Replicating Balenciaga Harry Potter Advertisement

Warning

This post is more than a year old. Information may be outdated.

An Gen AI Experiment done by SuppMon104
Tested generative AIs to create content for a fictional scenario involving Harry Potter and Donald Trump.
Utilized various AI tools for character design, voice generation, and animation.
Although facing challenges in the process, the result was satisfactory.

Key Points

Used ChatGPT to generate prompts for dressing 10 Harry Potter characters in 1990s Balenciaga fashion pieces.
Used Midjourney to create character images, but getting the desired image was difficult, with mid-quality results being more straightforward than high-quality ones.
Used Eleven Labs for voice generation. Extracting a good voice for Donald Trump proved challenging and time-consuming, and the result could have been better.
D-ID was used for animation, with no issues encountered.

In voice generation, finding noise-free training data can be complex. The process can be labor-intensive, requiring multiple iterations to achieve the desired tone and quality.
The generated script's pacing and pauses may not match the user's expectations, resulting in a defective final product.
Using the voice of a character like Harry Potter proved difficult due to the need for more distinctiveness in the adult version of the character's voice. Characters with more recognizable voices may yield better results.