Behind the Scenes of Beautiful Marge | AI Music Video Creation with Kling AI

Making a music video used to mean cameras, sets, and a production budget. This behind-the-scenes walkthrough shows a different path: building the AI music video Beautiful Marge entirely from AI-generated images and video clips, using a specific chain of tools from image prompt to final edit.

Step one: creating the images

The foundation of the whole video is the image work, done with a self-built plugin called Zebonastic, which works alongside Midjourney to craft the visuals. The core principle driving this step is simple: the better the quality of the image input, the better the eventual video output. That meant spending real time tweaking prompts in Zebonastic to get each image exactly right, including a specific image of the character Marge with her iconic tall blue hair, established early so it could stay consistent across the video.

Step two: directing scenes with a custom GPT

Once the images existed, the next step was turning static visuals into motion. That's where a second custom tool comes in: Hitchcock, a custom GPT built specifically as a Kling AI text-to-video director. Hitchcock takes the images alongside the song's lyrics and turns them into video scene descriptions, complete with detailed camera shots and effects, functioning like an AI-powered version of a director planning coverage for each shot.

Step three: generating clips with Kling AI

For actually generating the video clips, the primary tool was Kling AI, chosen because it's accurate and often nails a scene on the first attempt. Runway Gen-3 served as a secondary option when more variations were needed to test, since it's well suited to fast, less constrained generation depending on the plan in use. The practical distinction drawn here is about budget and access: Kling AI is the stronger choice if you don't have unlimited generation access, since it tends to deliver usable results without needing many attempts, while Runway Gen-3 makes more sense if you're on an unlimited plan and are comfortable generating many variations and picking the best one afterward.

Step four: editing it all together

With image and video generation complete, the final stage was traditional editing, using tools like Camtasia or Adobe Premiere to bring everything together. This is where the music gets synced to the visuals, pacing gets fine-tuned shot by shot, and any additional polish gets layered on to make the finished video work as a coherent whole rather than a sequence of disconnected AI-generated clips.

The full chain, start to finish

Laid out end to end, the pipeline runs from Midjourney images shaped by Zebonastic prompts, through scene direction by the custom Hitchcock GPT, into video generation with Kling AI, and finally into traditional video editing software for the finished cut. Each step depends on the quality of the one before it, which is why the emphasis throughout stays on getting the image prompts right at the very start.

Key takeaways

• Image quality is the foundation of the whole pipeline: better input images produce better AI-generated video output.
• Zebonastic, a self-built plugin, works with Midjourney to generate the source images used throughout the video.
• Hitchcock, a custom GPT built as a Kling AI text-to-video director, converts images and lyrics into detailed scene and camera direction.
• Kling AI is the stronger choice without an unlimited generation plan; Runway Gen-3 fits better when you can afford to generate many variations.
• Traditional editing tools like Camtasia and Adobe Premiere remain essential for syncing music, pacing, and final polish.

Who this is for

This walkthrough is aimed at creators experimenting with AI image-and-video pipelines for their own music or short-form video projects. It pairs well with the companion video on using the Zebonastic plugin directly, and it reflects the kind of hands-on AI production work associated with Humanitarians AI's broader creative and technical content.