AI Research Engineer – Diffusion Model Training

We're building generative image models for character-based children's entertainment, and we're looking for an AI Research Engineer to train, fine-tune, and improve them.

This is a research and model-training role. You'll work from datasets prepared by a separate data cleaning and preparation team. Your job is to make the models better: improve image quality, character consistency, style consistency, and prompt adherence through structured experimentation.

We're looking for someone who reasons scientifically about model behavior, who can look at a failing output and determine why, then fix and test the solution. This requires more skills than simply training a model.

We're delivering industry-leading character and style consistency. You'll be a key part of this team, pushing the boundaries of what's possible for some of the biggest names in entertainment.

What you'll do:

• Train and fine-tune diffusion models (LoRA and related methods)
• Design and run controlled ML experiments: architectures, hyperparameters, training strategies
• Evaluate output quality using visual, quantitative, and human-evaluation methods
• Diagnose failed generations and identify the model-side cause (identity consistency, captions, regularization, LoRA rank, learning rate, architecture limits)
• Improve character consistency, style consistency, and prompt adherence
• Compare outputs across experiments and keep clear experiment logs
• Give the data team specific, actionable feedback on what data improves model performance
• Recommend changes to training strategy based on results

Requirements

• Strong Python and PyTorch
• Hands-on experience training or fine-tuning deep learning models
• Experience with diffusion models and generative image systems
• Experience with LoRA, DreamBooth, or similar fine-tuning methods
• Solid grounding in computer vision and deep learning fundamentals
• Ability to design controlled experiments and isolate variables
• Comfortable interpreting loss curves and debugging training behavior (overfitting, underfitting, dataset bias)
• Strong documentation and experiment-tracking habits
• Able to work independently in a remote, startup-style environment

Nice to have

• Stable Diffusion, SDXL, Flux, ControlNet, IP-Adapter
• Character-consistent and multi-character image generation
• Experiment tracking with Weights & Biases, MLflow, or TensorBoard
• Cloud GPU infrastructure (RunPod, AWS, GCP), Docker, Linux
• Experience with children's media, animation, or illustration content
• MSc, PhD, or research background in AI / computer vision / ML; publications or a strong image-generation portfolio