Jedrick Kosinski

Jedrick Kosinski is a generative-AI practitioner associated with ComfyUI, the open-source, node-based platform for creating image, video, audio, and 3D workflows. At the 2025 AI Engineer World’s Fair workshop, he represented Comfy Org alongside ComfyUI’s creator, ComfyAnonymous, and explained how visual pipelines give creators precise control over increasingly sophisticated models.

• Portable, reproducible workflows: ComfyUI can embed workflow parameters in generated media, allowing users to reload, inspect, modify, and rerun the process behind an image. Its open-source repository also supports reusable subgraphs, custom nodes, and API integrations.

• Fine-grained creative control: Kosinski described applying separate low-rank adaptations to different image regions, allowing distinct characters or styles to coexist while giving creators control over each modification’s strength.

• Local versus hosted inference: Locally executed models provide greater visibility into code and data handling, while API-based inference makes hosted models available without equivalent local hardware. Connectivity, accessibility, and sensitive creative material shape the tradeoff.

• Usability as infrastructure: Community-built custom nodes extend ComfyUI but can complicate onboarding and dependency management. Kosinski emphasized reusable subgraphs, clearer interfaces, and reliable extension compatibility as prerequisites for smoother local operation and future cloud workflows.

For virtual try-on and character consistency, he contrasted older combinations of masking, segmentation, and inpainting with FLUX.1 Kontext, which simplifies image editing through reference images and natural-language instructions. Explicit workflow graphs remain useful when creators need tighter control over precisely what changes.