The second edition of creating book introduction videos with Codex + HyperFrames is here! I’ve left the reference Douyin blogger in the comments. Let me know what you think of this version. Codex handled the audio perfectly, giving it a strong storytelling feel. Reference workflow: feed the Douyin blogger’s video to Codex for analysis, prepare the materials after analysis, then HyperFrames.

Codex + HyperFrames: Deconstructing Douyin Hits – Making a Living from Self-Media!

Previously, I used Codex and HyperFrames to deconstruct a viral video from a book account on Douyin. The audio effect received widespread praise. Today, I’m writing a long post to teach you how to deconstruct viral Douyin videos step by step, lock in a repeatable workflow, and empower yourself to create videos with AI.

Serena@369Serena·Jul 3 The second edition of using Codex + HyperFrames for book introduction videos is here! The reference Douyin blogger is in the comments. Let me know how this version looks. Codex handled the audio perfectly, giving it a strong storytelling feel. Reference workflow: feed the Douyin blogger’s video to Codex for analysis, prepare the materials after analysis, then HyperFrames.Show more799654738K

1. Preparation

First, make sure you have the tools ready:

1. You need an agent like Codex.

2. Then install two tools from the plugin library:

   • (a) HyperFrames: Creates videos using HTML, ideal for corporate presentations, data animations, or showcase videos.
   • (b) Remotion: Built with React, it turns video into code. Once you lock in a theme/style, it’s perfect for high-volume reproduction via code. I’ve seen many bloggers use Remotion to create animations similar to “Xiao Lin Says,” which work great alongside voiceover videos.

Additionally, you can install some Skills to assist with deconstruction. If you want to understand the structure of a viral video – like how the first few seconds hook viewers, the audio effect, or the rhythm between narration and visuals – use these Skills to deconstruct and replicate:

1. Claude Video: Used to deconstruct the video workflow, making it easy to replicate next.
2. Video Use: I highly recommend this for editing beginners. It acts like a “video editing director,” calling the right tools based on your needs. For newbies who don’t know when to use which tool, it plays the role of a master director.

That’s the preparation work.

2. Video Production Workflow

Creating videos with Codex + HyperFrames can follow this order:

1️⃣ Set video specs

• e.g., 30 seconds, 9:16, Chinese, knowledge-based, book introduction, for X/TikTok/Xiaohongshu.

2️⃣ Write the voiceover script

• First, ask Codex to write a 30-second Chinese voiceover. Don’t rush into visuals.

3️⃣ Determine visual style

• e.g., dark editorial knowledge video, advanced knowledge type, dark background, book cover as main visual, slight camera push-in.

4️⃣ Create timestamps/storyboard

• Split the voiceover into 4-6 visual segments. Each segment corresponds to one key message.

5️⃣ Prepare materials

• Book cover, title, keywords, logo, background music, voiceover text.

6️⃣ Generate voiceover

• Use Edge TTS, ElevenLabs, or HyperFrames media-related workflow.

7️⃣ Transcribe the voiceover

• Generate a transcript from the final audio. Don’t guess subtitle timings manually.

8️⃣ Create HyperFrames composition

• Write index.html with correct data-start, data-duration, data-track-index.

9️⃣ Add subtitles

• Automatically sync subtitles based on the transcript.

🔟 Preview

• Check the preview first, don’t directly render the final video.

1️⃣1️⃣ Validate / Inspect

• Check for errors, fonts, layout overflow, subtitle occlusion.

1️⃣2️⃣ Render MP4

• Only render after confirming everything is fine.

Those are the general steps. In reality, many of these don’t require manual input. So just get familiar with the process; manual work is minimal.

3. First-time Manual Work: Video Deconstruction + Script Generation

Now we need to deconstruct the video. First, use Claude Video to analyze its structure. The video I deconstructed was from a book account, so the structure was simple: hook the audience in the first few seconds, then introduce or review the book.

The next steps for voiceover script generation:

1. Collect information:

   • (a) Use Codex to gather book information.
   • (b) Use the WeChat Reading Skill to check commonly highlighted sections and reviews.
   • (c) Ask Codex to search for famous book reviews.

2. Generate an introduction:

   • Ask Codex to write an introduction based on book facts. On self-media platforms, this must sound “non-AI.” Adjust according to personal aesthetic preferences. Specific requirements:
     • (a) Avoid constructions like “not… but…” which are typical AI phrasing.
     • (b) Avoid overly parallel sentence structures.
     • (c) The expression should be smooth and natural, narrating the book in a straightforward manner – not with an arrogant tone.

This part requires you to:

1. Use the previously deconstructed video structure.
2. Personally review and approve the copy.

Once generated, you can save this workflow (using Codex to collect and organize book info) as a Skill. That way, the voiceover copy can be reused in the future.

Everything should be based on facts. After this operation, Codex will output a voiceover script. You’ll have your first complete script ready.

4. Audio TTS Processing: Achieving a Natural, Magnetic Voice

Once the voiceover is done, there’s another major part I haven’t automated yet – connecting to voice generation APIs. But I think it’s doable, so I’ll write it down: how to adjust the voice to make it sound more natural, magnetic, and story-like.

I’ve shared before that I used ElevenLabs recommended by Codex. Its English voices are perfect – you can find voices from different countries. However, when generating Chinese audio, I found it doesn’t recognize Chinese well and produces many mispronunciations. That’s not just an “AI feel” problem – it’s unusable. Codex suggested using Jianying (CapCut) or Volcano Engine’s API as alternatives.

Since I didn’t have Volcano Engine’s API handy, I used the CapCut app on my computer.

My method is simple:

1. Create a new project, then find the Text option.
2. Use the “Text to Audio” feature, pick a decent voice.
3. Copy the voiceover script, click generate.
4. Export the audio (CapCut only exports MP4 files).
5. Feed this MP4 file back to Codex.

Based on Codex’s earlier audio analysis, I ask it to process the voice to sound as natural, story-like, and magnetic as the reference video.

Codex’s audio processing workflow:

1. Automatic analysis: Codex first analyzes the original video’s audio parameters and modifies my MP4 file accordingly.
2. Parameter settings: I’ll include screenshots of relevant parameters here. You can send those screenshots to Codex, and it will handle the processing to achieve that effect.
3. Simple method: Or just give it a short reference video and say “I want the audio to sound like this,” and Codex will handle it.

The audio processed this way is perfect. Once you master this voice processing, you can reuse it for other video types, such as:

• (a) Movie explanations or story narrations
• (b) Life philosophy, uplifting content
• (c) Science/education accounts

In these scenarios, this voice processing approach works perfectly.

That said, I believe this part should also be automated. For self-media creation, minimize manual work – you’ll stick with it much easier. So later I’ll try using Volcano Engine’s API to let Codex automatically generate voiceover and TTS files.

5. HyperFrames: Aligning Visuals, Subtitles, and Voiceover

Next is to create the visuals and match the audio with subtitles. At this point, I didn’t intervene at all – Codex already had a general plan. Since I had previously fed it a reference video and now had the voiceover and audio, Codex would create the video on its own.

The generated version won’t be perfect the first time. You might encounter:

1. Audio/subtitle mismatch

   • Communicate with Codex to let it fix the alignment. Codex will automatically check frames to see why they don’t match. This may require two rounds of adjustment.

2. Visual issues

   • Including content, subtitle position, and animation styles. You can iterate with Codex repeatedly, reminding it to get closer to the reference video.

Once the workflow is smooth, you can generate videos with a high degree of replication. If you can automate the whole workflow, you can batch-run accounts. Whether book accounts or other types, you can deconstruct and replicate using this process.

6. Summary

Finally, I hope this article is helpful. I’m also learning as I go, so let’s communicate more. Often, the problem isn’t that we can’t create – it’s that we haven’t started. With Codex, just take action, talk to it in natural language, and find solutions step by step. I believe once you start, everyone can leverage AI to make a living from self-media.

I hope everyone, in this AI era – an era where ordinary people can approach productivity like never before – can create content, find reusable workflows, and make wealth even while you sleep. Approach freedom!

I’m Serena, exploring AI × Web3 × Self-Media, documenting how ordinary people can take back control of their lives. Hope this article helps you! 👏