Skip to content
Raipur businesses still relying on manual content creation are already behind—this AI video pipeline shows exactly how to turn raw footage into high-quality marketing assets at scale with minimal effort.
Drop in a raw talking-head video. Get back a trimmed, animated, subtitle-ready clip — orchestrated entirely by Claude Code.
Pipeline: Raw Video → video-use (trim) → Hyperframes (motion graphics) → Claude Code → Final Clip
PHASE 1 — SETUP
Step 1: Understand What You’re Building
Your goal is simple: drop a raw video in, get a polished edited clip with motion graphics and subtitles out. Three tools make this work together:
- video-use — trims filler words, silences, and retakes from your raw footage
- Hyperframes — adds motion graphics and handles rendering
- Claude Code — the orchestrator that ties everything together
Step 2: Install Claude Desktop and Choose a Project Folder
Download Claude Desktop from the official page and sign in with a paid plan that includes Claude Code access. Create a new empty folder — this becomes your video studio.
VS Code tip: Claude Desktop is simpler to start with, but opening your project in VS Code alongside it gives you much better visibility into generated files and folder structure.
Step 3: Pull in Hyperframes and video-use
In Claude Desktop, paste both GitHub repo URLs and prompt Claude to scan, clone, and wire them up as your editing studio. Alternatively, use a pre-built starter kit from the community. Let Claude run until it confirms everything is connected and ready.
Step 4: Configure Transcription
The pipeline needs word-level timestamps to sync motion graphics with speech. Choose one of these three options:
- 11Labs API (recommended — better at identifying clean cut points)
- OpenAI Whisper API
- A local transcription tool (free, but requires local installation)
Security note: Store your API key in an .env file — never paste it directly into the chat, where it would be stored in conversation history.
PHASE 2 — EDITING
Step 5: Record and Import a Raw Test Clip
Record 30–60 seconds with intentional mistakes — false starts, filler words, pauses. This gives you realistic material to work with. Save it into your project’s assets folder. In Claude Desktop, use @ to reference the file by name.
Step 6: Auto-Trim with video-use
Tell Claude to run the video-use tool on your clip, removing filler words, silences, and retakes. It will transcribe the file, identify cuts, and ask you small editorial questions along the way. Answer in plain English:
“Make it as punchy as possible.”
Output: An edited.mp4 and a transcript JSON with word-level timestamps. Ask Claude for the file path to verify the result yourself before moving on.
Step 7: Inspect the Transcript and Timestamps
Open the transcript JSON from your project folder. Confirm you see the full text with precise start times per word. These timestamps are what allow motion graphics to snap perfectly to specific spoken words — they are the connective tissue of the whole pipeline.
PHASE 3 — MOTION GRAPHICS
Step 8: Define Your Motion Graphics Scene by Scene
Watch your edited clip and describe what you want — use voice-to-text or type directly into Claude. Be specific about three things for each scene:
- Placement: left side, right side, lower-third, full-width
- Style: liquid glass card, text overlay, animated illustration
- Text content: exactly what should appear on screen
End your prompt with: “Make everything sync to the exact second where it makes sense.” Send it as one detailed prompt so Claude has a full brief before it starts planning.
Step 9: Use Plan Mode — Refine Before Generating
Switch Claude into plan mode so it maps out “beats” with timing ranges, color palettes, and word anchors before writing any code. Review each beat for placement, typography, and content. Add revisions at this stage — it is far cheaper than iterating after generation.
Token tip: More time spent in plan mode equals fewer tokens wasted regenerating scenes you end up changing. Plan thoroughly, then generate once.
Step 10: Generate, Preview, and Revise
Approve the plan and let Claude generate the Hyperframes HTML compositions. Preview inside Claude Desktop or in your browser. Watch for common issues:
- Cards covering your face on screen
- Unwanted background overlays or grid patterns
- Incorrect frame cropping, especially for the final shot
Write one specific revision prompt that lists all the fixes at once, then let it regenerate.
Step 11: Fine-Tune Timing in the Hyperframes Editor
Open the timeline editor inside the preview to move, shorten, or delete beats without re-prompting Claude. Claude can see these changes directly, so minor timing tweaks don’t consume a new generation cycle. Re-preview and iterate until you’re happy.
PHASE 4 — SYSTEMATIZE
Step 12: Render and Organize Your Project
Ask Claude to produce the final render. Your project folder should now be cleanly structured:
- assets/ — raw clips, edited clips, and transcript JSON files
- compositions/ — the beat HTML files that define motion graphics
- components/ — reusable visual elements
- Final renders and verification screenshots
Pro tip: Let Claude take screenshots of scenes to self-check whether the output looks correct before you do a full review.
Step 13: Turn Finished Videos into Reusable Style Files
Ask Claude to build a style philosophy markdown file from each type of video you produce — lessons, intros, shorts, hype videos, demos. Over time, aim to reach the point where you can drop in a raw file and say:
“Make a YouTube short in my usual style.”
And get a high-quality result with minimal extra prompting.
Step 14: Manage Token Cost Efficiently
Complex runs can consume 200,000+ tokens. Keep costs under control by:
- Spending more time in plan mode before asking Claude to generate code and assets
- Avoiding building complicated scenes you are not sure you want
- Using the planning and iteration approach to avoid going down the wrong path
Practice Plan: 4 Sessions to a Working Pipeline
Session 1 — Setup: Install Claude Desktop, configure your project, import Hyperframes + video-use, configure transcription.
Session 2 — First full pipeline: Record a test clip, run video-use to auto-trim, verify the edited clip and transcript.
Session 3 — Motion graphics: Define 3–4 beats, use plan mode, approve the plan, generate a preview, and refine with the timeline editor.
Session 4 — Systematize: Turn your videos into style philosophy files. Try a new raw clip with minimal prompting.
If you’re not systemizing your video content like this, your competitors eventually will—Technowild can set this up for you before that gap becomes permanent.
