SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Krea is building next-generation AI creative tools with a mission to empower human creativity through intuitive and controllable AI. The company has raised over $83M from top-tier investors including a16z and Bain Capital, and recently launched Krea 2, their first foundation model built from scratch for aesthetic diversity and stylistic control.
As an ML Researcher focused on posttraining, you will lead the enhancement of diffusion models through large-scale finetuning and advanced posttraining techniques. Your responsibilities include:
- Finetuning diffusion models at scale to improve image aesthetics and quality
- Implementing posttraining techniques including supervised finetuning, preference optimization (DPO, GRPO, PPO), reinforcement learning, on-policy distillation, and model acceleration strategies
- Designing comprehensive evaluation suites and reward models for reinforcement learning in image space
- Training custom vision-language models (VLMs) as reward models and custom LLMs for prompt expansion
- Coordinating with data teams to manage preference data collection and model evaluation
- Working on safety alignment for open-source model releases
- Collaborating with AI research and engineering teams to integrate advancements into products
You should have proven experience posttraining diffusion models for image or video generation, deep expertise in large-scale model training and optimization, and strong proficiency in PyTorch. The ideal candidate understands distributed training paradigms (FSDP, tensor parallelism, pipeline parallelism), low-precision training techniques (FP8, NVFP4), fast inference engines (vLLM, sglang), and RL infrastructure optimization. Experience with VLM training, knowledge of LLM/representation learning research, and strong research taste are essential. You'll work in a goal-oriented environment requiring comfort with ambiguous research objectives and the ability to translate them into concrete experiments and execution plans.
The team includes musicians, designers, visual artists, and engineers working full-time in-person at Krea's North Beach office in San Francisco. The company emphasizes fast iteration, execution speed, and demonstrated interest in the creative space.
