Skill instructions
create-short-clips4.85 KB
View saved version →
---
name: create-short-clips
description: "Clipping: turn webinars, interviews, demos or uploaded recordings into captioned business clips; review moments, choose, repair or export clips with BlitzReels."
---
# Clipping
Repurpose the identified recording into complete, source-grounded moments for the user's business.
## 1. Resolve source and context
- Use the selected asset, `media_upload_file` for a ChatGPT attachment, `media_list` for workspace media,
or the user's supported URL. Use `media_import_inspect` to verify uncertain URL support and source metadata.
- Read `workspace_context_get` for the explicit workspace. Use `workspaces_list` only when selection is unresolved.
- Reuse reviewed audience, offer, purpose, language, protected and avoided terms, captions and Series; ask only for missing material choices.
- For a social URL with multiple audio tracks, use the language from `media_import_inspect` as selectedAudioLanguage.
Null preserves the default track; direct files and existing assets preserve source audio.
- State the source, duration when known, intended clip count, duration bounds, platform, purpose and export constraints.
- For a short recording the user wants preserved, route to edit-and-export-video; clipping selects excerpts.
- Treat each independent source as its own batch. Reuse approved settings and retain every source-to-batch mapping.
Completion: identified sources, authorized workspace and resolved business outcome.
## 2. Review moments and create
- When the source has candidates, read `clips_candidates_list`; use `media_transcript_get` for needed context.
- Prefer complete explanations, examples, demonstrations, proof or relevant objections for the requested audience.
- Show candidate IDs, source ranges, factual hooks and a brief reason. Distinguish the spoken excerpt from draft copy.
- Apply topic instructions by selecting matching verified candidate IDs; show when no matching candidate exists.
- Call `clips_create` with one asset or URL, clipCount, durationBounds and the chosen captionThemeId.
- Use selection.includeIds for approved candidates; use selection.excludeIds for rejected candidates.
- Omitted captions inherit reviewed defaults. Preserve the chosen Series; use null for explicitly standalone clips.
- Keep createSharePage false unless public sharing is explicitly requested.
- Create once with a stable idempotencyKey. Retry identical inputs with the same key; inspect after ambiguous responses.
Completion: one accepted batch per source with verified identifiers and requested settings.
## 3. Follow and revise
- Read batch.polling and follow `clips_status_get` within its delay and attempt/time limits.
- Show the batch with `clips_get` when playback is requested; status polling keeps the current review widget mounted.
- Deliver ready clips incrementally. Processing or QA warnings alone do not justify recreating accepted clips.
- Use `clips_list` and `clips_inspect` for bounded state; use `clips_reselect`, `clips_repair`, `clips_export`
or `captions_look_apply` for the identified clip and explicit requested change.
- Selected widget clip/time context identifies the target; route precise project edits to edit-and-export-video.
- Use `clips_recover` only for a terminal batch's missing count; retain accepted clips and rejected moments.
Completion: every clip accounted for as ready, active, requiring review or failed.
## 4. Deliver and reuse
- Present inline playback and ready MP4s or the ZIP. Refresh `clips_status_get` when signed links expire.
- State the actual quality/watermark and any unresolved framing, caption or audio issues.
- Save user-requested ratings with `video_feedback_save` for the exact completed export; preserve accepted output.
- Draft titles, covers and platform copy from the verified source; keep invented statistics and claims out.
- For LinkedIn posts or carousels, reuse approved source excerpts; use the existing carousel tools to render/export.
- Organize repeat
Preview of the first 4,000 characters. Read full instructions →
edit-and-export-video3.96 KB
View saved version →
---
name: edit-and-export-video
description: "Editing: polish a short recording or revise a BlitzReels video's captions, framing, text, media or sound; preview changes, undo edits and export finished files."
---
# Editing
Resolve the selected clip and time, preview the precise change, apply once and deliver a verified export.
## 1. Inspect
- Use the selected clip's projectId and timeSeconds when provided by the review widget.
- Otherwise resolve the project with `projects_get` or bounded `projects_list`.
- Read `editor_snapshot_get` for current revision and stable item IDs; use `projects_inspect` for needed details.
- For an uploaded short recording, upload/resolve the source, create the project if needed, then use
`editor_assets_insert` to preserve the complete recording. Use `captions_source_prepare` after transcription completes; an explicit no-speech result needs no captions.
- Read `workspace_context_get` and reuse reviewed captions, brand, language and protected names.
- For synchronized camera/screen sources, create a recording set, attach each known camera/screen asset,
finalize, then inspect alignment and warnings before `recording_sets_clips_create`.
Existing access is required; separate audio-only tracks are unsupported.
Completion: one authorized project, current revision and explicit target IDs/time ranges.
## 2. Preview the focused change
- Use dryRun=true with `editor_asset_replace`, `editor_audio_set`, `editor_text_set`, `editor_crop_set`,
`editor_assets_insert`, `editor_items_move` or `editor_items_remove` as appropriate.
- Resolve every target from the current snapshot; include expectedRevision and a stable idempotencyKey.
- For trimming, use `timeline_trim_preview`, then `timeline_trim_apply`.
- For splitting an item, use `timeline_split_preview`, then `timeline_split_apply`.
- For caption corrections, read `captions_words_list`, then use `captions_words_correct` with verified word IDs.
This changes written captions and preserves timing; it does not change the spoken audio.
- For caption style, use `captions_look_apply`; use actual workspace logo/media assets for branding.
- Use the existing focused layout or visual tools for supported layouts and safe-zone changes.
- Keep unaffected clips and accepted edits. State any linked captions or items included in a removal.
Completion: valid preview or an unambiguous user-requested change covering every affected target.
## 3. Apply and verify
- Apply the previewed focused editor change with dryRun=false and the same expectedRevision.
- Read the updated snapshot; compare the requested target, timing and unaffected neighboring items.
- A stale revision requires a new snapshot and preview before mutation.
- After an ambiguous response, inspect state and retry only identical inputs with the original key.
- For undo, read `editor_history_list` and call `editor_edit_undo` on the identified action at the current revision.
- For a generated shot replacement, generate the requested asset once, verify it, then replace only the selected item.
Completion: every requested edit verified by current state and mutation receipts.
## 4. Review and export
- Review the relevant before/after media when available; distinguish timeline-state proof from rendered playback.
- Verify spelling, complete endings, framing, readable demos, sound and watermark before calling a video ready.
- Start `exports_start` for the user-requested format and entitled resolution. Follow `exports_list` to terminal state.
- Return only ready download URLs; read export state again to renew an expired link.
- For additional platform versions, preserve the approved base project; create/export explicit requested variants.
- A download link or download request is not proof that the user transferred or watched the file.
Completion: ready export and usable link, or explicit active/blocked state with next action.
## Account access
Use existing entitlements. Explain current quality, watermark, cr
Preview of the first 4,000 characters. Read full instructions →
generate-video-assets6.22 KB
View saved version →
---
name: generate-video-assets
description: "Generation: replace characters in a source video, transfer motion, create a complete video from an idea or script, or generate images, clips, voiceovers, music or sound effects with BlitzReels."
---
# Generation
Choose one generation branch, confirm material cost or likeness choices, and return verified results.
## 1. Route the request
- Preserve the user's prompt, script, narration, and aspect ratio.
- Call `generation_options_list` when model, duration, voice, style, or aspect ratio remains unresolved.
- Ask one focused question when a missing choice changes the output or cost.
- For a reusable visual system, inspect it with `story_kits_list` and `story_kits_get`.
- For an existing character, inspect it with `characters_list` and `characters_get`; call `characters_apply` after
the user confirms the rights and likeness consent required by the tool.
Completion: one supported branch with every material input resolved.
## Source-video editing branch
- Inspect the selected source video and identity photos; retain its duration, soundtrack and total people count.
- Call `generation_plan_brief` with the user's `rawPrompt` and source duration, then inspect `brief.execution`.
Transformation execution submits an asset job; its empty story modes do not require `generation_plan_validate`.
- Call `generation_video_create` with the chosen model, one processed `referenceVideoAssetIds` source and
`transformation.characters` mapping each selected `sourceRole` to its `frontalAssetId` and `referenceAssetIds`.
Use model-specific target-image inputs when the returned execution schema requires them.
- Preserve unselected people and source choreography, gestures, camera and cuts. Use actual assets for logos.
Set `transformation.keepOriginalAudio: true` to restore the source music; `generateAudio` is a separate control.
- Check live people/duration limits and cost. Submit within the user's authorization and retain the job receipt.
- Compare output against the source for identity, motion, background and music synchronization before acceptance.
Exact likeness and timing are not guaranteed. A provider rejection ends the attempt without an automatic retry.
Completion: a terminal, playable result with verification findings; workflow attachment follows user validation.
## 2. Complete-video branch
- Read `workspace_context_get`; reuse reviewed business facts, tone, purpose, avoided terms and Series or story-kit references.
Apply reviewed pronunciation notes when drafting narration; source speech remains unchanged.
- Approve the factual script and constraints first: aspect ratio, duration, voice, captions and credit budget.
- Read `generation_options_list` for capabilities and costs; use `generation_plan_brief` and
`generation_plan_validate` to resolve supported choices and warnings.
- Create or identify the editable project, then call `video_storyboard_create` with the approved script,
confirmScriptApproved=true and a stable idempotencyKey. Planning consumes credits.
- Review the returned scene narration, references, factual warnings and production credit estimate.
- If voice or pronunciation needs review, use `generation_voiceover_create` for a short approved sample first.
Show its cost and actual audio; reuse the accepted voice for production. A voice sample is a separate paid job.
- Use `video_storyboard_revise` for selected scene changes; retain the new immutable revision ID.
- After user approval of previews and their cost, call `video_storyboard_approve` with confirmApproved=true.
- Follow `video_storyboard_get` until awaiting_asset_approval; show actual keyframes and reference previews.
- A `scene_keyframe` regeneration changes one preview. A character anchor refreshes dependent previews;
a style anchor refreshes all previews. Explain the affected scope/cost before the confirmed regeneration.
- After approval of the actual previews and animation cost, call `video_assets_ap
Preview of the first 4,000 characters. Read full instructions →