At Visualskies, we don’t just follow industry trends, we define them. Over the past six months, we’ve been pioneering a revolutionary approach to Ethical AI Training for asset development and previs that’s transforming how creative teams work.
Our breakthrough? LoRA-based generation powered entirely by our proprietary 3D scan data.
The results speak for themselves: dramatically accelerated workflows, unprecedented creative possibilities, and, most importantly, complete ethical integrity in every pixel we generate.
What Is LoRA—and Why Ethics Matter
LoRA (Low-Rank Adaptation) models can learn specific visual styles from relatively small datasets. Train a LoRA on a particular prop, character or environment, and it can generate consistent visuals for concept work and previs when integrated into workflows like ComfyUI. Check out our recent LinkedIn post on this workflow
The challenge isn’t the technology, it’s the data. Too many AI models are trained on unlicensed, scraped content from artists who receive no credit or compensation.
Our approach has been different.
Training on Data You Own: Real Results from Ethical Sources
Whilst tools like Runway, Veo & Kling are great platforms for generating content. Data security and scraping are known issues. At Visualskies, we process all of our content locally and offline using our bespoke workflow. This allows us to maintain the strict level of data security that studios require.
We exclusively use data we’ve collected in-house: high-resolution 3D scans of people, objects, and locations created for our projects or under client agreements. All ethically sourced, all permission-cleared.
The results? In some cases, our previs turnaround time dropped by 40% for character placement studies. We can now generate 20+ consistent character variations in different lighting conditions within an hour—work that previously took 2-3 days.
More importantly, every model was trained on content we own outright. No ownership ambiguity, no consent issues.
Full-body photogrammetry
Hero Assets & Objects
Ground Truth Landscapes
AI as Creative Extension… With Measurable Benefits
Faster concept visualisation: Initial character placement studies now take hours, not days
Consistent variations: Generate lighting and angle studies without manual recreation
Rapid iteration: Test creative decisions during early production stages
Our workflow doesn’t generate final VFX assets. Instead, it creates previs materials that help visualise how characters appear in specific environments or how objects behave under various lighting conditions.
This process helps studios optimise and save costs during the early stages of production. The time savings are real. But the creative foundation (the scans, environments, and artistic direction) remains entirely human-driven.
Follow this link for an explainer on some the the basics around gaussian splatting.
From Idea to Final Pixel
Capture Speed
Gaussian Splatting enables high-speed capture, delivering environment studies in hours.
Precision
Meticulous scanning fidelity ensures pixel-perfect digital doubles for any project scale.
Creativity
Customised solutions allow creators to iterate rapidly across all sectors.
Transparency Builds Trust
We tell collaborators and clients exactly where AI is used and what role it played. We don’t present AI-generated results as purely human work.
This honesty matters for building trust in AI workflows. Every frame generated by our LoRA models represents the work of our artists and technicians: scanning subjects, labelling data, training models, and refining workflows.
Building Better Pipelines from the Ground Up
As AI tools evolve, the temptation to “move fast and break things” will intensify. But innovation doesn’t require cutting ethical corners.
By training models on our own data, crediting all contributors, and using AI to enhance rather than replace human creativity, we’re establishing a framework others can follow.
We envision a future where:
- AI assists artists without sidelining them
- LoRA models are trained ethically with full source control
- Previs workflows accelerate without losing quality or integrity
The ethical development of AI tools raises broader questions about authorship, trust, and the evolving role of creative professionals. Our goal isn’t just to keep pace with innovation, it’s to help shape its direction, ensuring technological progress aligns with the responsibility and rigor that define effective creative practice.
Interested in our approach?
We’re happy to share our LoRA training methodology with fellow studios working on similar challenges. Reach out to discuss collaboration opportunities.
WHERE TO FIND US?
Visualskies is proud to offer our expert Photogrammetry services for VFX across multiple locations worldwide. Our presence in key cities enables us to provide prompt and efficient service to our clients. You can find us in the following locations.
GLOBAL BASES
LONDON
Unit 4S Hewlett House, Battersea
SW8 4AS, UK
BERLIN
Wilhelminenhofstrasse 92
12459, Germany
NYC
546 West 48th St
NY 10036, USA
ITALY
Via Fratelli Lumiere, 86/c
52100 Arezzo
HONG KONG
20 Wang Hoi Road
Kowloon Bay, HK
