AI Fine-tuning vs. FaceGen
Fine-tuning for AI Avatars
The traditional way to build pipelines for AI-generated avatars is to train a custom model for each user, which takes ~14mins of GPU time and costs ~$1500 credits per run. This gives you the best quality on the market, is the most predictable, and is how most of our customers doing AI Headshots have implemented into production.
FaceGen for AI Avatars
Alternatively, we just released new model called FaceGen, which costs ~$.15-.20 per user and generates in 60 seconds.
The quality is a bit less predictable, but it’s much faster/cheaper. This output only required 1 input image:
Choosing the right option for your use-case
Choosing the right option depends on your specs for cost and quality.
Fine-tuning will deliver higher quality results, and remains the best option for producing images of different subjects — including pets, styles, products, or objects. See here for what’s possible with fine-tuning styles or objects:
How To Generate Style-Consistent Assets: Finetuning on Leap
Style-consistent asset generation using AI is revolutionizing capabilities for design, marketing, and product teams. Using custom image model training on Leap, teams can now create an endless library of visual assets with just a handful of reference images reflecting your their design style. Let's take a look at a few
Alternatively, FaceGen has unlocked new use-cases as a result as a 10x cheaper and faster solution, and we’ve noticed FaceGen does best for photorealistic styles and simple prompts.
For more information on pricing, please see here.