Image & Video

Comfy-Org's Lens model lands native ComfyUI support with 10 aspect ratios

Lens model goes native in ComfyUI, supporting 10 aspect ratios at two base resolutions.

Deep Dive

Comfy-Org has unveiled Lens, a fresh image generation model now available on HuggingFace, with a critical pull request (#14077) set to merge native support into ComfyUI. The PR allows users to git fetch and checkout the branch for early testing. Lens supports two base resolutions—1024 and 1440—each offering 10 aspect ratios ranging from portrait (1:2) to landscape (2:1). Resolutions like 768×1376 for 9:16 or 2080×1040 for 2:1 give users granular control over output dimensions.

Initial tests show the model handles JSON prompts well, handling detailed scene descriptions including main subjects, secondary elements, environment, style, and camera details. For example, a prompt for an anthropomorphic badger playing retro games in a 1990s basement produced acceptable results. This native integration simplifies ComfyUI workflows and reduces the need for custom node workarounds, making Lens a practical addition for users who want structured, aspect-ratio-aware image generation.

Key Points
  • PR #14077 adds native ComfyUI support for the Lens model from Comfy-Org.
  • Lens supports 10 aspect ratios at two base resolutions (1024 and 1440).
  • Model accepts detailed JSON prompts for precise control over scene composition, style, and camera details.

Why It Matters

Offers ComfyUI users a flexible, natively integrated model for aspect-ratio-controlled image generation with structured prompts.