A 3-Step Workflow to Reverse-Engineer Viral AI Video Prompts Without Paying for Them
A practical three-step workflow shared by Awa K. Penn (@TawohAwa) for downloading viral social media AI videos, analyzing them frame by frame in ChatGPT, Claude
In an X thread published on October 19, 2026, AI video creator Awa K. Penn (@TawohAwa) outlined a three-step workflow for reverse-engineering viral AI videos into custom prompts for Seedance 2.5 rather than begging for or buying prompts from other creators.

Image source: https://Publer.io
Many creators pay for prompt packs or leave comment requests trying to replicate trending AI clips on social media. Penn argues that multimodal large language models can inspect a downloaded video frame by frame and reconstruct a detailed prompt tailored to a target video generator. Below is a faithful walkthrough of the author's three steps, the exact prompt template from the thread, plain-language context for the tools involved, and the caveats noted in the discussion.
Step 1: Download the Reference Video from Social Media
The workflow begins by saving the target video file so it can be uploaded into an AI assistant. Penn recommends Publer.io ([https://Publer.io](https://Publer.io)), a social media utility accessible on both web browsers and mobile apps.
The author lists the following sequence for downloading a clip:
- Open
[https://Publer.io](https://Publer.io). - Open the menu in the top-right corner, select free tools, and choose video and photo downloader.
- Paste the social media post link and follow the on-screen instructions to save the video.
Downloading the raw file removes reliance on third-party prompt sellers and gives you a local copy that multimodal AI tools can inspect directly. When saving clips from public profiles, keep usage limited to personal study and structural analysis rather than redistributing someone else's footage.
Step 2: Analyze the Video Frame by Frame in a Multimodal LLM
Next, upload the downloaded video directly into Claude, ChatGPT, or Gemini and instruct the model to break down the visuals for a specific video generator. In this context, reverse-engineering means working backward from a finished video's pacing, framing, and subject motion to write a text prompt capable of reproducing a similar look.
Penn shared the following exact prompt in the thread:
Watch this video frame by frame. Then create a prompt for me specifically for Seedance 2.5 to recreate this video perfectly.
(Optional) Use me from the reference photo as the main character instead of the character in the video. (Upload a photo of yourself)
The first instruction directs the assistant to examine the clip sequentially (Watch this video frame by frame) rather than giving a high-level summary, then formats the resulting description specifically for Seedance 2.5.
The optional second line shows how to swap the subject. By uploading a reference photo of yourself alongside the video, you can instruct the LLM to replace the original character in the generated prompt with your own likeness. Always ensure you have rights and platform permission when using photos of real people.
Step 3: Generate and Publish the Video
In the final step, copy the generated prompt into a video generation platform. Penn notes using Syllaby because it aggregates major AI video models in a single workspace and supports scheduling and publishing across multiple social media platforms.
As an alternative to copying and pasting between browser tabs, the thread also points to using the Syllaby MCP inside Claude or ChatGPT. MCP (Model Context Protocol) is an open standard that lets AI assistants connect directly to external tools and trigger actions on your behalf. With Syllaby MCP configured, you can generate the video inside the same Claude or ChatGPT session where you reverse-engineered the prompt.
Caveats and Practical Notes from the Thread
Before relying on this workflow for production work, keep several verified details and thread caveats in mind:
- Affiliate link disclosure: The Syllaby URL included in Step 3 of the original thread is explicitly labeled by the author as an affiliate link. You do not need to use Syllaby specifically to apply the prompt extraction technique if you already use another AI video generator.
- Manual editing still helps: In a reply to the thread, user
@socialwithaayannoted still rewriting the prompt by hand after Step 2. LLM-generated prompts serve as a structured starting draft rather than a guaranteed one-click clone. - Cross-model testing and camera details: Reply notes in the thread discussion also highlight testing prompts across multiple models—such as Kling, Seedance, or Veo—and breaking down shot composition, lens choice, and lighting separately for finer control.
- Unverified claims: The source thread shares a creator's personal workflow and does not provide quantitative benchmarks comparing reproduction accuracy across ChatGPT, Claude, Gemini, or different video generation models.
Original source
- Awa K. Penn (
@TawohAwa) on X: 3-step workflow to reverse-engineer viral AI video prompts