Workflow templates
A workflow is the recipe ReelBolt follows to turn your files into a video. A template is a ready-made workflow you can add to a project in one click and then run as-is or adjust.
ReelBolt ships sixteen templates in two families:
- Promo from code — reads the source code of a web app you upload and produces an animated promotional video about it. Four templates.
- Edit real footage — takes the video clips you upload and cuts, narrates, decorates and finishes them. Twelve templates, all built on the same three-step core: analyze the footage, decide what to keep, produce the edited file.
Every template in this page lists its display name, its key (the short name the assistant and the API use), whether it is added automatically, its steps in order, what it produces, and what you need to upload or set up before you run it.
Every template asks for your brief when you run it. Clicking Run opens a box where you say what video you want — the topic, who it is for, how long it should be and the format (for example "15 s vertical reel about the timeline editor"). The AI steps read it; a reel, short, story or TikTok request makes the promo templates produce a vertical 9:16 video. You can leave the box empty to let the AI decide.
How to add a template to a project
Templates are added to a project from the workflow builder:
- Open the project and choose New workflow (on the Overview panel or the Workflows tab).
- Under Start from a template, click a card. Each card shows the template's name and a one-line description of what you get and what you need to upload first.
- Click Use this template. ReelBolt creates the complete workflow straight away, named after the template with its version (for example "Highlight edit from footage v1").
The list in the builder shows the opt-in templates only. The one automatically created template,
Promo video from your code (quick-win-promo), is already added to every new project, so it does not
appear in that list.
You can also ask the assistant for a template. The assistant can look up any template by its
key and get its full, ready-to-propose list of steps (through its GetWorkflowTemplate tool), then
propose it to you as a new workflow or as changes to an existing one. Nothing is saved until you
review and apply the proposal. See the assistant.
After a template is added it is an ordinary workflow: you can rename it, change any step's settings, add or remove steps, and save. Changing it never changes the template itself.
Choosing a template: quick guide
Pick a template by what you start from and what you want to end up with:
| You have | You want | Template (key) |
|---|---|---|
| Web app source code | A clean promo video (silent unless you add narration) | Promo video from your code (quick-win-promo) |
| Web app source code, and you care about cost | The same promo with fewer AI tokens spent | Promo video from your code (lighter) (lean-context-promo) |
| Web app source code and a speech provider | A short narrated reel, vertical when you ask for one | Quick vertical reel (cinematic-feature-spotlight) |
| Web app source code and a speech provider | A narrated, story-shaped launch trailer | Narrated launch trailer (story-arc-launch-trailer) |
| Video clips (screen recordings, product shots, b-roll) and a speech provider | A narrated story written from your brief, with captions | Narrated product story (narrated-story) |
| A filmed video (talking head, vlog, interview) | A tightened edit with dead air removed | Highlight edit from footage (video-derush-edit) |
| A filmed video | The same, decided by a panel of AI editors | Highlight edit, decided by an editing team (video-derush-edit-room) |
| A filmed video | The edit plus titles, lower-thirds and callouts | Footage edit with titles and callouts (video-derush-edit-graphics) |
| A filmed video | The edit plus graphics, planned by a panel of AI artists | Footage edit with titles, designed by a graphics team (video-derush-edit-graphics-room) |
| A filmed video and music files you uploaded | The edit with a background music bed | Footage edit with background music (video-derush-edit-music) |
| A filmed video and short sound files you uploaded | The edit with whooshes, clicks and stingers | Footage edit with sound effects (video-derush-edit-sfx) |
| A filmed video | The edit with a colour grade, chosen by a panel | Footage edit with a colour grade (video-derush-edit-grade-room) |
| A filmed video and a paid video-generation account | The edit with AI-generated b-roll | Footage edit with AI-generated b-roll (video-derush-edit-broll) |
| A filmed video and a speech provider | The edit with narration and captions written after the cut | Footage edit with narration and captions (video-derush-edit-voiceover) |
| Talking-head footage with speech, a speech provider and a consented cloned voice | The edit with flubbed sentences re-recorded in the presenter's voice | Fix flubbed lines in a talking-head video (video-derush-edit-pickups) |
| Footage with speech and a speech provider | The edit dubbed into another language, with subtitles | Dubbed video in another language (video-derush-edit-dub) |
There is no built-in music or sound-effects library yet: the music and sound-effects templates choose only from audio files you upload to the project first. With none uploaded, the result has no music or effects, and the run says so.
There is no dedicated template for tracked screen inserts (putting a rendered animation onto a green screen in your footage, such as a phone screen). You build that workflow yourself in four steps, and the assistant can draft it for you; see Tracked screen insert recipe.
Four templates add narration to footage (narrated-story, video-derush-edit-voiceover,
video-derush-edit-pickups and video-derush-edit-dub), and two promo templates are narrated
(cinematic-feature-spotlight and story-arc-launch-trailer). The other two promo templates are
silent until you insert a Voiceover step; see Voiceover recipe.
Promo video from your code (quick-win-promo)
Promo video from your code is the starter template. It is the only template added
automatically to every new project. When you run it, it asks for your brief (audience, message,
length, format); leave it empty to let the AI decide. The video is silent — the only sound a
promo can carry is narration, and this template has no narration step. The finished video's notes
say so; insert a Voiceover step after step 8 to narrate it (see
Voiceover recipe).
What it makes: an animated promotional video about a web app, built with Remotion (the code-based animation tool ReelBolt uses to draw and render promo videos). The finished video appears as the output of step 10.
What it needs: the web app's source code uploaded as project files, and a working chat AI provider.
| # | Step label | Step type | Agent |
|---|---|---|---|
| 1 | Analyze code structure | Agent | CodeStructureAnalyzer |
| 2 | Analyze UI dependencies | Agent | DependencyAnalyzer |
| 3 | Inventory components | Agent | ComponentInventoryAnalyzer |
| 4 | Map routes and API | Agent | RouteAndApiAnalyzer |
| 5 | Extract style tokens | Agent | StyleAndThemeExtractor |
| 6 | Translate to Remotion scenes | Agent | RemotionComponentTranslator |
| 7 | Create animation strategy | Agent | AnimationStrategyAgent |
| 8 | Draft narration and captions | Agent | ScriptwriterAgent |
| 9 | Build shot direction | Agent | DirectorAgent |
| 10 | Assemble and render master composition | Agent | AuthorAgent |
| 11 | Review quality and compliance | ReviewLoop | ReviewAgent |
Step 11 scores the result out of 10. Below 8 it sends the work back to step 10 for another try, at most 2 review rounds in total.
Choose it when you want a reliable first promo with no setup. See Promo from code recipe.
Promo video from your code, lighter (lean-context-promo)
Promo video from your code (lighter) is an opt-in version of the starter promo that spends fewer AI tokens. Like the starter, it asks for your brief and is silent unless you add narration. It adds one non-AI step that trims the list of app components down to the essentials (at most 40 components, each reduced to name, file and purpose) before later steps read it, and it gives each later step only the earlier results it actually needs.
What it makes: the same kind of animated promo as the starter, rendered by step 11.
What it needs: the web app's source code uploaded as project files.
| # | Step label | Step type | Agent |
|---|---|---|---|
| 1 | Analyze code structure | Agent | CodeStructureAnalyzer |
| 2 | Analyze UI dependencies | Agent | DependencyAnalyzer |
| 3 | Inventory components | Agent | ComponentInventoryAnalyzer |
| 4 | Map routes and API | Agent | RouteAndApiAnalyzer |
| 5 | Extract style tokens | Agent | StyleAndThemeExtractor |
| 6 | Reduce component inventory to a lean view | Extract | ExtractTransform (not an AI) |
| 7 | Translate to Remotion scenes (lean context) | Agent | RemotionComponentTranslator |
| 8 | Create animation strategy (lean context) | Agent | AnimationStrategyAgent |
| 9 | Draft narration and captions | Agent | ScriptwriterAgent |
| 10 | Build shot direction (lean context) | Agent | DirectorAgent |
| 11 | Assemble and render master composition (lean context) | Agent | AuthorAgent |
| 12 | Review quality and compliance | ReviewLoop | ReviewAgent |
Step 12 sends work back to step 11 when the score is below 8, at most 2 rounds.
Choose it when your app is large and the starter promo is slow or expensive.
Quick vertical reel (cinematic-feature-spotlight)
Quick vertical reel is a short, punchy, narrated feature spotlight for social media, with reveal transitions and a strong call-to-action ending. It asks you for a brief before each run (for example "15 s vertical reel about the new dashboard, end with a sign-up call to action"), and every AI step sees it. Asking for a reel, short, story, TikTok or vertical video makes a 1080x1920 (9:16) video; "square" makes 1080x1080; otherwise it is landscape 1920x1080. A stated length is the video's length.
What it makes: a narrated animated promo rendered by step 9. The narration is the script's own voiceover, recorded by step 7 before the director times the shots to it.
What it needs: the web app's source code uploaded as project files, a speech provider set up by an admin (the run fails at step 7 without one), and a short brief typed when you run it.
| # | Step label | Step type | Agent |
|---|---|---|---|
| 1 | Analyze cinematic candidate surfaces | Agent | CodeStructureAnalyzer |
| 2 | Find hero components | Agent | ComponentInventoryAnalyzer |
| 3 | Extract brand visuals | Agent | StyleAndThemeExtractor |
| 4 | Translate hero scenes | Agent | RemotionComponentTranslator |
| 5 | Plan cinematic transitions | Agent | AnimationStrategyAgent |
| 6 | Write emotional feature hooks | Agent | ScriptwriterAgent |
| 7 | Record the narration ("kind": "ScriptScenes") | Voiceover | VideoTransform (not an AI; calls the speech provider) |
| 8 | Define reveal sequence | Agent | DirectorAgent |
| 9 | Assemble cinematic master cut | Agent | AuthorAgent |
| 10 | Cinematic quality review | ReviewLoop | ReviewAgent |
Step 10 sends work back to step 9 when the score is below 9, at most 3 rounds.
Narrated launch trailer (story-arc-launch-trailer)
Narrated launch trailer is a promo template shaped like a narrative trailer: set-up, rising tension, a payoff reveal and a final launch call to action, spoken over the picture. It asks you for a brief before each run; format and length follow the brief exactly as in the quick vertical reel above.
What it makes: a narrated animated trailer rendered by step 9.
What it needs: the web app's source code uploaded as project files, a speech provider set up by an admin, and a short brief typed when you run it.
| # | Step label | Step type | Agent |
|---|---|---|---|
| 1 | Map narrative assets | Agent | CodeStructureAnalyzer |
| 2 | Extract user journey flow | Agent | RouteAndApiAnalyzer |
| 3 | Extract mood and tone tokens | Agent | StyleAndThemeExtractor |
| 4 | Translate story scenes | Agent | RemotionComponentTranslator |
| 5 | Write narrative arc script | Agent | ScriptwriterAgent |
| 6 | Record the narration ("kind": "ScriptScenes") | Voiceover | VideoTransform (not an AI; calls the speech provider) |
| 7 | Create trailer shotlist | Agent | DirectorAgent |
| 8 | Plan suspense pacing | Agent | AnimationStrategyAgent |
| 9 | Assemble and render trailer | Agent | AuthorAgent |
| 10 | Trailer impact review | ReviewLoop | ReviewAgent |
Step 10 sends work back to step 9 when the score is below 9, at most 3 rounds.
What every promo template does with your copy
The four promo templates share three rules, so the video reads well on the screen it is made for:
- Plain text on screen. Words are drawn exactly as written, so the AI never writes markdown
such as
*word*. Any that slips into the script is removed automatically, and a render whose declared on-screen text still carries it is refused before it starts. - Readable at phone size. Body text is at least about 48 pixels tall on a 1080-wide frame, and a vertical video zooms into the part of your app's screen that matters instead of shrinking the whole desktop layout.
- Nothing invented. Names, quotes, ratings, prices and numbers come only from your code, your files or your brief. A screen that greets a user says "Welcome back", never a made-up name.
Highlight edit from footage (video-derush-edit)
Highlight edit from footage is the basic footage-editing template. "Derushing" means going through raw footage and keeping only the good parts. It finds pauses, shot changes and (optionally) what is said, lets an AI editor choose which parts to keep, and then produces the edited video file.
What it makes: an edited MP4 (edited.mp4), produced by step 3. The edited video is also added
to the project's files so you can edit it again later.
What it needs:
- A video file uploaded to the project, selected as the source in step 1 (the template leaves the file empty on purpose; the step shows "no file selected" until you pick one).
- For the editor to understand speech, a transcription provider set up by an admin (the
bundled local
whisperservice works). Without one, the analysis still runs but the editor only sees pauses and shot changes.
| # | Step label | Step type | Agent |
|---|---|---|---|
| 1 | Analyze source video | VideoAnalyze | VideoTransform (not an AI) |
| 2 | Decide which spans to keep | Agent | VideoStoryEditor |
| 3 | Compile edited video | VideoCompile | VideoTransform (not an AI) |
| 4 | Review edit quality | ReviewLoop | VideoReviewAgent |
The compile step uses gentle automatic transitions between cuts ("transitionPolicy": "Auto"), a
short fade in and out at the start and end, and drops any kept piece shorter than 0.8 seconds.
Step 4 sends work back to step 2 when the score is below 8, at most 3 rounds.
Choose it when you have a talking-head video, interview or vlog with dead air, false starts or retakes. See Derush edit recipe.
Highlight edit, decided by an editing team (video-derush-edit-room)
Highlight edit, decided by an editing team works like the highlight edit, but the decision about what to keep is made by a small panel of AI editors instead of one. Three editor "seats" (pacing, story and craft) discuss the footage in a short group conversation, and a director then writes one final decision. If the panel fails for any reason, a single AI editor makes the decision instead.
What it makes: an edited MP4 produced by step 3.
What it needs: a video file selected as the source in step 1; a transcription provider is recommended.
| # | Step label | Step type | Agent |
|---|---|---|---|
| 1 | Analyze source video | VideoAnalyze | VideoTransform (not an AI) |
| 2 | Edit room deliberation | EditRoom | panel: three editor seats plus VideoEditDirector |
| 3 | Compile edited video | VideoCompile | VideoTransform (not an AI) |
| 4 | Review edit quality | ReviewLoop | VideoReviewAgent |
Step 4 sends work back to step 2 when the score is below 8, at most 3 rounds. The panel costs more AI calls than a single editor.
Choose it when the edit needs judgement (a story to tell, several takes to choose between) and you can afford a slower, more thorough decision.
Footage edit with titles and callouts (video-derush-edit-graphics)
Footage edit with titles and callouts adds on-screen graphics (titles, lower-thirds, callouts) to the basic edit. The analysis step also finds good, uncluttered places on screen for graphics, and a motion-graphics AI chooses where to put graphics and designs them. By default it designs each graphic as a real animation that it renders itself with Remotion; plain text on a box is the fallback.
What it makes: an edited MP4 with graphics, produced by step 4.
What it needs: a video file selected as the source in step 1; a transcription provider is recommended so graphics match what is being said.
| # | Step label | Step type | Agent |
|---|---|---|---|
| 1 | Analyze source video ("emitOverlayPlacements": true) | VideoAnalyze | VideoTransform (not an AI) |
| 2 | Decide which spans to keep | Agent | VideoStoryEditor |
| 3 | Plan motion graphics overlays | Agent | MotionGraphicsPlanner |
| 4 | Compile edited video with graphics ("enableGraphics": true) | VideoCompile | VideoTransform (not an AI) |
| 5 | Review edit quality | ReviewLoop | VideoReviewAgent |
Step 5 sends work back to step 2 when the score is below 8, at most 3 rounds. Step 3 can take a long time because it writes and renders small animations.
This template is also the easiest starting point for a tracked screen insert: turn on insert detection in step 1 and inserts in step 4. See Tracked screen insert recipe and Motion graphics recipe.
Footage edit with titles, designed by a graphics team (video-derush-edit-graphics-room)
Footage edit with titles, designed by a graphics team replaces the single motion-graphics AI with a panel of AI artists: a layout artist, a timing artist, a copy artist and an insert artist discuss the options, then a lead director writes one final graphics plan and renders the graphics. If the panel fails, a single motion-graphics AI plans instead.
What it makes: an edited MP4 with graphics, produced by step 4.
What it needs: a video file selected as the source in step 1.
| # | Step label | Step type | Agent |
|---|---|---|---|
| 1 | Analyze source video ("emitOverlayPlacements": true) | VideoAnalyze | VideoTransform (not an AI) |
| 2 | Decide which spans to keep | Agent | VideoStoryEditor |
| 3 | Graphics room deliberation (reads step 1) | GraphicsRoom | panel: four artist seats plus MotionGraphicsDirector |
| 4 | Compile edited video with graphics ("enableGraphics": true) | VideoCompile | VideoTransform (not an AI) |
| 5 | Review edit quality | ReviewLoop | VideoReviewAgent |
Step 5 sends work back to step 2 when the score is below 8, at most 3 rounds.
Footage edit with background music (video-derush-edit-music)
Footage edit with background music adds a background music bed to the basic edit. There is no built-in music library yet: upload a music track to the project's files first. Every audio file you uploaded to the project is offered as a music choice, a music-supervisor AI picks at most one track and how loud and how strongly it should dip under speech, and the compile step mixes it in.
What it makes: an edited MP4 with music, produced by step 4. With no audio files uploaded the video has no music, and the music step's reason says so in plain words.
What it needs: a video file selected as the source in step 1, and one or more music files (audio files) uploaded to the project.
| # | Step label | Step type | Agent |
|---|---|---|---|
| 1 | Analyze source video ("offerMusicTracks": true) | VideoAnalyze | VideoTransform (not an AI) |
| 2 | Decide which spans to keep | Agent | VideoStoryEditor |
| 3 | Pick background music | Agent | MusicSupervisor |
| 4 | Compile edited video with music ("enableMusic": true) | VideoCompile | VideoTransform (not an AI) |
| 5 | Review edit quality | ReviewLoop | VideoReviewAgent |
Step 5 sends work back to step 2 when the score is below 8, at most 3 rounds. See Music, sound effects and colour grade recipe.
Footage edit with sound effects (video-derush-edit-sfx)
Footage edit with sound effects adds short sound effects (whooshes, clicks, stingers) at chosen moments in the edit. There is no built-in sound-effects library yet: upload your effects to the project's files first. Every audio file in the project is offered as a possible effect, and a sound-designer AI decides which effect plays when, and how loud. Choosing no effects at all is a valid result.
What it makes: an edited MP4 with sound effects, produced by step 4.
What it needs: a video file selected as the source in step 1, and short sound-effect files
uploaded to the project (name them clearly, for example whoosh.wav; the AI chooses by file name).
| # | Step label | Step type | Agent |
|---|---|---|---|
| 1 | Analyze source video ("offerSfxClips": true) | VideoAnalyze | VideoTransform (not an AI) |
| 2 | Decide which spans to keep | Agent | VideoStoryEditor |
| 3 | Plan sound-effect cues | Agent | SoundDesigner |
| 4 | Compile edited video with sound effects ("enableSfx": true) | VideoCompile | VideoTransform (not an AI) |
| 5 | Review edit quality | ReviewLoop | VideoReviewAgent |
Step 5 sends work back to step 2 when the score is below 8, at most 3 rounds.
Footage edit with a colour grade (video-derush-edit-grade-room)
Footage edit with a colour grade adds a colour grade (an overall "look", such as warm, cool, filmic or muted) to the edit. A panel of AI colourists discusses the measured colour of each shot, then a supervising colourist picks one look for the whole video, as words only (for example "Warm, Subtle"). ReelBolt turns those words into the actual colour adjustment. "None" (no grade) is a valid result.
What it makes: an edited, colour-graded MP4 produced by step 4.
What it needs: a video file selected as the source in step 1.
| # | Step label | Step type | Agent |
|---|---|---|---|
| 1 | Analyze source video | VideoAnalyze | VideoTransform (not an AI) |
| 2 | Decide which spans to keep | Agent | VideoStoryEditor |
| 3 | Color grade room deliberation (reads step 1) | ColorGradeRoom | panel: three colourist seats plus ColorGradeDirector |
| 4 | Compile edited video with grade ("enableColorGrade": true) | VideoCompile | VideoTransform (not an AI) |
| 5 | Review edit quality | ReviewLoop | VideoReviewAgent |
Step 5 sends work back to step 2 when the score is below 8, at most 3 rounds.
Footage edit with AI-generated b-roll (video-derush-edit-broll)
Footage edit with AI-generated b-roll adds AI-generated video clips to the edit: cutaways over speech, a cold open at the start, or an end card. A shot-director AI plans the clips, a generation step pays a video-generation service to make them (with a hard spending cap), and the compile step places them.
What it makes: an edited MP4 with generated clips, produced by step 5.
What it needs:
- A video file selected as the source in step 1.
- A video-generation provider (MiniMax) set up by an admin. This costs real money.
- Spending limits: this template's generation step allows at most 4 clips and at most 5 US
dollars per run (
"maxClips": 4,"maxSpendUsd": 5.0), and admins set daily limits on top.
| # | Step label | Step type | Agent |
|---|---|---|---|
| 1 | Analyze source video | VideoAnalyze | VideoTransform (not an AI) |
| 2 | Decide which spans to keep | Agent | VideoStoryEditor |
| 3 | Plan generated b-roll shots | Agent | ShotDirector |
| 4 | Generate planned b-roll clips | VideoGenerate | VideoTransform (not an AI; calls the paid provider) |
| 5 | Compile edited video with generated clips ("enableGeneratedClips": true) | VideoCompile | VideoTransform (not an AI) |
| 6 | Review edit quality | ReviewLoop | VideoReviewAgent |
Step 6 sends work back to step 2 when the score is below 8, at most 3 rounds. When you run or retry this workflow, ReelBolt asks whether to reuse previously generated clips (free) or regenerate them (costs money). See Generated b-roll recipe.
Footage edit with narration and captions (video-derush-edit-voiceover)
Footage edit with narration and captions adds spoken narration written by an AI after the
cut, with captions burned in. It asks for your brief (topic, tone, length) when you run it. A
narration-writer agent writes zero or more lines, each tied to a shot, pause or spoken phrase of the edit (never a
time), a Voiceover step turns them into speech and works out where each line starts, and the
compile step mixes them over the original sound.
What it makes: an edited MP4 with narration and karaoke-style captions, produced by step 5.
What it needs:
- A video file selected as the source in step 1.
- A speech provider (Fish Audio, or a local OpenAI-compatible speech server) set up by an
admin, and a voice id entered as
defaultVoiceIdin step 4 (or a voice from the project's Voices tab asprojectVoiceId). - A transcription provider for the captions: karaoke captions follow each spoken word, and
without word timings a line gets no caption (the compile step reports
no_word_alignment). Switch step 5 to"captionStyle": "Subtitle"if you have no transcription provider.
| # | Step label | Step type | Agent |
|---|---|---|---|
| 1 | Analyze source video | VideoAnalyze | VideoTransform (not an AI) |
| 2 | Decide which spans to keep | Agent | VideoStoryEditor |
| 3 | Write narration lines | Agent | NarrationWriter |
| 4 | Synthesize narration ("kind": "NarrationPlan") | Voiceover | VideoTransform (not an AI; calls the speech provider) |
| 5 | Compile edited video with narration ("enableVoiceover": true, "enableCaptions": true) | VideoCompile | VideoTransform (not an AI) |
| 6 | Review edit quality | ReviewLoop | VideoReviewAgent |
Step 6 sends work back to step 2 when the score is below 8, at most 3 rounds. A line that does not
fit in its moment is reported, and the review asks the narration writer for shorter words; nothing
re-cuts the picture to the voice. Captions are on ("captionStyle": "Karaoke"); set
"enableCaptions": false in step 5 to turn them off. Because the editor cuts first and the writer
follows, the narration describes the kept clips; for a story written first, use Narrated product
story below. See Voiceover recipe.
Fix flubbed lines in a talking-head video (video-derush-edit-pickups)
This template needs talking-head footage with clear speech. Fix flubbed lines in a
talking-head video re-records the sentences the presenter flubbed, in the
presenter's own cloned voice. A pickup-planner agent names the spoken phrases to replace and the
corrected sentence for each, a Voiceover step speaks them, and the compile step mutes the
original take for exactly that phrase and plays the pickup there. The picture is untouched, so the
mouth is still the original take's: lip sync is not available, and the result says so.
What it makes: an edited MP4 with pickups, produced by step 5.
What it needs:
- A video file selected as the source in step 1, and a transcription provider, because
pickups replace spoken phrases (
t0,t1, ...) and there are none without a transcript. - A cloned voice with recorded consent in the project's Voices tab (needs a Fish Audio
provider). Enter it as
projectVoiceIdin step 4; a voice whose consent is missing, revoked or expired fails the step instead of using another voice.
| # | Step label | Step type | Agent |
|---|---|---|---|
| 1 | Analyze source video | VideoAnalyze | VideoTransform (not an AI) |
| 2 | Decide which spans to keep | Agent | VideoStoryEditor |
| 3 | Plan pickup lines | Agent | PickupPlanner |
| 4 | Record pickups ("kind": "PickupPlan") | Voiceover | VideoTransform (not an AI; calls the speech provider) |
| 5 | Compile edited video with pickups ("enablePickups": true) | VideoCompile | VideoTransform (not an AI) |
| 6 | Review edit quality | ReviewLoop | VideoReviewAgent |
Step 6 sends work back to step 2 when the score is below 8, at most 3 rounds.
Dubbed video in another language (video-derush-edit-dub)
This template needs footage with speech. Dubbed video in another language makes a version of the edit in another language. The narration
writer writes the lines, a translator agent rewrites the same lines in the language you name when
you run the workflow, a Voiceover step speaks the translation, and the compile step replaces
the original speech under every line and burns subtitles of the words actually spoken. The
picture is untouched: lip sync is not available, and the result says so.
What it makes: a dubbed, subtitled MP4, produced by step 6.
What it needs:
- A video file selected as the source in step 1, and a speech provider.
- Text when you run it: the target language (for example "Spanish"), in the request box.
| # | Step label | Step type | Agent |
|---|---|---|---|
| 1 | Analyze source video | VideoAnalyze | VideoTransform (not an AI) |
| 2 | Decide which spans to keep | Agent | VideoStoryEditor |
| 3 | Write narration lines | Agent | NarrationWriter |
| 4 | Translate narration | Agent | NarrationTranslator |
| 5 | Synthesize dub ("kind": "NarrationPlan", pointing at step 4) | Voiceover | VideoTransform (not an AI; calls the speech provider) |
| 6 | Compile dubbed video ("voiceoverMode": "ReplaceDialogue", "enableCaptions": true) | VideoCompile | VideoTransform (not an AI) |
| 7 | Review edit quality | ReviewLoop | VideoReviewAgent |
Step 7 sends work back to step 2 when the score is below 8, at most 3 rounds.
Narrated product story (narrated-story)
Narrated product story writes the story first and then cuts your clips to it. A narration writer reads your brief and what the analysis saw in each clip, and writes a short script with a beginning, a middle and an end, tying each line to the clip it should play over. The editor then keeps those clips in the script's order, the narration is recorded, and the compile step mixes it in with karaoke captions. It asks for your brief when you run it: what the video is about, who it is for, the tone and how long it should be.
What it makes: a narrated, captioned MP4 produced by step 5.
What it needs:
- Your video clips (screen recordings, product shots, b-roll), uploaded and selected as the sources in step 1. Clips with no sound of their own are fine.
- A speech provider set up by an admin, and a transcription provider for the karaoke captions (see the narration template above).
- Optional: a vision provider, so the analysis can describe what each clip shows and the writer can pick the right clip for each line. Without one the run still works.
- The output format: step 5 keeps your clips' own shape by default. Choose vertical, square or landscape there before you run it.
| # | Step label | Step type | Agent |
|---|---|---|---|
| 1 | Analyze your clips ("vision": "Optional") | VideoAnalyze | VideoTransform (not an AI) |
| 2 | Write the story | Agent | NarrationWriter |
| 3 | Choose and order clips for the story | Agent | VideoStoryEditor |
| 4 | Record the narration ("kind": "NarrationPlan", pointing at step 2) | Voiceover | VideoTransform (not an AI; calls the speech provider) |
| 5 | Compile the narrated video ("enableVoiceover": true, "enableCaptions": true) | VideoCompile | VideoTransform (not an AI) |
| 6 | Review the finished video | ReviewLoop | VideoReviewAgent |
Step 6 sends work back to step 2 when the score is below 8, at most 3 rounds, so a weak result gets a rewritten story as well as a new cut. Clips play in the script's order across different clips; two moments of the same clip always play in that clip's own order.
What every footage template has in common
All twelve footage templates (video-derush-edit, its ten variants and narrated-story) share
these facts:
- Step 1 must point at your video. It is set to a project file with no file chosen, so a run
fails straight away with
SOURCE_UNRESOLVEDuntil you open step 1 and select the uploaded video. - Later steps point back at earlier steps by number. For example the compile step in the music template reads the editing decision from step 2 and the music choice from step 3. If you insert or remove steps, check these numbers (see Step types).
- The AI never writes times or positions. The analysis step gives every shot, pause, spoken
phrase, graphics spot and green-screen region a short id (such as
s3,g1,t7,p2,r0), and the AI agents only choose among those ids. ReelBolt itself works out the exact frames. This is why the AI cannot place something at a moment that does not exist. - The review step checks the result and sends the work back to step 2 (the editing decision,
or the story in
narrated-story) when the score is too low, up to 3 rounds; after that the run finishes with the best result it has. - They ask for your brief before running. Type instructions for the AI editor (for example "keep it under 60 seconds"), or leave the box empty. To make the brief optional, open the workflow, switch off Require a brief before each run and save.