Help

Open the Bridge panel in Premiere and click "Send to Paper Edit" — there are no files to export or upload.

• The Bridge reads your sequence and its transcripts directly.
• You build the paper edit by selecting words in the browser.
• The Bridge builds the finished sequence back in Premiere when you're done.

Drag across words in the transcript, exactly like selecting text — release and that word range becomes a block.

• Snaps to whole words; a plain click does nothing, so you can't create a block by accident.
• A selection can cross speaker boundaries and still stays one block.

A block must sit within a single clip — you can't drag from the tail of one clip into the start of the next, since each block's timecodes are relative to one clip's media time.

Click a block, then drag its handles.

1. Click a block in the paper edit panel — its words highlight in the transcript and drag handles appear at the start and end.
2. Drag a handle word-by-word to trim or extend.
3. Click the block again (or elsewhere) to finish.

Three view-only toggles above the transcripts, applying everywhere at once — they never change what's stored.

• Timecodes — prefixes each paragraph with its start time. In the paper edit panel each block shows two: its source in–out timecode, then a second “@ mm:ss” showing how far into the finished paper edit that block begins. Reordering updates this instantly.
• Pauses — reveals silence and filler gaps flagged by Premiere (hidden by default).
• Tints — turns per-speaker background shading on or off (speaker labels always stay).
• Trims — shows or hides words you've cut with the pencil, struck through, so you can see what was removed and click to put it back.

Premiere flags silence and filler gaps but doesn't export the words — Paper Edit shows them as […] or [Xs], hidden until you toggle Pauses on.

1. In Premiere: Window → Text → Transcript
2. Click the ••• menu → Transcript view options…
3. Enable filler-word detection
4. Re-transcribe the sequence
5. Re-send from the Bridge panel

Transcripts created outside Premiere, or older ones, usually have no filler-word flags.

Each block has a row of icons — colours match the Premiere timeline.

• Tag — adds a label; becomes a blue marker in Premiere.
• Speech-bubble — click a word to drop a precise orange marker at that exact source time, with a note that becomes the marker's comment in Premiere. Imported Frame.io comments (see below) show the same way.
• Pencil — Edit mode, for crossing out words.
• Stacked speech-bubbles — collaborator comments for that block; separate from markers and never touches the Premiere timeline.
• Trash — deletes the block.

Click the pencil to open Edit mode, then click or drag across words to mark them for removal — they turn red and struck through.

• "Apply" removes the marked words (splitting the block if they were in the middle).
• "Suggest" proposes the same trim to collaborators instead of applying it.
• "Cancel" discards the marks.

Closing Edit mode with the pencil does none of these — your marks are kept, the block shows a red "N marked · not applied" pill, and you can click it any time to carry on. Marks are not saved into a version; saving one will warn you first.

Yes — cut words stay visible in the transcript, struck through, so you can tell deliberate cuts from material you simply never selected.

• Hover to see who cut it; click any struck-through word to put it straight back, rejoining the blocks the cut split apart.
• The Trims toggle hides and shows them.
• Cut words belong to the version they were made in — save a version and they save with it; load an older version and its cuts come back.

Cuts are never included in a Premiere build — only what's actually in the paper edit is built.

Yes — double-click a speaker chip to rename it, right-click to reassign everyone with that speaker to another.

• Updates the transcript, tints, and any blocks already using that speaker.
• The original transcript is never altered — overrides are reversible.

Cmd+Z undoes any change — create, delete, reorder, label, note, boundary, clear, or restoring a cut word. Cmd+Shift+Z redoes it.

• Drag the grip handle to reorder blocks.
• "Delete: Blocks | Labels | Notes" in the header deletes each (with a confirm) — also undoable.
• If a save ever fails, the change is rolled back and a message appears.

In a shared project your undo only affects your own changes, never a collaborator's.

The Settings button in the top-right of the page header controls what the built sequence is set to.

• With several sources at different frame rates or resolutions, it warns you and lets you choose: match the first source, match the last, or set frame rate and resolution manually.

The Export button downloads a Paper Edit JSON or prints your edit; the actual sequence is built from the Bridge panel back in Premiere.

• In Premiere, click Build in the Bridge panel — it pulls your paper edit and reconstructs the trimmed clips in order, with linked video/audio and your markers.
• By default it builds the current paper edit; the Version picker in the Bridge lets you build a specific saved version instead.

There's no XML or SRT in this workflow.

Yes — a shared paper edit is live. Changes appear in each other's window within about a second.

• You each control your own transcript columns and layout, but you're all shaping the same paper edit.
• It also works asynchronously — changes made while you're away show up next time you open the project.

A green dot and avatar showing their initials appears in the page header when someone else has the project open.

• Hover it to see their email; it disappears when they leave.
• Up to three avatars stack side-by-side, with "+N" if more are present.

There are no live cursors — it's a light "who's here" indicator.

Each block is handled independently, so two people working on different blocks never interfere.

• If you both change the very same block at the same moment, the most recent save wins.
• Your Undo only ever affects your own actions — it can't remove a collaborator's change.

Cmd+Z (Ctrl+Z) undoes your last change; Cmd+Shift+Z (Ctrl+Shift+Z) redoes it — both also available as buttons in the paper edit header.

• Every kind of change is covered — create, delete, reorder, label, marker, trim, clear.

In a shared project, undo affects only your own changes, never a collaborator's.

Open Comments from the page header, or click a block's stacked-speech-bubbles icon — threads sit in a margin beside the block they're attached to, like Google Docs.

• A composer bar at the top writes new comments; comments not tied to a block ("Whole paper edit") group there too.
• Post a comment and collaborators see it live and get a notification.
• Reply to build a thread, Resolve it once handled (Reopen later), and delete your own comments — owners can delete any.
• A block with comments shows a small blue count badge on its comments icon.

A comment is discussion. A suggestion is a proposed edit that can be applied with one click.

• In the composer bar, switch the title toggle from "Comments" to "Suggest", then choose an op.
• With a block focused, propose removing it, relabelling it, or moving it elsewhere in the paper edit.
• To propose adding a block: pick "Add block", choose where it goes, then "Select words in transcript" to drag-select the words.
• From a block's Edit mode: mark words to remove and click "Suggest" instead of applying directly (a trim suggestion).

Suggestions are ideal for a reviewer who wants to propose changes without altering the edit directly.

Open Comments — each open suggestion shows what it proposes, with Accept and Reject buttons.

• Accept applies the change immediately, exactly as if you'd made it yourself — syncs to everyone and can be undone with Cmd+Z.
• Reject leaves the paper edit untouched and marks the suggestion rejected.
• Anyone on the project can accept or reject; the suggester gets a notification either way.

If the target block has changed too much since the suggestion was made, Accept will tell you it can no longer be applied.

A collaborator was the last person to change that block — a small "edited by [name]" note appears in its header.

• Only shows when someone other than you made the last change, so you can see at a glance what a collaborator has touched.

Click Request review in the paper edit header — everyone else gets a notification that you'd like them to take a look.

• The lightweight handoff for asynchronous work; they can pick it up whenever they're next free.
• Producers and clients have the same button in their header.

Request review = "take a look and give me feedback." Sign off = "this is final — go build it in Premiere."

• Request review is the everyday, mid-process handoff — either editor or producer can send it any time.
• Sign off is only available to a producer or client (an editor never signs off their own work to themselves), and is meant for the very end of a producer's pass.

Both just send a notification — signing off doesn't lock the paper edit or stop anyone editing it further.

Your paper edit saves itself continuously — a "version" is a separate, deliberate named snapshot.

• Click Versions → "Save to version…" and give it a name (e.g. "Rough cut", "Producer's pass").
• Anyone with access to the project can save, load, or delete a version.
• Loading a version replaces the live paper edit with that snapshot — destructive, can't be undone, asks you to confirm first.
• If this browser session already saved or loaded that version, Save to version offers to update it in place instead of creating a duplicate; choose "Save as a new version" to keep the old one too.

The Versions list shows a "Current paper edit" row for the live state, and marks whichever saved version it was last saved as or loaded from — though further edits since then won't be reflected until you save again.

Yes — the Bridge's Build section has a Version picker under "Build from", defaulting to "Current paper edit".

• Pick any saved version to build that instead, with no need to load or restore it first.
• The live paper edit is never touched by building a version.

If any of that version's sources have since been removed, the Bridge tells you how many blocks can't be built and lets you build the rest with one extra click.

No — live editing, presence, comments, and suggestions all work without any AI key.

AI keys are only needed for AI search, the Topics tab, and the Refine tab.

Use the search bar at the top of each transcript column.

• Jump between matches with ↓/↑, Enter, or the arrow buttons next to the search bar.
• Press Escape to clear. Matches highlight in yellow.

Describe what you're looking for in plain language instead of an exact keyword.

• Type a phrase like "the bit where he talks about losing his job" into the search bar, then click the ✦ AI button.
• The AI reads the full transcript and returns every segment matching the idea, not just the literal words.

Results highlight in yellow using the same navigation as keyword search.

Go to Settings → AI API Keys and paste a key for any supported provider (Claude, OpenAI, or Gemini) — you only need one.

• Saved keys for more than one provider each show a green Active dot on the one in use; click "Set as active" on any to switch.
• With no preference set, the app defaults to Claude first, then OpenAI, then Gemini.

Keys are stored securely on our servers and never sent to the browser.

Claude (Anthropic), OpenAI, and Gemini (Google) — you pay the provider directly, with no markup.

• A typical search on a 500-block transcript costs less than half a US cent.
• Gemini via Google AI Studio has a free tier and is the easiest way to get started at no cost.

Each error message tells you what to fix.

• "Your AI key appears to be invalid" — check the key in Settings; it may be mistyped or expired.
• "Your AI account has run out of credits" — top up your balance with your provider.
• "Rate limit reached" — you've hit your provider's per-minute cap; wait and try again.

If AI search fails entirely, keyword search always works without a key.

An AI editorial assistant that reads your existing paper edit and surfaces structured observations — it never makes changes automatically.

• Every suggestion requires a deliberate action on your part.
• Switch to the Refine tab (above the transcript columns) once you have a draft paper edit to review.

Four types of observation.

• Repetition — two or more spans covering the same idea or near-identical phrasing.
• Pacing — a section taking disproportionate runtime relative to its ideas.
• Trim candidate — spans that feel redundant relative to what surrounds them (only when you've set a time target).
• Arc — a high-level structural observation, like resolving before it builds or opening with the conclusion.

For multi-speaker edits, the AI includes speaker context — observations may note which speaker a span comes from, and suggestions about balance or repetition can reference speakers by name.

Switch to Refine, optionally enter a target duration (e.g. "1:30" or "90s"), then click Analyse.

• The AI returns observation cards: a neutral observation, a question to sit with, and a "Block N" button per relevant block that jumps you to it.
• Some cards include a "Remove block N" button — hover to see why, and it shows how much runtime removing it would save.
• Dismiss any card with the ✕ button.

Keep editing after analysing and an amber banner appears ("Your paper edit changed since this analysis") with a one-click Re-analyse.

Switches you to the Sources tab and pre-fills the AI search with the relevant span's text — firing automatically so you can browse alternatives from the full rushes.

• On multi-source projects, a "Find alts: This source / All" toggle appears so you can scope the search.
• Found something better? Select it as normal and switch back to Refine to re-analyse.

The block is deleted the normal way — Cmd+Z immediately recovers it.

• The suggestion card disappears once the removal succeeds.

There's no "Apply all" — each suggestion is acted on individually or dismissed.

Yes — the same key you save for AI search.

• Go to Settings → AI API Keys and paste a key for Claude, OpenAI, or Gemini.

Analysis of a 20-span paper edit costs roughly $0.002 at current Haiku / GPT-4o-mini pricing.

Analyses your transcript with AI to produce a structured map: broad themes broken into subtopics, each with the segments that cover it.

• Sits alongside Sources and Refine — click Topics above the paper edit panel to open it.
• Designed for documentary and interview editing, where you need to understand what was said across many hours of rushes before shaping a structure.

A switcher at the top of the Topics tab generates three different kinds of map from the same transcript.

• Topics (themes and subtopics), Emotions (emotional moments), Story Beats (narrative structure).
• Each is generated, searched, and added to the paper edit exactly the same way — switching kind just changes what the AI looks for.

Each source keeps a separate map per kind, so generating one doesn't affect the others.

Open Topics and click "Map topics" (or "Map emotions" / "Map beats") on any source card.

• Returns a two-level hierarchy in a few seconds.
• Requires an AI key saved in Settings — the same key used for AI search.

Costs roughly $0.01–0.03 per transcript at current provider pricing.

Each source gets its own collapsible card — click a topic to see subtopics, click a subtopic to see individual segments.

• Topics are labelled in short, editor-friendly title-case names (e.g. "Career Beginnings", "On Collaboration").
• Each subtopic includes a brief description of what was said.
• For multi-speaker sources, each segment shows the speaker's name.

Click the + icon next to any segment to add it — it switches to a green tick.

• Click a subtopic's count button (e.g. "+ 4") to add every not-yet-added segment in that subtopic at once.

There's no single button to add an entire top-level topic — add subtopic by subtopic, or use the block count to gauge how much a topic covers before committing.

Yes — the resulting block is labelled with its subtopic's name, e.g. "First Job in TV".

• Edit or remove the label in the paper edit panel as normal.
• Works the same for a normal transcript or a "Selects" (derived) column.

When two or more sources share a broad topic, it appears in "Across sources" at the top of the Topics tab.

• Each entry shows how many sources cover it and how many segments exist across them.
• Expand to see a row per source with normal add controls, or use the group's own add-all button.

The extraction prompt assigns a consistent category slug (e.g. career_beginnings) to topics covering the same theme, even if named differently across sources.

• The app groups by that slug client-side — no extra AI call.

Use the search bar at the top of the Topics tab, visible once any source has a map for the selected kind.

• Filters the hierarchy live, keeping the topic → subtopic structure, matching topic name, subtopic name, or subtopic description.

A plain local filter, not an AI search — there's no separate AI-powered topic search.

Click the "txt" button in a source's header to download every topic currently generated for that source and kind as a single plain-text file.

• Includes each subtopic's description and segment text/timecodes.

No per-topic export and no JSON/Paper-Edit export from this tab — add segments individually (or via a subtopic's add-all), then export or build from the paper edit panel.

Yes — click "Regenerate" (the button that said "Map topics" before a map existed) to run a fresh analysis; the old map is replaced.

• Segments already added to your paper edit are not affected by a regeneration.

There's no automatic warning if the transcript has changed since the map was generated — it's on you to remember.

Yes for generating a map (any kind) — needs an AI key in Settings → AI API Keys.

Text search (live local filtering) works without a key.

In the Bridge panel, tick every sequence you want, pick the existing project in the Send to dropdown, and click Add sources.

• Each source gets its own colour-coded transcript column (up to 8 colours), so you always know which camera or interview a line came from.
• Each column header shows a source picker (name + position, e.g. 1/3) so you can see and switch which source is loaded without losing your scroll position.

Yes — the Sequences field is a checklist, not a single picker. Tick as many as you want (or "Select all") and click Add sources once.

• The active sequence is ticked by default, so a one-click add of just that one still works — untick it if you don't want it.
• Each ticked sequence is still sent and validated as its own separate upload — batching doesn't change the per-sequence size limit (see "Ingest failed (413)" below).
• If one sequence in the batch fails or is a duplicate, the rest still go through — you'll see a summary like "Added 3 · 1 already present".

Transcript columns sit in a scrollable area to the left, with the paper edit panel fixed on the right.

• Up to four columns open at once — use the Column button above the transcripts to add more.
• Each column has a source picker (switch or move it left/right) and a ⊗ icon to close it — this only removes it from your view, not the underlying source or your selections.

Click the source picker in the column header, then "Delete source" near the bottom.

• Different from closing a column — deleting removes the source and every block built from it, everywhere in the project, and can't be undone.
• You'll be asked to confirm first. Owners only.

Click the source picker in any column header, then Move left or Move right.

A local display change — it doesn't affect the underlying source order or any existing selections.

Yes — click the source name in the column header to edit it inline; Enter (or click away) to save.

Safe to rename — only changes the display label. All clip matching, timecodes, and exports use internal IDs, not the name.

Yes — in the Bridge panel, tick the sequence(s), choose the existing project in Send to, and click Add sources.

The new columns appear in the browser immediately — no files to export or upload.

Yes — export the review's comments as a CSV from Frame.io and import it against the matching source.

• Each comment lands at the right point in that source's transcript, whether or not the footage is in your paper edit yet — shown as a small note pinned to the word it was made on.
• Add that word to a block and the comment appears in the block's marker row too, becoming a real marker in Premiere on build.
• Re-uploading the same export later updates existing comments instead of duplicating them.

1. In Frame.io: review link → comments panel → export as CSV.
2. In Paper Edit App: click the source picker for the matching footage → "Import Frame.io comments…" → paste the CSV or pick the file.

Owners only.

Open Sequence Settings (next to Export) — a mismatch is flagged with an amber warning and lets you choose how the built sequence is set up.

• Match a source, or pick your own frame rate and resolution.
• Your choice is saved to the project and honoured by the Bridge — Premiere conforms off-rate clips to the chosen sequence rate.

Yes — select blocks from any column in any order.

The export cuts between sources according to your paper edit order, not the original source order.

Yes — the Bridge finds a multicam source sequence, even inside a bin folder, and places it in the built sequence.

• One manual step after building: select all clips in the new sequence, right-click → Multi-Camera → Enable.

This restores camera-switching — the timecodes and selections are already correct.

Yes — stereo pairs, dual-mono, or multi-channel recordings all place on the correct tracks in the built sequence, with video and audio correctly linked.

No — only transcribe one track per conversation.

• Transcribing every isolated mic (or both channels of the same stereo recording) sends the same conversation multiple times over, which does nothing for your paper edit but can trigger the upload size error on a long sequence.
• In Premiere, run Window → Text → Transcript → Transcribe Sequence only on the one clip (or stereo pair, one channel) with the cleanest recording of everyone — ideally with multi-speaker detection enabled, so each person still gets their own label.
• Leave other camera angles and isolated mic tracks untranscribed — the Bridge still places all of them correctly on build, regardless of which one was transcribed.

Rule of thumb: a single transcribed track of up to about 3 hours sends reliably. That budget is a total across every track transcribed, not per track — three isolated mics on the same 90-minute conversation adds up to 4.5 hours of transcribed audio, not 90 minutes.

Check the meta line in the column header, next to the duration.

• Names the source file directly (e.g. "INT_MicB.wav") when the whole column comes from one clip — the normal case for a single transcribed track.
• If a source is spliced from several takes or files, the header shows a count (e.g. "3 files") instead, with a subdued divider in the transcript at each point the clip changes.

The quickest way to confirm exactly which physical file to open in Premiere if you want to fix a diarisation error at the root instead of using Split.

They come through automatically from Premiere's transcription (Text panel → Transcript) — the Bridge sends them along with the words.

If Premiere has speakers wrong, rename, reassign, or merge them here (see below) without touching the Premiere transcript.

Each speaker in a multi-speaker transcript gets a distinct opacity of the source's colour — up to four speakers, lightest to strongest.

• Appears as the block background and on the speaker chip, so you can see who's speaking at a glance.
• Assigned in the order speakers first appear in the transcript. Works in both light and dark mode.

Click a speaker chip in the transcript header — the transcript shows only that speaker's segments, and search/AI search operate only within them.

Click the same chip again, or click All, to return to the full transcript.

Double-click a speaker chip — it becomes an editable field. Type the new name, Enter (or click away) to save.

• Updates across every transcript block and in the paper edit panel.

Non-destructive — stores a mapping on top of the original Premiere label, so you can rename again any time.

Right-click the speaker name above a paragraph to open the reassign menu.

• This is a per-paragraph override — it doesn't alter any other paragraph or the original Premiere labels.
• If the paragraph already has an override, the menu also offers "Reset to original".

Separate from right-clicking a speaker chip, which reassigns everyone with that label. Not available on a "Selects" column.

Yes — right-click the speaker name and choose "Split — reassign part of this paragraph…".

1. Drag across just the words that belong to the other speaker.
2. Pick who they should be reassigned to.

The paragraph splits around your selection — words before and after stay with the original speaker. If the range crosses more than one source clip, you'll be asked to keep it within a single clip.

Right-click a speaker chip to open the bulk reassign menu, then pick the speaker to move all those segments to.

• Useful when Premiere has split one person across multiple speaker labels.
• The merged-from speaker disappears from the header once no segments are left attributed to it.

Like individual reassignments, this is an override — it doesn't alter the original Premiere labels.

Open the project, click Share (top right), enter their email, click Invite.

They'll receive an email with a link to create their account and access the project.

They don't need the email — a notification appears in the bell icon in the top nav as soon as you share.

Clicking it automatically accepts the invite and takes them straight to the project.

They get the full paper edit workflow — viewing, selecting, reordering, labelling, and Topics/Refine.

• Edits sync live; they can leave comments and propose suggestions on any block.
• Can rename speakers, reassign segments, bulk-merge speakers, and save or load versions.
• AI search, topic mapping, and Refine work if they've saved their own AI key.
• Can click Sign off, Request review, or post a comment when ready.

They cannot add, rename, or delete sources, cannot export in any format, and cannot see any of your other projects.

Alongside your own on the main Projects page, under "Shared with you" — there's no separate dashboard.

Shows only projects shared with you; your editor's other projects aren't visible.

They click Sign off (final) or Request review (still want your input) — either way you get a notification in the bell.

Click it to open the project, review their changes, export the paper edit, and build via the Bridge panel.

Yes — a Leave Project button is available from within the shared project. Once used, they no longer have access.

Click Invite again for the same email address — sends a fresh link and invalidates the old one.

On your projects dashboard, click "Invite an editor" (or "Copy invite link" in the banner) to copy a personal invite URL.

• Anyone who follows it can sign up and their first project is free.
• The link is yours and multi-use — share it anywhere; it stays active until you generate a new one.

Yes — a Quick Guide overlay appears automatically the first time, covering how to build a paper edit, mark up blocks, undo/redo and live editing, working with a producer, and the Topics/Refine tabs.

Once dismissed it won't show again for that project, but it can be re-opened any time via the Quick Guide link above the Sources/Topics/Refine tabs.

In the Bridge panel, choose the project and click Build Paper Edit.

• Builds the sequence directly from the media already in your project — trimmed clips in order, video/audio linked, labels and notes placed as timeline markers.

Nothing is exported, uploaded, or re-imported.

Click Export → "Paper Edit JSON" to download a file describing your edit — block order, word ranges, timecodes, labels and notes.

A portable record of the edit: useful as a backup, for support requests, or for building your own tooling. The Bridge itself doesn't need it — it pulls your edit directly.

Single-source projects inherit everything from the source automatically.

• For multi-source projects, open Sequence Settings (next to Export) to match a source or pick your own frame rate and resolution.
• If sources differ, the dialog flags it with an amber warning — Premiere conforms off-rate clips to your chosen rate.

Your choice persists for the project and is honoured by the Bridge build.

Yes — Export → "Print / Save as PDF", then choose "Save as PDF" in your browser's print dialog.

• Only the paper edit panel prints — transcript columns, buttons, and controls are hidden.
• Each block shows its timecode and transcript text; labelled spans show their label.
• All spans expand for printing and return to their previous state when you close the dialog.

A "Made with Paper Edit App" footer is added at the end.

Yes — Premiere's scripting API doesn't allow a panel to enable multicam mode automatically.

1. Select all clips in the new sequence.
2. Right-click → Multi-Camera → Enable.

The Bridge places clips with correct timecodes first, then shows a tip in its log. Camera-switching is then fully active.

The Bridge re-finds clips through the source timeline they came from.

1. Check the source timeline still exists in the project (must appear in the panel's sequence picker).
2. If you renamed or replaced it since sending, re-send it as a source and re-select the affected blocks.
3. Retry the build.

The panel lives under Window → UXP Plugins → Paper Edit Bridge (not Window → Extensions).

• Confirm the install completed in Creative Cloud Desktop (requires Premiere Pro 25.0+), or unzip the .ccx and load its manifest.json with Adobe's free UXP Developer Tool — Add Plugin only accepts manifest.json, not the packaged .ccx itself.
• Restart Premiere Pro after installing.

Full instructions are on the Bridge page.

Go to Settings → Display Name.

Your name appears in notifications and shared project messages instead of your email address, so collaborators see 'Jane has shared…' rather than a raw email.

Go to Settings (link in the top nav) — you can change your password (requires your current one) and update your email address.

Go to Settings → AI API Keys and paste a key for any supported provider — Claude, OpenAI, or Gemini. You only need one.

• Multiple saved keys show a green Active dot on the one in use; click "Set as active" on any to switch.
• With no preference set, defaults to Claude first, then OpenAI, then Gemini.

Keys are never exposed to the browser. Gemini via Google AI Studio is free to start and takes about two minutes to set up.

Sign up at console.anthropic.com — separate from a normal claude.ai account — and generate a key.

1. Go to console.anthropic.com and sign up or log in.
2. Settings → Billing: add a payment method. No subscription — billed monthly for exactly what you use.
3. Settings → API Keys: click "Create Key" and name it (e.g. "Paper Edit App").
4. Copy the key — shown only once — and paste it into Paper Edit App's Settings → AI API Keys.

Most editors spend well under $1 a month on this app's AI features. Prefer not to add a card yet? Gemini via Google AI Studio has a genuinely free tier and works everywhere Claude does in this app.

You're billed by the provider directly, at their normal rates — no markup.

• AI search on a 500-block transcript: under half a US cent.
• A Refine pass over a 20-span paper edit: about $0.002.
• Generating a topic map: $0.01–0.03 per transcript.

A typical documentary project — regular AI search, a handful of topic maps, a few Refine passes — comes to well under $1 total, often just a few cents. No minimum spend, no wasted cost if you stop.

Click "New Folder" from your projects page to create one.

• Move a project into a folder using the three-dot menu on its card.
• Use the search box in the header to find projects by name across all folders.

Your drag started in one clip and ended in a different one.

Each block's timecodes are relative to a single clip's media, so spanning two clips would misplace it when the Bridge builds the sequence. Release and try again within one clip — clip boundaries appear as paragraph breaks in the transcript.

A network error prevented the last change from reaching the server — the UI has already rolled it back.

Check your connection and try again. If it keeps failing, reload the page — your last successfully saved state will load.

The server couldn't create the block from your selection — usually a transient network error.

Try the selection again. If it persists, reload the page.

Word removal (pencil / inline edit mode) failed to save — the block has been left unchanged.

Check your connection and try again.

An undo or redo couldn't reach the server.

Reload the page — your last saved state (including any collaborators' changes) loads from the database, and you can re-apply anything you wanted to keep.

A network error stopped a comment or suggestion from saving — nothing is lost on your side.

Check your connection and try again. If it keeps failing, reload the page; existing comments and suggestions reload from the server.

The block a suggestion targets has since been edited or deleted (often by a collaborator), so the proposed change no longer fits.

The suggestion is left open rather than applied. Review the block as it stands now and edit it directly, or reject the suggestion.

The Topics and Refine tabs require an AI API key (Claude, OpenAI, or Gemini).

Go to Settings → AI API Keys, paste a key, and mark it active. Gemini via Google AI Studio is free and takes about two minutes.

The AI call failed — most likely a revoked or out-of-credit key, a provider outage, or a rate limit.

Check your key in Settings, wait a moment, and try again. If the provider is down, switching to a different active key (e.g. Claude to OpenAI) works around it.

Premiere hasn't transcribed the clips in this sequence, or transcription hasn't completed yet.

1. In Premiere: Window → Text → Transcript.
2. Click Transcribe Sequence.
3. Wait for it to finish (words appear in the panel).
4. Click Send again in the Bridge panel.

Your plan allows a set number of projects.

Delete projects you no longer need from the dashboard, or upgrade your plan from Settings → Account.

413 means the sequence is too large in one go — almost always because more audio has been transcribed than actually needs to be.

• Size is driven by total minutes of transcribed audio, not sequence length — transcribing the same conversation on several tracks (isolated mics, both channels of a stereo pair, multiple camera angles) multiplies the data for no benefit.
• "Ignore Transcript" does not fix this — it only tells Premiere's caption/speech features to disregard that track; the transcript stays attached and the Bridge still sends it. Premiere has no way to detach or delete a transcript from a clip in place.

1. The only real fix for a track you don't want transcribed: re-import that media and replace the clip on the timeline with the fresh copy — this loses that clip's transcript entirely, so only do it to tracks you don't need (see "should I transcribe every track?" above).
2. As a rule of thumb, a single transcribed track of up to about 3 hours sends reliably — that budget is a total across every transcribed track in the sequence, not per track.
3. Re-send from the Bridge panel.

The Build button was clicked but no Premiere project is open.

Open the .pproj file that contains the media you sent, then click Build again.

The sequence was sent from a different Premiere project than the one currently open.

Clicking Build again proceeds, but some clips may be offline (shown in red) if the media isn't in the current project's bin. To avoid this, open the correct .pproj file first, then click Build.

Check the connection first, then the developer console for detail.

1. Check you're connected — the panel shows your name and a Disconnect button at the top.
2. If disconnected, reconnect using your Bridge key from Settings → Bridge Connection.
3. If connected but builds fail, open the UXP developer console: Premiere → Window → Extensions → UXP Developer Tools.
4. Look for [bridge] log lines — these show exactly where the build stopped and what Premiere returned.

Premiere only flags filler words when the transcript was created with filler-word detection turned on.

1. In Premiere: Window → Text → Transcript.
2. Click the ••• menu → Transcript view options…
3. Enable filler-word detection.
4. Re-transcribe the sequence.
5. Re-send from the Bridge panel.

A word-removal edit left a run containing only empty punctuation tokens from the Premiere transcript.

Delete the empty block using its trash icon — it can't be edited or filled in after the fact. Future edits will not produce empty blocks.