AI Fashion – Lauren
The VERY Commercial - My AI fashion model creation workflow
Taking a look at the cost of overseas shoots:
Travel to far off exotic locations can be very expensive so I set myself the goal of creating a spec ad that maintained character and location consistency, and had the look and feel of a Vogue-style ad, that was shot in a swanky location in Dubai.
The tools I used on this production was Midjourney for the initial character design, Nano Banana for the outfit change and character model sheet, and Kling for the video output.
As usual, I chose music from Audiio.com and the final edit was cut together in Davinci Resolve.
The Takeaway
The takeaway from this process was the real benefit of using a multimodal model. In a ad campaign that uses a single model, it is vital that the look of the character is totally consistent. If one relies on the first frame/last frame approach, it is inevitable that the character or outfit will drift in it’s look. As with any job it’s vital to view the output with a critical eye, looking out for any deviations or coherence issues.
How I Created a FULL Fashion Campaign - Breaking Down MY AI Fashion Film Shot-by-Shot
The VERY Commercial - My AI fashion model creation workflow
The Goal:
The goal was to put together a fast, flexible, high quality production pipeline that would enable fast iteration, a consistent character look, combined with high quality, consistent AI generated environments. The model needed feel right for the brand. She needed to be beautiful but not overly glamourous. The type of woman that would feel comfortable and look great in the type of clothes that Very sell.
The Process;
Step one was to create the beautiful female model . Using Midjourney I experimented with different looks and different ethnicities. During the process I was referring to the kind of models that they typically film with so that I didn’t go too far away from their norm.
I find Midjourney a good tool to start with but the images can, depending on the requirement, feel a bit too cinematic. I love cinematic but there is a time and place to turn it down. The way I do that is that once I’ve landed on the character, I will take the images into Nano Banana and create a model sheet. The prompt will request the model sheet be produced is with the talent standing in a photographic studio, with all required views being rendered. I normally will roll the prompt a number of times until I get exactly the images that I’m looking for.
Next step was the environment. I knew I wanted one to be a runway of sorts and the other to feel like an exclusive hotel. I decided to create two quite different locations because I wanted quite different lighting for the outfits that the talent would be wearing. I set the runway scene in what looked like a grand hall or cathedral. A location that would be extremely expensive if not impossible to film in.
After creating the model sheets of the environments. Again using Nano Banana, I headed over the Kling. In Kling Omni I created two reference for the talent (one in each outfit) and a reference for each of the two locations.
Each sequence was split into 7 x 2 secs clips, with prompts requesting a focus on different areas of the talent…face, hands, shoes etc, combined with descriptions of the camera moves.
I ran the prompt 4 times for each outfit and location, and selected the best shots that were produced. After review the shots I then selected shots that would be suitable for a first frame/ last frame approach.
After all the shots were selected, I visited the Audiio website and found a suitable track.
Editing and final grade were done in Resolve.
Creative control for set-design using AI
The Takeaway
There are some amazing generative tools out there right now. They’re all slightly different and they can all be part of the pipeline. Like any project, planning is everything. The more research you can do before you start generating images, the better. Clear goals for the project will ensure that it hits the brief.
Full and consistent character model sheets