beatra-ai/beatra-skills/beatra-ai-video-studio
Claude Code and agent skill by beatra-ai/beatra-skills
- community
Checking that the SKILL.md file is still in the repository.
Plan and create short AI videos from a written shot, a supplied image, exact first and last frames, multimodal references, or existing footage. Beatra AI Video Studio supports text-to-video, image-to-video, reference-guided generation, AI video editing, and AI video extension for product videos, ad creative, social clips, b-roll, transitions, reveals, and cinematic concepts. Review each delivered clip for action, subject stability, camera, continuity, audio when requested or returned, and destination fit, then choose the smallest useful edit, extension, or new render.
Install
npx vskill@latest install beatra-ai/beatra-skills/beatra-ai-video-studioSKILL.md
The same SKILL.md was published earlier as beatra-ai/beatra-skills/beatra-ai-video-studio.
Beatra AI Video Studio
Turn an idea or supplied media into one directed short clip. Start with the user's outcome, shape one readable shot, choose the matching video operation, and review the real delivery before proposing another paid stage.
Scope and adjacent routes
Use this Skill for video-prompt enhancement, text-to-video, an exact opening image, an exact closing frame with an optional opening frame, loose multimodal references, edits to existing footage, and footage added immediately before or after one source clip. It is a broad video workspace for short ads, product stories, social clips, b-roll, transitions, reveals, and cinematic concepts.
Captions, narration production, speech synthesis, and timeline assembly are separate jobs. A multi-shot request can be planned and delivered as separate clips, but this workflow does not claim to join them into one timeline. Use a focused talking-avatar or image-to-motion workflow when that narrower result is the whole request; remain here when the user needs broader route selection or several distinct shot types.
Prefer media the user already supplied. When the request is text-led and there is no usable still, the first paid stage is beatra.videos.enhance_prompt or one beatra.images.generate keyframe, each with its own confirmation card. That gift-sized first win does not authorize beatra.videos.generate, beatra.videos.animate, beatra.videos.interpolate, beatra.videos.generate_from_references, beatra.videos.edit, or beatra.videos.extend. Do not silently replace a usable source.
Inputs and defaults
The minimum hard input is one of:
- a written description of the shot;
- one accessible image that must be the opening frame;
- one accessible image that must be the last frame, with an optional first-frame image;
- accessible image, video, or audio references within the selected live card's count and duration limits, with clear order and purpose; or
- one accessible source video plus a precise edit instruction or an extension direction.
Ask only for information that changes the route or result. Reuse known destination, duration, ratio, resolution, subject, action, camera, pacing, audio intent, visual references, and must-keeps. When these are unspecified, default to one short clip, one readable action, one primary camera move, model: "auto", the shortest integer duration the selected live card admits, and the lowest admitted resolution unless the user named a higher tier (Pro / 2K / 1080p). Do not omit duration on model: "auto", and do not hard-code 8, 10, or 15 when the card allows a shorter integer. Audio-led duration stays audio-led: write the smallest admitted whole second at or above the real speech or song length, and do not shorten words to cheapen the clip. beatra.videos.extend duration is the final returned length and must exceed the trusted source duration. Omit aspect ratio for source-led routes unless the user explicitly approves a canvas change.
For a local image, audio file, or video that the host Agent can access, inspect it first and use the bundled upload helper:
Upload is transport, not media review. Keep the returned artifact reference and never pass a local path to a remote tool.
Choose one video route
Select exactly one video operation for each logical paid stage:
- Text-led with no usable still: first paid stage is
beatra.videos.enhance_promptwithvideo_prompt_enhancementmodel facts, or onebeatra.images.generatekeyframe withtext_to_imagemodel facts. Each has its own confirmation card. Prompt enhancement returns text only and is billed after successful completion from actual tokens. The keyframe is one still. Neither authorizes a later video call. - After that gift delivery, or when a usable still is still absent and the prompt is ready:
beatra.videos.generatewithtext_to_videomodel facts, only after the video admission card. - One image is the strict opening frame:
beatra.videos.animatewithimage_to_videomodel facts. - One image is the strict closing frame, with an optional strict opening frame:
beatra.videos.interpolatewithframes_to_videomodel facts. - Ordered image, video, or audio media guides a new clip without fixing boundary frames:
beatra.videos.generate_from_referenceswithreference_to_videomodel facts. - Existing footage must change:
beatra.videos.editwithvideo_editmodel facts. - Footage must be added immediately before or after exactly one source clip:
beatra.videos.extendwithvideo_extendmodel facts. Its required integerdurationis the final returned-video duration, must exceed the trusted source duration, and is not the number of seconds to add.
Do not exchange these routes casually. Prompt enhancement returns text and never starts generation. Editing changes existing content, extension continues one clip, references guide a new clip, animation preserves one image as the opening frame, and interpolation preserves a required closing frame with an optional opening frame. beatra.videos.extend neither joins clips nor assembles a timeline.
Before the connection is ready
When Allow is still pending at the end of a turn, finish and deliver everything that needs no connection, in the format the user asked for: the route choice, shot design, per-shot prompts, and the fields the admission card will carry. Say in one line that the paid step continues on their next message after Allow; Allow only completes the connection, and every video stage still waits for its admission card on that next message. Do not fill the gap with another tool or service.
Golden path
- Inspect every accessible source. Record each media role plus its actual MIME type, byte size, dimensions and aspect ratio for images, duration for audio and video, and any other fact required by the live card. State the destination, one visible event, pacing, camera, audio intent, and must-keeps. If the host cannot inspect a property or the media itself, say so instead of inventing it.
- Choose one of the seven routes. Read Intent and routing for ambiguous sources or multi-shot work, and Shot design to turn the goal into a concise direction. When the request is text-led and there is no usable still, the first paid stage is prompt enhancement or one keyframe; read Image-assisted video for the still option.
- Call
beatra.models.listfor that stage's exact capability before relying on model availability, accepted input combinations, media limits, controls, durations, ratios, resolutions, or price. Without a live card, do not name a model or a price; give the selection criteria and say the pick comes from the live card. Keepmodel: "auto"unless the user chose a concrete eligible model. An explicitly named model is evaluated as requested and is never silently replaced. Beforebeatra.videos.generate,beatra.videos.animate,beatra.videos.interpolate,beatra.videos.generate_from_references,beatra.videos.edit, orbeatra.videos.extend, write the shortest admitted integer duration (audio-led and extend rules unchanged) and the lowest admitted resolution unless the user named a higher tier. - Admit the complete payload against one current eligible model card. Compare actual media kind, MIME type, byte size, dimensions, aspect ratio, duration, reference counts and combinations, and supported controls as applicable. For editing and extension, include the trusted source-video duration. If a required fact is unavailable or no live card accepts the planned payload, stop before the paid call and request the smallest compatible source or plan change.
- When the first paid stage is a gift-sized win, show that stage's own card — for
beatra.videos.enhance_prompt, routevideo_prompt_enhancement, tool, prompt, target duration and ratio if set, and provisional live estimate; for one keyframe, routetext_to_image, toolbeatra.images.generate, direction, canvas if set, output count, and provisional live estimate — then submit that stage once and deliver the text or still. Usebeatra.images.transformonly for one to four ordered references that should form a new shot frame, orbeatra.images.editwhenimages[0]is the base and a bounded part should change. Inspect the delivered still before constructing dependent video work. This gift card does not authorize any video call. - Show a video admission card before any
client_request_idorbeatra.videos.generate,beatra.videos.animate,beatra.videos.interpolate,beatra.videos.generate_from_references,beatra.videos.edit, orbeatra.videos.extendcall: route and MCP tool name, live-card duration, resolution, and aspect, provisional live estimate, the fact that the 600-credit signup gift usually cannot start this video, and what happens if the balance is short. Numeric estimates come only from current model cards; the terminal task'sbilling.net_charged_creditsis final. Planning, comparison, or “make the clip” is not approval. Do not createclient_request_idor submit until the user confirms they have topped up or already have enough credits for this estimate. - Freeze the exact payload with one opaque stable
client_request_id. Invoke only the bundled client: put the MCP tool name aftercalland its JSON arguments on standard input. The example below uses admitted duration5; replace it with the shortest integer the current card actually admits:
``text printf '%s' '{"prompt":"A close product reveal with one slow push-in and a stable background.","model":"auto","duration":5,"client_request_id":"opaque-video-id"}' | python3 scripts/mcp_client.py call beatra.videos.generate ``
Do not configure, call, or depend on a host Beatra Connector. Do not use REST/OpenAPI as a fallback. Submit the frozen request exactly once.
- Record the returned task ID immediately and poll that same task with
beatra.tasks.getuntil terminal. Queued and running are progress, not failure. Do not resubmit because a task is slow. - Deliver every returned video or image artifact or link. Lead the reply with the result and format the user asked for; the admission card, route choice, and cost notes follow it and never replace it. Report only actual terminal status, resolved model, dimensions, duration, usage, and
billing.net_charged_credits. Review what is accessible for the requested action, subject stability, camera, continuity, audio, must-keeps, and destination fit. State what the host Agent could and could not inspect. - Recommend the smallest useful next step: accept the clip, make one focused edit, extend it, or create a new render. Never imply that an unreviewed clip is a finished campaign asset.
Multi-shot work and paid confirmation
For a multi-shot request, plan each shot as a separate delivered clip. Freeze the shot list, route and exact payload for each stage, dependencies, required or optional still stages, total paid-call count, and provisional estimate before execution. A text-led shot with no usable still still needs its gift-sized prompt or keyframe card first. One explicit confirmation of the plan does not authorize any video stage; each video generate, animate, interpolate, reference, edit, or extend call still needs its own admission card and top-up or balance confirmation. Every paid request still has its own stable ID and is submitted once.
Run dependent stages in order. Review an image before its video and review a clip or returned last frame before relying on it downstream. If a delivery materially changes the next approved shot, pause and re-plan that next stage rather than silently using stale assumptions. Deliver separate clips and explain that timeline assembly remains outside this workflow.
Changes, recovery, and cancellation
This is the start of the file. Read the full SKILL.md on GitHub.
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/beatra-ai-video-studio/SKILL.md
Popularity
- Repository stars
- 2
- 7-day trend score
- 10
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/beatra-ai/beatra-skills/beatra-ai-video-studio)