Tools

What You'll Build

A YouTube production pipeline that takes an idea and turns it into a polished video without you ever opening a video editor. Six stages run in sequence:

  1. Idea research. Claude reads your channel's performance data plus the wider trends from VidIQ, picks the topic, and writes the brief.
  2. Script. Claude writes the full script in your voice, including the hook, the structure, the calls to action, and the cards.
  3. Voiceover. Either an AI avatar delivers it on camera (HeyGen) or a cloned voice delivers it faceless (ElevenLabs).
  4. Visuals. Runway and Higgsfield generate the B-roll. Remotion renders the title cards, lower thirds, and motion graphics from code.
  5. Assembly. The pipeline stitches voiceover plus visuals plus music into the final video.
  6. Upload and metadata. The pipeline writes the title, description, tags, thumbnail, and chapter markers, then publishes.

What you do, as the human: pick the channel direction once, review the script, sign off on the cut. Roughly 30-45 minutes per video instead of 8-12 hours.

Why This Works

YouTube is a content-volume game. The channels that grow are the ones that ship consistently. Burnout kills more channels than bad ideas. The traditional production model (write, record, edit, thumbnail, upload) takes an entire workday per video for a polished result. Most creators can sustain that for maybe six months before they hit the wall.

The no-code AI pipeline changes the math. Once it's set up, the marginal cost of one more video is the API time plus the music license. The marginal cost of your time on each video drops from 8 hours to 30 minutes. A channel that used to ship once a week can ship four times a week without you working any more hours.

The trade is that the videos are produced, not crafted. They look professional. They sound professional. They are not going to win a Cinema Eye award. For 90% of what YouTube rewards (informational, educational, list, news, opinion, niche reviews), produced is enough. For documentary-style storytelling, vlog content, or anything where the human element is the product, you keep producing those by hand.

The Stack, Layer by Layer

Layer 1: Research and Idea Selection

The pipeline starts with two data inputs:

The output is a ranked list of ideas. Each idea has a working title, a hook, an angle, and a confidence score based on the trend data plus your channel's history.

You review the list and pick. The pipeline takes it from there.

Layer 2: Script

Claude writes the full script. Not a bullet outline. A real script with:

The key to making this not generic is a "voice file" you write once. The voice file is your style guide: the phrases you use, the words you avoid, your stance on the niche, your typical opening, your typical closing. Claude reads the voice file every time it generates a script. The output sounds like you, not like Claude.

Layer 3: Voiceover (Two Paths)

Path A: AI avatar. HeyGen records a video of an AI presenter delivering the script. You can use a stock presenter or train one on yourself. The output is a talking-head video file. Works for educational, explainer, and list channels where a face on camera improves retention.

Path B: Faceless voiceover. ElevenLabs generates the voiceover from the script using a voice you've cloned (yours) or a voice you've licensed (a stock voice). The output is just the audio track. Works for news, opinion, reviews, and anywhere a faceless format is the convention.

For most channels, faceless is fine. The avatar route signals "this is AI" more strongly. Some audiences love that. Others reject it. Test both, pick the one your audience rewards.

Layer 4: Visuals

The visuals are the part most pipelines underdeliver on. The Bob Business stack uses two engines:

The pipeline picks visual segments to match the script's pacing. Talking-head segments get B-roll laid over them. Transition moments get motion graphics. The result feels intentional, not slapped together.

Layer 5: Assembly

This is where most "AI YouTube" stacks fall apart. Stitching audio plus video plus graphics plus music into a finished file usually means opening a video editor. The Bob Business stack does it in code:

You can open the MP4 to verify before upload. You do not have to touch a timeline.

Layer 6: Upload and Metadata

Claude writes the YouTube metadata package:

A small skill in Claude Code calls the YouTube Data API to upload the video, set the metadata, and schedule the release.

Step-by-Step Setup

Step 1: Write the Voice File

Before anything else, write a one-page voice file:

This is the single highest-leverage 30 minutes you'll spend on the whole pipeline. Every video the pipeline produces will inherit from this file.

Step 2: Set Up the Idea Research Loop

Connect Claude Code to:

Write a skill that runs weekly and produces the ranked idea list. The output is a markdown file or a Notion page you can review on your phone.

Step 3: Pick Your Voiceover Path

If you want a face on camera, set up HeyGen. If you want faceless, set up ElevenLabs with your cloned voice. Either way, test the output on a 60-second sample script before you build the rest.

Step 4: Build the Script Skill

Write a Claude Code skill that takes a topic and outputs a full script in your voice file's style. Test it on 5-10 topics. Tune the prompt until the output is close enough that your edit takes 5 minutes instead of 30.

Step 5: Configure the Visual Generators

Sign up for Runway and Higgsfield. Spend an afternoon prompting both. Find the visual style that matches your channel. Save the prompt patterns as templates so the pipeline can reuse them.

For motion graphics, install Remotion. Build three templates: an opening title card, a section transition, and an outro card. These three are enough for 90% of videos.

Step 6: Wire the Assembly Script

Write the FFmpeg pipeline script that takes the inputs (voiceover audio, visual clips, motion graphics renders, music) and produces the final MP4. This is the most technical step in the whole pipeline. If you do not code, this is where you hire a freelancer to write it once and then you reuse it forever.

Step 7: Connect to YouTube

Set up the YouTube Data API. Write the upload skill. Test it on a private upload first. Once you're confident the metadata lands correctly, switch to public uploads with scheduling.

Step 8: Run One End-to-End Video

The first time you run the pipeline end to end, expect to fix five things. The voiceover pacing might be off. A visual prompt might render wrong. The music might not match. Fix each one, write down what you fixed, and re-run. The second video takes half the manual work. By video five, the pipeline is producing without your intervention.

Adapting This for Your Channel

The same pipeline works for almost any non-vlog channel, with small adjustments per format.

Faceless documentary or essay channels. Use ElevenLabs for voiceover, lean heavily on Runway and Higgsfield for cinematic visuals, use a slower script pacing. Format examples: niche history, business case studies, science explainers.

Faceless news and opinion channels. Same voiceover stack, lean on motion graphics and stock footage instead of generative visuals. Faster script pacing. Daily cadence.

Educational and how-to channels. HeyGen avatar if you want a face, ElevenLabs if you don't. Screen recordings layered in via the assembly pipeline. Section markers especially important.

Review and unboxing channels. This is where the pipeline breaks down. The product has to physically exist in front of a real human. The pipeline can still write the script, generate the metadata, and assemble the final video; you have to record the product segments yourself.

Comedy and personality channels. Probably skip this. Personality channels are where the human element is the product. The pipeline can do the post-production work (cuts, captions, thumbnails) but the source footage has to be you.

Gotchas and Tips

What This Replaces

Before this stack:

After this stack:

For a channel that lives and dies on consistency (which is most of them), the pipeline removes the constraint that kills channels before they ever scale. Whether your channel grows past 10K subscribers is now a question of niche fit and idea quality, not whether you have the stamina to keep shipping.

Keep Reading