Verdict: For small teams that publish a steady stream of talking-head shorts, the best 2026 workflow is a two-stage pipeline. Use Descript's text editor to rough-cut footage and export scene-by-scene clips, then hand those clips to a Claude Code Remotion skill that adds captions, B-roll, motion graphics, and renders the final video from code. It is faster than a human editor for repetitive shorts, but it still needs a human reviewer before anything goes live.
Last verified: 2026-06-18
- Best for: high-volume talking-head clips (shorts, Reels, TikTok, YouTube Shorts).
- Core stack: Descript (rough cut) + Claude Code + Remotion skill (motion + render).
- Cost floor: Descript Hobbyist $16/mo (annual) + Remotion free for teams of 3 or fewer + Anthropic API usage.
- Honest limit: this is a build, not a buy; expect 5–10 hours to train the skill and a review loop on every render.
What this workflow actually does
The dream of "AI edits my video" usually means one of two things: either the AI makes every creative decision, or it automates the mechanical parts while a human makes the calls. This workflow is the second kind.
Stage one is Descript. You import your raw recording, use text-based editing to delete retakes and dead air, and export each scene as its own clip. Stage two is a Claude Code Remotion skill. You drop the clips into a folder, ask Claude to edit them, and Claude reads a SKILL.md file that encodes your brand rules: which captions to write, where to place logos, what B-roll to fetch, how to follow the speaker's head movement, and which music to lay underneath. Claude then writes React/TypeScript code, Remotion renders it, and you get back a finished video to review.
Why combine Descript with Claude Code instead of using one tool?
No single product currently handles the whole pipeline well. Descript's AI co-editor, Underlord, is excellent at text-based rough cuts, silence removal, filler-word deletion, and exporting clean clips (Descript Underlord). It is not a motion-graphics tool. Remotion, by contrast, is a developer framework that turns React components into MP4s, making it ideal for repeatable animations, captions, and branded overlays (Remotion Agent Skills). If you are weighing different ways to put AI agents to work in your business, our AI for Small Business complete guide covers the same build-vs-buy decision across writing, code, and operations. Claude Code is the glue: it can read your skill, inspect the clips, transcribe them, write the Remotion code, and render the output.
The combination gives you:
- Speed: one prompt can style dozens of clips once the skill is mature.
- Consistency: every export follows the same brand rules encoded in the skill.
- Version control: the skill, templates, and renders live as files you can diff and reuse.
The exact tools and their current costs
| Tool | Role | Cost (2026-06-18) | Source |
|---|---|---|---|
| Descript Hobbyist | Rough cut, retake removal, scene export | $16/person/mo (annual) / $24 monthly | Descript Pricing |
| Descript Creator | More media hours + 4K export + team of 3 | $24/person/mo (annual) / $35 monthly | Descript Pricing |
| Claude Code | Agent harness; reads skill, writes Remotion code | Free CLI; Anthropic API usage billed per token | Claude Code |
| Remotion skill | Code-based motion graphics + render | Free for individuals/teams up to 3 employees | Remotion License |
| Remotion Company License | Required for 4+ employees | Starts at $100/mo | Remotion Pricing |
Pricing changes often; check the source links before you budget.
Step 1: Rough-cut in Descript
Descript is the fastest way to turn a long recording into clean scenes.
- Import your recording into a Descript project.
- In the AI Tools panel, run Shorten word gaps to tighten long pauses. Set a threshold (for example, gaps longer than 300 ms) and target length (for example, 200 ms). Click Shorten all (Descript Help).
- Run Remove retakes to strip duplicate or restarted sentences.
- Review the transcript, delete anything you do not want, and split the script into scenes.
- Click Export → Local export → Video → Scenes. This exports each scene as an individual file (Descript Help).
If you are on the Free plan, exports are limited to 720p and include a watermark, so budget for at least Hobbyist for production work.
Step 2: Build the Claude Code Remotion skill
A Claude Code skill is just a folder with a SKILL.md, optional scripts/, assets/, and references/. Anthropic launched the skill system in October 2025 and published the official build guide in January 2026 (Claude Blog). Remotion published its own official skill in early 2026; you install it with npx skills add remotion-dev/skills (Remotion Agent Skills).
For video editing, the skill should define:
- Input: a folder of scene MP4s and a script or transcript.
- Brand assets: logo files, fonts, color hex codes, music tracks, lower-third templates.
- Rules: caption style, when to add B-roll, how to follow head movement, intro/outro length.
- Output: a Remotion project that renders the final MP4.
Install the skill once:
npx skills add remotion-dev/skills
Then create your own .claude/skills/video-edit/SKILL.md inside your project. The better the rules, the less manual correction you need later. Plan to spend your first 5–10 hours adding examples and correcting mistakes so the skill learns your taste.
Step 3: Let Claude edit the batch
With the clips exported and the skill in place, the workflow becomes conversational:
- Open Claude Code in your project folder.
- Drop the scene MP4s into an
input/folder. - Prompt Claude with the scene names and any special instructions: "Edit scenes 1 to 18. Use the brand kit. Add captions, fetch relevant B-roll for the businesses I mention, and render in low quality first so I can review."
- Claude inspects each clip (dimensions, duration), transcribes it, corrects transcription errors, reads the skill rules, picks templates, fetches images, writes the Remotion composition, and renders a low-quality draft.
- You review the draft, give feedback, and ask for a full-quality render.
The real value is scale: once the skill is trained, one prompt can process a whole batch of shorts.
What the rendered output looks like
A mature skill produces videos that include:
- Burned-in captions synchronized to the corrected transcript.
- B-roll and stock images matched to entities you mention (logos, products, news events).
- Motion graphics from your template library: lower thirds, subscribe buttons, animated charts.
- Head tracking or subtle zoom that follows the speaker instead of a static frame.
- Music and outro pulled from your asset folder.
This is the same agentic idea behind our recent look at Claude Fable 5 for video editing, but built with the shipping tools available today rather than a frontier research model.
The quality ceiling is the template library and the rules you wrote, not the model. If your templates look good, the output looks good. If they do not, the AI will not fix that for you.
Where it works and where it breaks
| Works well | Breaks or needs help |
|---|---|
| Talking-head shorts with a clear script | Highly narrative pieces that need precise timing |
| Repetitive formats (news reactions, explainers) | Emotional or comedic timing |
| Scenes you already rough-cut in Descript | Clips with bad audio, multiple speakers, or heavy crosstalk |
| Strong brand templates | One-off creative experiments with no templates |
Use this as a first-pass factory, not a final editor. The human review is where quality actually happens. If you are building a broader AI content pipeline, see our guide to running an AI SEO funnel with Hermes subagents for how the same agent pattern applies to written content.
Cost reality check for a small team
Assume you publish 20 shorts per month:
- Descript Creator: ~$24/person/mo (annual) for 30 media hours, 4K export, and full AI tools.
- Remotion: $0 if your team is 1–3 people.
- Anthropic API for Claude Code: varies with clip length and complexity. Heavy video-generation sessions can run $10–$50 in API usage; simple edits are much less.
That is usually cheaper than a part-time human editor for a high-volume content pipeline, but it is not free. Time saved on edits can be spent on skill maintenance and review.
What this means for you
If you run a small business, agency, or personal brand that lives on short-form video, this pipeline is worth building in 2026. Start with Descript for rough cuts, add a Remotion skill in Claude Code for styling, and keep a tight review loop. If you are new to Claude, read our five-level progression for using Claude in 2026 before you build the skill. For teams that already use AI for writing, the same operational discipline applies to automating customer support with AI on a budget.
Do not skip the human review. The AI is fast; the editor is still the quality gate.
FAQ
Q: Can one AI tool do all of this without Descript? A: Not reliably in mid-2026. Tools like Remotion can render motion graphics from prompts, but you still need clean source clips and a way to remove retakes, dead air, and filler words. Descript remains the fastest text-based rough-cut tool for that step.
Q: Do I need to know how to code to use Claude Code with Remotion? A: You do not write the code by hand, but you need to understand what Remotion produces so you can review it and fix mistakes. The skill handles the typing; you handle the direction.
Q: How long does it take to build a usable skill? A: Budget 5–10 hours of setup and feedback loops for your first polished output. Anthropic estimates 15–30 minutes for a basic skill, but a video-editing skill with brand templates, B-roll rules, and music cues needs more time (Claude Blog).
Q: Is Remotion free for commercial use? A: Yes, if you are an individual or a company with up to three employees. For four or more employees you need a Company License (Remotion License).
Q: What should I never automate? A: Fact-checking, legal claims, brand-sensitive messaging, and final publish approval. The AI can speed up production; it should not be the last set of eyes on content that represents your business.
Q: Can I use this for long-form videos too? A: The same stack works, but the render time and review surface grow quickly. Most teams start with shorts, refine the skill, and then expand to longer edits.
Discussion
0 comments