Reverse engineer the viral AI video effects filters, generating everything from physical effects ...
Hello everyone, still searching for AI effects filters or tutorials on viral videos? Today, we're skipping the endless search for trending AI video tutorials on social media. Using the Gemini 3 Pro's powerful visual analysis capabilities, combined with the highly stable Google Flow engine, we'll recreate all those viral "mermaid transformation," "building time-lapse," and "fruit self-eating" effects!
What will you learn in this episode?
✅ Understand how to communicate with Gemini, provide visual content, and recreate production techniques.
✅ Google Flow raw images and one-stop video creation.
✅ Structural recreating: Using Nano Banana 2 to recreate the original image from a finished renovation, ensuring no frame drops during time-lapse photography.
🛠️ Toolkit for this episode:
Brain analysis: Gemini
Image generation: Google Flow
Structure reconstruction: Nano Banana 2 (for structure reconstruction)
If you enjoy this kind of simple and easy-to-understand AI tutorial, please subscribe to the channel.
👉 Subscribe to AI Vision Lab: / @ai%e8%a6%96%e8%a6%ba%e5%af%a6%e9%a9%97%e5...
#AIVisionLab #ReverseEngineering #Gemini #GoogleFlow #PikaLabs #NanoBanana2 #AIVideoTutorial
Commands used in the video:
Method 1: Provide an AI video and add the following command:
Role: You are now a top AI video technology expert. Task: Please perform "AI production workflow reverse engineering" on the video (or link) I provided, and output the analysis results according to the following structure. No further explanation is needed:
Recommended Workflow: Determine the most suitable production method ([Single Image Generation I2V] / [First and Last Frame Generation S&E] / [Plain Text Generation T2V]).
Material Preparation and Bilingual Image Prompts:
If it is [Single Image Generation I2V]: Describe the characteristics of the starting master image and provide a set of *"Chinese and English" bilingual* image prompts.
If it is [First and Last Frame Generation S&E]:
Describe the characteristics of the starting frame and the ending frame respectively, and provide *two sets of "Chinese and English" bilingual* image prompts.
【⚠️ Core Operation Instructions】: If the video involves building growth, interior decoration, or any structural evolution (e.g., renovation reverting to unrenovated state), you must forcefully instruct users: "To ensure 100% structural accuracy, please directly upload your first (or last) frame image to Nano Banana 2 and use its revert function to generate the corresponding second image. It is strictly forbidden to directly use AI to generate two independent images out of thin air."
Core Video Generation Commands (Bilingual Video Prompt): Provide a precise set of *"Chinese-English bilingual" video prompts (applicable to Grok Imagine / Kling / Luma / Runway)* for the video's dynamics, camera movements, and material changes.
Goal: To reproduce the video effect to the highest degree with the lowest point cost, and all instructions must include both Chinese and English.
Method Two: Provide an AI effects filter URL and add the following instructions:
【Target URL】: [Paste the AI effects URL]
【Role Definition】: You are now a senior image physics simulation engineer with "real-time full-network detection" capabilities. Your task is to deeply analyze all video effects appearing in the target URL and recreate them as 100% practically usable prompts specifically designed for "brute-force unlocking" of Google Flow's highly stable engine.
【Execution Guidelines: Visual Detection and Physical Reconstruction】
*Deep Scan:* Once a URL is received, you must search online for all effects currently displayed on that URL. Do not apply any preset data; create a physical model based on "visual facts."
*Silhouette Destruction:* To break Google Flow's "first-frame lock," any effects involving deformation, liquefaction, or disintegration must include: 'Total structural failure,' 'No solid parts remaining,' and the strongest physical intervention: 'The original silhouette is 100% erased and obliterated from the very first frame.'
*Pixel Evolution:* Do not use "Maintain," which can cause locking; instead, use 'Repurposing all source pixels into [new state]'. This forces the AI to use the original colors to create new objects, instead of maintaining the old shapes.
Physics Anchoring:
Downward Movement: Lock 'Gravity-anchored' and describe it as a horizontal plane.
Upward Movement: Lock 'Weightless' or 'Buoyant'.
Outward Movement: Emphasize 'Centrifugal propulsion'.
No Placeholders: The object is uniformly defined as "the object".
Quality Assurance Syntax (Flow Fixed Version Only): Each English prompt ends with: ", maintaining all its original colors and fine textures from the source image." (Note: The AI must ensure that this instruction does not interfere with the aforementioned "Contour Annihilation," with physical destruction having the highest priority).
[Output Form...