ComfyUI Workflow Architect

We’re hiring a ComfyUI Workflow Architect for a Swedish AI Film Production startup.

About company:

We are building the future of animation and digital IP creation. We combine world-class storytelling with cutting-edge AI tools to produce high-quality animated content at scale for YouTube, TikTok, streaming platforms, and global licensing partners.
We are now looking for a ComfyUI Workflow Architect — someone who lives and breathes pipeline design, prompt engineering, and multi-model orchestration — to help push our productions to the next level.

What You'll Be Doing:

• Design and optimize production-scale diffusion model pipelines: Image-to-video synthesis architectures
• Style transfer, LoRA adaptation, and identity-preserving generation
• Multi-stage conditioning systems (depth, pose, optical flow)
• High-resolution upsampling and noise reduction pipelines
• Develop custom model components: fine-tune ControlNets, train domain-specific LoRAs, implement custom attention mechanisms and conditioning layers
• Collaborate with production engineers to ensure visual consistency through latent space control, embedding management, and multi-shot coherence strategies
• Integrate ML pipelines with training infrastructure, model versioning systems, and asset management tools

Requirements:

• Speaking English on B2-C1 level
• Deeply experienced with ComfyUI's model architecture and node-based inference graphs (required)
• Strong foundations in computer vision and diffusion models: DDPM, DDIM, latent diffusion, conditioning mechanisms
• Proficient with ControlNet, LoRA, and custom conditioning architectures
• Expert in PyTorch/CUDA optimization: memory management, kernel fusion, distributed inference
• Strong communicator able to bridge ML engineering and production requirements

Bonus Points For:

• Custom PyTorch node/operator development for ComfyUI or contributions to the ComfyUI ecosystem
• Published research or novel implementations in generative models, video synthesis, or controllable generation
• Model fine-tuning and training pipelines: experience with distributed training, LoRA/DreamBooth workflows, or dataset curation
• MLOps infrastructure: experiment tracking, model versioning systems, automated evaluation pipelines
• Low-level optimization: custom CUDA kernels, quantization strategies, or inference acceleration techniques
• Adjacent pipeline experience: real-time rendering engines, VFX compositing systems, or production asset management
• Perceptual quality modeling: experience with CLIP embeddings, aesthetic scoring, or automated quality assessment

Working schedule: 9:00-17:00 European time

Socials package:  TBA