Friends who know me well know that I've been updating videos recently, and the data has been quite good.
2 videos on Douyin opened the selection plan, and 2 videos on Xiaohongshu landed nearly four-figure commercial orders. Currently, both accounts have only posted 3 videos, and there have already been no fewer than 10 commercial inquiries. There are also creation invitations from Bilibili and course invitations from some platforms.
Hi friends, I'm here to submit my work again~
Today let's talk about how to achieve Codex automatic editing for complex and high-quality knowledge videos.
Actually, this content already had a video version two days ago, which received nearly 200,000 views on Douyin.
But I feel the video length is limited and many things couldn't be explained clearly, so I spent time writing this article to dismantle the complete method and logic.
I've also organized the relevant prompts, tools, and material libraries into an attachment for easy one-click use.
As usual, I suggest you practice it once, but the word count is high as always. If you can't finish reading, you can just throw it to AI to make a Skill.
If there are no problems, let's begin.
01. Why can I get commercial orders?
Since the main revenue for these videos comes from platform traffic and brand deals, I added this part upfront, hoping it's useful.
From my experience, there are two reasons why brands approached me after just 2 videos.
First, the visual level.
This type of video isn't unique, but it's relatively rare, so the format itself is a differentiator. I also did specific visual designs, like a TV-like character theater and various visual effects. These designs make the visual layers richer and the experience better. I believe only if people can finish watching will your content have traffic.
Second, the content level.
I feel that brands in the AI field currently prefer tutorial content. This content can seamlessly integrate products. Since it's a specific tutorial, users must use the product to practice, so both traffic and conversion are good.
02. What is the editing logic?
In my videos, the screen is split into two forms: A-roll and B-roll.
A-roll focuses on situational expression, turning abstract text into understandable scenarios using my IP character. B-roll is supplementary, using charts and graphics to help viewers understand knowledge more intuitively. Alternating between them keeps the audience engaged and improves AI editing accuracy.
03. What tools are needed?
Primarily Codex, which has a built-in image generation model. You also need the HyperFrames plugin for motion effects.
For voiceover, I use Volcengine's Doubao Voice Synthesis 2.0. It's stable and affordable.
04. How to use Codex for editing
Step 1: Copywriting + Voiceover
Prepare your script using Claude or DeepSeek. Then, use Codex to synthesize the voiceover and perform acoustic alignment to get a millisecond-level timeline document.
Step 2: Visual Arrangement Table
Codex acts as a director to split the script into specific visual shots, deciding between A-roll and B-roll for each segment.
Step 3: Formal Editing
For A-roll, ensure IP consistency using character sheets and detail references. Use multiple perspectives (Host, Protagonist, Supporting, First-person) to keep it interesting.
For B-roll, use local motion templates or open-source libraries like Hyperframes-launches to maintain efficiency.
Step 4: Sample and Final Cut
Produce a 1-minute sample first to check the rhythm. Once confirmed, complete the full video and add background music and sound effects.
Final Thoughts
If you don't want to show your face but want to make videos that land commercial deals, this format is for you. In the age of Agents, any idea is worth trying with AI. If you can think it, you can do it.
Good luck! 🍀
