Lens: Rethinking Training Efficiency for Foundational Text-to-Image Models

Contributors (Alphabetical Order):
Baining Guo, Chong Luo, Dong Chen†, Dongdong Chen, Fangyun Wei†, Ji Li, Jianmin Bao, Jiawei Zhang*, Jinjing Zhao*, Lei Shi, Qinhong Yang, Sirui Zhang*, Xiuyu Wu, Xuelu Feng, Yan Lu, Yanchen Dong, Yang Yue*, Yitong Wang, Yunuo Chen, Zhiyang Liang*, Ziyu Wan†
Microsoft  |  *Core Contributors  |  †Project Lead

     

---

Lens is a 3.8B-parameter foundational text-to-image model designed for efficient training and fast high-resolution generation. It combines dense-caption pre-training, mixed-resolution learning, GPT-OSS multi-layer text features, and the FLUX.2 semantic VAE to reach competitive quality with substantially less training compute than larger T2I models.

This repository provides the minimal inference code for generating images from Lens DiT checkpoints.

Highlights

• Efficient Foundation — Trained on Lens-800M, an 800M image-text corpus with long GPT-4.1 captions, maximizing information density per training batch.
• Compact & Expressive — A 48-block MMDiT denoiser leverages FLUX.2 latents and concatenated multi-layer GPT-OSS features for stronger prompt following and multilingual generalization.
• Flexible Resolution — Mixed-resolution training enables inference across aspect ratios from 1:2 to 2:1 and resolutions up to 1440×1440.
• Post-trained Variants — RL tuning improves visual quality and artifact suppression; the distilled Lens-Turbo supports fast 4-step generation.

Gallery

Page 1 / 6   samples 000-005

Sample 000 · 1440x1440 A generous portion of classic British fish and chips served on a sheet of white paper, golden crispy beer-battered cod fillet alongside thick-cut chips, a wedge of lemon, mushy peas in a small dish, malt vinegar bottle nearby, wooden pub table, overhead shot	Sample 001 · 1440x1440 The iconic Big Ben clock tower and the Houses of Parliament in London at golden hour, the River Thames reflecting warm amber light, Westminster Bridge in the foreground, a classic red double-decker bus crossing, dramatic clouds lit by sunset	Sample 002 · 1440x1440 La Tour Eiffel au crépuscule vue depuis le Trocadéro, la structure en fer illuminée de milliers de lumières dorées scintillantes, le ciel passant du bleu profond au violet, les fontaines du Trocadéro au premier plan avec des reflets dorés, silhouettes de promeneurs
Sample 003 · 1248x1664 A crystal dragon soaring through an aurora borealis sky, its entire body made of transparent faceted crystal refracting the green and purple aurora light into rainbow spectra, ice particles trailing from its wings, high fantasy digital art	Sample 004 · 1664x1248 Aerial view of Yuanyang rice terraces in Yunnan province at sunrise, thousands of cascading water-filled paddies reflecting golden and pink sky colors, morning mist weaving between terrace layers, lush green hillside with scattered palm trees, drone photography	Sample 005 · 1664x1248 A green iguana basking on a moss-covered fallen log in a tropical rainforest, every scale and spine rendered in sharp detail, dewdrops clinging to its skin, a blurred waterfall and lush tropical foliage in the background, National Geographic wildlife photography style

Page 2 / 6   samples 006-011

Sample 006 · 1248x1664 Oil painting portrait of a Renaissance noblewoman in a deep blue velvet dress with pearl drop earrings, soft chiaroscuro lighting revealing delicate skin, craquelure texture on the painted surface, in the style of Vermeer	Sample 007 · 1440x1440 An artisan honey jar with a hand-illustrated vintage botanical label reading "Mountain Wildflower Honey" in brown serif letterpress-style typography with decorative flourishes, detailed ink drawings of wildflowers, clover and honeybees surrounding the text, kraft paper label on clear glass jar	Sample 008 · 1440x1440 Watercolor portrait of a thoughtful young man reading a worn leather book in a Parisian cafe, loose wet-on-wet brushstrokes bleeding into warm amber and burnt sienna washes, visible paper grain texture
Sample 009 · 1664x1248 An explorer's oak desk with an aged world map spread open, a brass sextant, leather-bound navigation journal with handwritten entries, melting candle in a copper holder, scattered compass and quill pen, warm window light, still life photography	Sample 010 · 1664x1248 New York Grand Central Terminal subway station with the classic station name "GRAND CENTRAL" spelled out in elegant white ceramic mosaic tile letters embedded in a dark green tile wall, each letter approximately eight inches tall, ornate tile border frames, the S-curve of train tracks visible	Sample 011 · 1664x1248 A ruby-throated hummingbird hovering in front of a bright red heliconia flower, wings frozen in a figure-eight pattern showing iridescent feather detail, individual water droplets suspended around the bird, high-speed macro photography with dark background

Page 3 / 6   samples 012-017

Sample 012 · 1664x1248 An old Remington typewriter with a sheet of cream-colored paper rolled into the carriage, the typed words "Chapter One: The Beginning" visible in slightly uneven Courier typeface with characteristic ink density variations, some letters slightly misaligned, warm desk lamp lighting	Sample 013 · 1664x1248 The Great Wildebeest Migration crossing the Mara River at golden hour, hundreds of animals plunging into churning water sending spray everywhere, dust clouds rising from the riverbank, dramatic backlit scene, National Geographic documentary style	Sample 014 · 1248x1664 A charming flower shop storefront window with hand-painted white script lettering on the glass reading "Fresh Flowers Daily" in flowing connected cursive with decorative swashes, roses and peonies arranged in buckets visible through the lettering, morning sunlight catching the painted letters
Sample 015 · 1248x1664 A steampunk floating sky-city built on massive gear-driven platforms, brass and copper towers connected by chain bridges, steam-powered airships and hot air balloons docking at various levels, sunset clouds below the city, detailed concept art	Sample 016 · 1664x1248 Milford Sound in New Zealand at dawn, a perfect mirror reflection of steep fjord walls on glass-still water, waterfalls streaming down thousand-foot cliffs, morning mist hovering above the water surface, panoramic landscape photography	Sample 017 · 1248x1664 An Indian Bharatanatyam classical dancer in the aramandi pose, bronze ankle bells and elaborate hand mudra gestures, rich silk costume with gold temple jewelry, captured mid-performance with dramatic stage lighting

Page 4 / 6   samples 018-023

Sample 018 · 1248x1664 A narrow alleyway in Marrakech's old medina with walls painted in vivid cobalt blue, colorful handwoven rugs and ceramic plates displayed along the walls, ornate wooden doors, warm sunlight from above creating dramatic shadows, Moroccan architecture	Sample 019 · 1664x1248 A rustic wooden sign at a fishing village dock reading "Fresh Catch of the Day" in hand-carved letters painted nautical blue, thick hemp rope threaded through the sign as a border, fishing nets and lobster traps stacked in the background, seaside atmosphere	Sample 020 · 1664x1248 A sunken shipwreck on the ocean floor completely overgrown with colorful coral formations, schools of tropical fish swimming through the broken hull and portholes, shafts of sunlight streaming down from the surface above, underwater archaeology photography
Sample 021 · 1664x1248 Zhangjiajie pillar mountains rising above a sea of clouds at sunrise, golden light painting the sandstone peaks, the surreal Avatar-like floating mountain landscape stretching to the horizon, aerial drone photography capturing immense vertical scale	Sample 022 · 1440x1440 A red-eyed tree frog perched on a bright red bromeliad flower in the Costa Rican cloud forest, its neon green body contrasting with blue-striped flanks and orange feet, water droplets on its smooth skin, extreme macro with ring flash lighting	Sample 023 · 1248x1664 Inside a massive limestone cave, ancient stalactites and stalagmites meeting to form columns, an underground river reflecting the formations like a mirror, subtle warm lighting revealing millions of years of mineral deposits, spelunking exploration photography

Page 5 / 6   samples 024-029

Sample 024 · 1664x1248 A weathered 1960s gas station with a large roadside sign reading "ROUTE 66 GAS & GO" in retro rounded sans-serif letters with a red and white color scheme, vintage gas pumps with analog dials in the foreground, a classic Chevrolet parked to the side, Americana nostalgia	Sample 025 · 1664x1248 Construction site hoarding covered in unauthorized street art with "ART IS EVERYWHERE" spray-painted in large freehand capital letters using multiple overlapping colors of red, yellow and blue, paint drips running down from each letter, chaotic beautiful urban canvas	Sample 026 · 1664x1248 Top-down view of a koi pond, dozens of ornamental koi fish in vivid red white orange and gold patterns swimming through crystal-clear emerald water, fallen cherry blossom petals floating on the surface, Japanese garden aerial photography
Sample 027 · 1664x1248 The Potala Palace in Lhasa under a canopy of stars with the Milky Way arching overhead, Tibetan prayer wheels and butter lamps in the foreground casting warm golden light, the massive white and red palace walls glowing in moonlight, night photography	Sample 028 · 1248x1664 Yellowstone's Grand Prismatic Spring shot from directly above by drone, concentric rings of vivid blue turquoise green yellow and orange created by thermophilic bacteria, steam rising from the surface, abstract natural color palette	Sample 029 · 1664x1248 A herd of African elephants walking in a line across the savanna with Mount Kilimanjaro's snow-capped peak behind them, golden sunset dust kicked up by their feet creating a hazy atmosphere, telephoto wildlife photography showing massive scale

Page 6 / 6   samples 030-031

Sample 030 · 1664x1248 The Hall of Mirrors at the Palace of Versailles, hundreds of candles reflected infinitely in the massive gilded mirrors, crystal chandeliers casting prismatic light across painted ceilings and gold leaf ornamentation, Baroque opulence	Sample 031 · 1664x1248 A pirate captain's cabin, navigation charts pinned to the wall, a brass telescope and astrolabe on the desk, stacks of gold coins and a jewel-encrusted goblet, rum bottle, warm swinging lantern light casting shadows with the ship's motion

Installation

Tested environment: Python 3.12 · CUDA 12.6 · PyTorch 2.11.0+cu126 · TorchVision 0.26.0+cu126

conda create -n lens python=3.12 -y
conda activate lens

uv pip install torch==2.11.0+cu126 torchvision==0.26.0+cu126 \
    --index-url https://download.pytorch.org/whl/cu126
uv pip install -r requirements.txt

The default GPT-OSS encoder and FLUX.2 VAE are loaded from Hugging Face. Make sure your environment has access to any gated model repositories you use.

Checkpoints

Repo	Description	Steps	CFG
microsoft/Lens	Default. RL-tuned for visual quality	20	5.0
microsoft/Lens-Turbo	Distilled from the RL model for fast 4-step sampling	4	1.0
microsoft/Lens-Base	Supervised base model (no RL, no distillation)	50	5.0

Pick a variant by passing its repo id to --repo_id (CLI) or LensPipeline.from_pretrained(...) (Python).

Inference

Important: run from the cloned repo root so from lens import LensPipeline resolves to this package — importing lens is what registers LensGptOssEncoder / LensTransformer2DModel with the transformers and diffusers namespaces that model_index.json references.

Python API:

import torch
from lens import LensPipeline

pipe = LensPipeline.from_pretrained(
    "microsoft/Lens", torch_dtype=torch.bfloat16
).to("cuda")

image = pipe(
    prompt="A cat holding a sign that says \"hello world\"",
    base_resolution=1440, aspect_ratio="1:1",
    num_inference_steps=20, guidance_scale=5.0,
    generator=torch.Generator("cuda").manual_seed(0),
).images[0]
image.save("lens.png")

To trade speed for VRAM, replace .to("cuda") with pipe.enable_model_cpu_offload().

CLI — basic usage:

python inference.py \
    --repo_id "microsoft/Lens" \
    --prompt "A cinematic mountain lake at sunrise, soft mist, detailed reflections" \
    --base_resolution 1440 --aspect_ratio 1:1 \
    --steps 20 --cfg 5.0 --n 1 --seed 42 \
    --out ./outputs

Batch generation — join multiple prompts with |:

python inference.py \
    --repo_id "microsoft/Lens" \
    --steps 20 --cfg 5.0 \
    --prompt "a red fox in snow|a glass greenhouse at night"

A100 / V100 (no MXFP4 kernels) — dequantize the GPT-OSS encoder to bf16:

python inference.py \
    --repo_id "microsoft/Lens" \
    --steps 20 --cfg 5.0 \
    --prompt "a cat" \
    --disable_mxfp4 --offload

Options

Flag	Description	Default
--repo_id	HF repo id (or local path) of the assembled Lens pipeline	microsoft/Lens
--base_resolution	1024 or 1440	1440
--aspect_ratio	1:2, 9:16, 2:3, 3:4, 1:1, 4:3, 3:2, 16:9, 2:1	1:1
--steps	Number of denoising steps	20
--cfg	Classifier-free guidance scale	5.0
--n	Number of images per prompt	1
--seed	Random seed (omit for non-deterministic)	—
--out	Output directory	./outputs
--dtype	Compute dtype: bfloat16, float16, float32	bfloat16
--disable_mxfp4	Dequantize the GPT-OSS text encoder to --dtype (required on A100 / V100; Hopper+ keeps MXFP4 by default for less VRAM)	—
--offload	Enable diffusers CPU offload (text_encoder->transformer->vae) to reduce peak VRAM	—
--reasoner	Refine prompts with the loaded GPT-OSS encoder before generation	—
--api_url / --api_key / --api_model	Use an OpenAI-compatible API for prompt refinement (takes precedence over --reasoner)	—

Citation

@article{zhao2026lens,
  title   = {Lens: Rethinking Training Efficiency for Foundational Text-to-Image Models},
  author  = {Guo, Baining and Luo, Chong and Chen, Dong and Chen, Dongdong and Wei, Fangyun and Li, Ji and Bao, Jianmin and Zhang, Jiawei and Zhao, Jinjing and Shi, Lei and Yang, Qinhong and Zhang, Sirui and Wu, Xiuyu and Feng, Xuelu and Lu, Yan and Dong, Yanchen and Yue, Yang and Wang, Yitong and Chen, Yunuo and Liang, Zhiyang and Wan, Ziyu},
  journal = {arXiv preprint arXiv:2605.21573},
  year    = {2026}
}

Responsible AI

The model is released for research purposes only and is not intended for product or service deployment. Responsible AI considerations were incorporated throughout the development process, including data selection, model training, and evaluation. The training data includes a combination of public, licensed, and internal datasets that were processed to remove clearly identifiable personal information and reduce harmful content where possible. However, as the data is largely sourced from web-scale collections, it may contain biases or uneven representation. As a result, the model may generate outputs that are inaccurate, biased, or inappropriate under certain prompts, including content that could be misleading or raise copyright or IP-related concerns. Given these limitations, the model should be used in controlled research settings, with appropriate human oversight. Downstream users are responsible for applying additional safeguards, such as content moderation, validation, and compliance checks, before using the model in broader applications.

Privacy

This project does not collect any usage data. For more information, see the Microsoft Privacy Statement.

License

This project is released under the MIT License.