Mood Ring Queen - (Official Music Video)

Mood Ring Queen - Cover Image

Inspiration

Just wanted to make some enjoyable fun music for the target audience of 2025. The lyrics came out for teenagers. Edited further, enhanced it for Suno metatags and music prompts. Came the output I couldn't stop listening. It'd be more enjoyable visually, so took me a 1 full month and this is my first debut Music Video using many AI tools and workflows in the process.

What it does

Resonates well with the music, lyrics, expressions, and visually entertaining.

How I built it

The main character, face was initially produced on Grok Imagine.
Different variations of it were created using Flux Kontext, Qwen-Image-Edit, Seedream 4 Edit, Nano Banana, Higgsfield Soul, ImagineArt.
Trained this character model lora for Flux and Qwen-Image using tools like Ostris.
Generated storyboard using Grok.
Generated keyframe images for image-to-video using multiple ways. 5.1 Used Midjourney, Qwen-Image, ImagineArt for initial image. 5.2 Then image-to-image generation with character lora on Qwen-Image model, to have matching character face. 5.3 Upscale to near 4k resolution with high crips detail using custom workflow in ComfyUI using models like Wan 2.2 + Qwen-Image.
Fed as image-to-video models for different models on Higgsfield AI, ImagineArt. Models used:

Luma Labs Ray3
Kling 2.1
Seedream Pro
Alibaba Wan2.5
Google Veo3
Higgsfield Video

Also used Sync Labs lipsync-2-pro, but found Wan2.5 much superior in outputs by giving audio for image-to-video.

Challenges I ran into

Initial training of the character model. Took about 1 week for accurate dataset creation, finding optimal parameters and settings to get a good consistent model output.
Lip syncing character to the song with natural looking visual. Sync Labs has limitations if something is partially covering the mouth, or multiple characters in the video. Wan 2.5 worked flawlessly as solution. No other models I tried came close to Wan 2.5 for lip-synced video.

Accomplishments that I'm proud of

Having a well edited video which is not a simple a stitched video of multiple visuals only video. I edited the video in Final Cut Pro heavily. Generate multiple shots with different angles, and situations to have rhythemic flow of the video. Kept extra audio generated from videos to keep the vibes of music video alive, which are not initially from audio song, but from video generated using models like Wan 2.5.

What I learned

Learned a lot about training custom AI character model. Enhanced prompting for creating AI music using tools such as Suno 4.5+. Used different models to generate videos and hitting their limits of creativity to my current skills, which I'm still learning and enhancing.