Behind The AI: ‘Tequila Heads’

We’re taking the hypothetical and making it real. Dive into our AI-production pipeline and experiment along with us as we highlight our process for ‘Tequila Heads’– our first fully AI social video for The Bad Stuff Tequila.

1 Leonardo.ai

We used Leonardo.ai’s ‘Realtime Generation’ to rapidly ideate the spot’s imagery. Leonardo’s model, true to its name, produces results in realtime as the prompt is entered. Images can then be refined with their additional models or create 4-second ‘Image to Video’ clips.

2 Magnific.ai

Once we landed on which images we wanted to transform into videos, we first brought them into Magnific.ai– self-described as “The most advanced AI upscaler & enhancer.”

Their results were consistently better than just about every other upscaler we tried. Their sliders– Creativity, HDR, Resemblance, and Fractality– allowed for greater overall precision as well.

3 Photoshop Beta

It’s best to remove those extra appendages before you take your image into video. Adobe Photoshop Beta has a whole suite of additional Generative tools that make the retouching process easier than ever before. These tools will eventually make their way into the default version, but why not buy stock early?

4 Runwayml

‘Image to Video’ generation is still very hit or miss. Sora experiments have given a very skewed idea of what can be accomplished with public-facing platforms. For this stage of the process, we ran each image through both Runway 2 and Leonardo.ai. It was a literal coin flip with results (as of April 30, 2024).

Runway (hypothetically) allows for more camera and motion control, but the quality is very dependent on which random ‘seed’ assigned to “provide a stylistic starting point.” If you find one that delivers, save it! You can lock in your preferred seed on future video generations.

Leonardo was better for creating more motion in a static frame. People would move their arms and legs more, without transforming into entirely new objects or people.

*Updates happen so frequently on these platforms and their models that this information may only be relevant for the next 4 - 6 weeks.

5 Topaz Video Ai

Topaz Video AI offers studio-grade video enhancement for fixing poor resolution, stabilization, motion blurs and more. It’s the most expensive of any of the platforms listed, but makes a big difference when starting with a 768x768px video. Topaz utilizes multiple models to accomplish its corrections and improvements. The render times are relatively quick for short clips.