Hey {{first_name | Friend}} 👋

Adam here, back with another spectacular student highlight.

What happens when AI isn’t a shortcut… But a collaborator?

Trace of One is one of the most ambitious projects to come out of Lighthouse AI Academy: a full fashion film, from concept to color grade, built with AI.

Marta Musial and her team created an entire fashion film (from concept to color grade) using nothing but AI tools.

Yes, the model and outfits were as real as I am. 

Everything else? Generated.

And the kicker: it’s repeatable. Below is the exact path—concept, LoRAs, image and video generation, sound, and 6K finishing.

Curious how far AI can take you when you slow down and do it right?

Keep reading, my friends 👇

Let’s walk through the process that brought it all to life and unpack the many lessons from this massive artistic undertaking.

The team spent over two weeks researching Mai Gidah’s brand identity, sketching references, and gathering poems, films, and cultural visuals onto a shared Miro board.

They used ChatGPT as a thinking partner, drafting treatments, refining storyboards, and building out the journey of a character moving through four lands — each one altering the design and geometry of his outfit.

Unlike a traditional fashion campaign, the dataset photoshoot focused on training material, not final images.

Neutral backdrops. Minimal lighting variation. Dozens of full-body, medium, and close-up shots of model Yoshua in each outfit.

The aim: Consistency. Expression. Range.

But details matter: a wrinkled shirt, a smudge of makeup — these become part of the dataset and replicate across generations.

The team trained multiple LoRAs: one for the model’s identity and four for the outfits.

Each needed to balance flexibility with fidelity — consistent results, across scenes.

They experimented with captioning strategies and lighting conditions, noting how even small biases in the dataset would affect the final look.

The image stage wasn’t about perfection, but rather about flow.

The team thought in sequences. Every frame was designed to work in motion.
They used controlnets (pose, depth, canny) to ground consistency, inpainting for refining character and outfit accuracy, and ComfyUI as the central tool for merging all layers.

Upscaling was a key hurdle: subtle changes in texture or tone could completely shift the garment or face. A second round of inpainting was often required just to hold visual consistency before video gen.

They used Kling 1.6 for most videos, later testing Kling 2 when it became available. Prompting was everything: to simulate real fabric movement, each outfit was described in careful detail, from texture to weight.

Every shot required iteration. The team would test 10 versions of a clip, compare outcomes, and edit only the strongest into the final film.

Postproduction followed a surprisingly traditional path: smart upscaling, DaVinci Resolve for color, soundtracks blended between stock and ElevenLabs-generated music.

Trace of One is much more than an AI experiment; it’s a production-grade proof of concept.

It shows what happens when creators don’t just prompt, but craft.

It proves that AI workflows can carry emotion, identity, and consistency if you treat them with the same rigor you’d apply to any creative job.

You’ve read the process, now it’s time to see the result.

This 2.5-minute fashion campaign was built entirely with AI: from concept to camera movement, from character to color grade.

📽️ Watch on YouTube  

Trace of One wasn’t built by one person: it was a collaboration across disciplines, continents, and craft. 

Here's the team behind the magic:

That’s your challenge. Not to finish anything but to practice the workflow, stage by stage.

Iteration = mastery 🤺

Applications are officially open for all three Lighthouse AI tracks:

Build visual workflows from scratch with ComfyUI—no prior node experience required.

Who it’s for: Artists, designers, and creatives who want solid AI foundations.

👉 Basic ComfyUI Course

Scale from prototype to pipeline: batching, automation, quality control.

Who it’s for: Technical artists, pipeline builders, VFX supervisors, and advanced users.

👉 Advanced ComfyUI Course

Design systems, not just outputs: workflow design, ethics, team integration, portfolio project.

Who it’s for: Creative directors, studio leads, strategists, innovation teams.

👉 AI for Creative Leaders

Your referrals can write your name down in the application process, and we will take care of the rest.

This is community-driven growth. And by helping others step into AI, you earn in the process.

This issue was a simple truth in motion: 

AI doesn’t replace your craft; it extends it. 

Start with concept, design for consistency (datasets → LoRAs → ControlNets → inpainting → upscaling), and think in sequences, not single frames. 

Most of all, create together; collaboration makes the work deeper and a whole lot more enjoyable, too!

That’s it for now: Thanks for reading and for building this new era with us. 

AI moves fast, but there’s always room to make something honest, useful, and yours.

If you’re ready to level up in 2026, applications are open for Basic ComfyUI, Advanced ComfyUI, and AI for Creative Leaders. 

Bring a friend through the Affiliate Program (they get 10% off, you earn 10% back) and grow alongside people who care about creating as much as you do.

Step by step, node by node — and away we go.

Keep creating and always remember to have fun.

— Adam & the Lighthouse AI Academy Team ☀️