Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.comfy.org/llms.txt

Use this file to discover all available pages before exploring further.

Lens is an open text-to-image model by Microsoft, licensed under MIT. With 3.8 billion parameters, it uses a dual-stream MMDiT architecture with GPT-OSS-20B text encoder features and the FLUX.2 semantic VAE to deliver competitive image quality with substantially less training compute than larger T2I models. Model highlights:
  • Dual-stream MMDiT architecture — joint image-text processing for coherent generation
  • GPT-OSS-20B multi-layer text features — rich text understanding for accurate prompt following
  • FLUX.2 semantic VAE — high-fidelity latent representation
  • Mixed-resolution training — supports various aspect ratios without cropping
  • Compact 3.8B footprint — runs on consumer GPUs with 24 GB VRAM
Related links:

​
Lens text-to-image workflow

Both the standard and Turbo variants use a Subgraph node to manage the text-to-image generation pipeline. You can open the subgraph to inspect or customize the internal nodes.

Learn about Subgraph

This workflow uses Subgraph nodes for modular processing. Check out the Subgraph documentation to learn how to customize and extend the workflow.

​
Lens

Download Workflow

Download JSON or search “Lens” in Template Library
Make sure your ComfyUI is updated.Workflows in this guide can be found in the Workflow Templates. If you can’t find them in the template, your ComfyUI may be outdated. (Desktop version’s update will delay sometime)If nodes are missing when loading a workflow, possible reasons:
  1. You are not using the latest ComfyUI version (Nightly version)
  2. Some nodes failed to import at startup

​
Get started

  1. Update ComfyUI to the latest version
  2. Go to Template and search for Lens
  3. Select the Lens workflow
  4. Download any missing models (see model downloads), enter your prompt, and click Queue

​
Example output

Lens text-to-image example output

​
Lens Turbo

Lens Turbo is a distilled variant that generates images in fewer sampling steps for faster inference.

Download Workflow

Download JSON or search “Lens Turbo” in Template Library

​
Get started

  1. Update ComfyUI to the latest version
  2. Go to Template and search for Lens Turbo
  3. Select the Lens Turbo workflow
  4. Download any missing models (see model downloads), enter your prompt, and click Queue

​
Example output

Lens Turbo text-to-image example output

​
Lens model downloads

You can find all model files at Comfy-Org/Lens on Hugging Face.

lens_bf16.safetensors

Diffusion model for Lens (BF16).

lens_turbo_bf16.safetensors

Diffusion model for Lens Turbo (BF16).

gpt_oss_20b_nvfp4.safetensors

Text encoder (GPT-OSS-20B) shared by both Lens and Lens Turbo.

flux2-vae.safetensors

VAE (FLUX.2) shared by both Lens and Lens Turbo.
Model storage location 
📂 ComfyUI/
├── 📂 models/
│   ├── 📂 diffusion_models/
│   │   ├── lens_bf16.safetensors (or lens_mxfp8.safetensors)
│   │   └── lens_turbo_bf16.safetensors (or lens_turbo_mxfp8.safetensors)
│   ├── 📂 text_encoders/
│   │   └── gpt_oss_20b_nvfp4.safetensors
│   └── 📂 vae/
│       └── flux2-vae.safetensors

​
Available models

ModelDescriptionInference stepsLink
LensStandard 3.8B model — higher quality, more steps~50Hugging Face
Lens TurboDistilled variant — faster generation with fewer steps~8Hugging Face