Krea
ML Researcher - Posttraining
San Francisco · On-site · Full-time · Posted Sep 1, 2026
- Visa sponsorship
- Open to sponsoring international visas where possible, including STEM OPT, OPT, H-1B, O-1, and E-3.
About the job
The researcher post-trains and fine-tunes diffusion and language models to improve image quality, aesthetics, controllability, and safety, and integrates the resulting advances into Krea’s products.
What you'll do
- Fine-tune diffusion models at scale and implement post-training methods, including preference optimization, reinforcement learning, and distillation.
- Design image-focused evaluations and reward systems, including training custom VLM reward models and LLMs for prompt expansion.
- Coordinate preference-data collection and model evaluation, support safety alignment, and collaborate with research and engineering teams on product integration.
What they ask for
- Proven experience post-training diffusion models for image or video generation.
- Experience with large-scale model training, inference, and optimization.
- Strong understanding of LLM and diffusion post-training pipelines and algorithms such as PPO, GRPO, DPO, OPD, and MOPD.
- Strong proficiency in PyTorch and distributed training; knowledge of low-precision training/inference, inference engines, and RL infrastructure.
Nice to have
- Experience training VLM models.
- Interest in creative fields and keeping up with LLM, VLM, representation learning, and robotics research.
Tools
PyTorch, PPO, GRPO, DPO, OPD, MOPD, vLLM, sglang
AI focus
Post-training diffusion models for image/video generation, including supervised fine-tuning, preference optimization, reinforcement learning, distillation, reward-model design, prompt expansion, evaluation, and safety alignment.
Model-written summary · the original posting is authoritative