Only temporary failures are caught.
Auth errors (bad API key) never fall back — those surface as errors so you can fix them. Only temporary failures are caught.
No fallbacks this session.
KIE API — Video Generation
Send B-Roll prompts directly to kie.ai to generate videos and import them into your timeline.
Note: not all models support every resolution. Hailuo defaults to 768p, WAN/Seedance to 720p, Sora 2 Pro supports up to 1080p.
Analysis Types
Toggle which types the AI looks for. Default 5 types are on; opt-in extras below.
Extras (opt-in)
Asset APIs
Used by Claude when fetching brand logos, product photos, or web captures into your Premiere project. Add the keys you have — leave the rest empty.
Auto-rotates: when one account hits its 50/month limit, the next key takes over automatically. 3 accounts = 150 free cut-outs/month.
Plus, key-less sources are always on : Openverse (CC photos) and Clearbit Logo (brand logos when AI detects a tool/app name).
Feedback
Found a bug, got an idea, or just want to say hi? Drop it here. Stored locally for now — nothing is sent yet.