Descriptions:
🚀 A 3.8B image model that can generate 1440px images?
Microsoft Lens is small, but its results are surprisingly serious.
In this video, we test Lens inside ComfyUI and see whether this lightweight text-to-image model is actually worth using.
🔥 What you’ll see:
🧠 What Microsoft Lens is and why 3.8B matters
🖼️ How it handles 1440px image generation
⚙️ How to load Lens in ComfyUI step by step
⚡ Normal Lens vs Turbo Lens: 20 steps vs 4 steps
🎯 Key settings: GPT-OSS, FLUX2 VAE, CLIP type, CFG, denoise
🧩 Why ModelSamplingFlux and CFGNorm are easy to miss
📸 Real tests: cinematic realism, Chinese prompts, English text, fantasy scenes, aspect ratios, and object counting
✅ My takeaway: Lens is not perfect, especially for exact object counts and Chinese text inside images.
But for a small model, its detail density, long-prompt understanding, and resolution flexibility are genuinely impressive.
#runninghub #ComfyUI #MicrosoftLens #LensAI #AIImageGeneration #TextToImage #ComfyUITutorial #ComfyUIWorkflow #AIArt #GenerativeAI #ImageGenerationModel #LensModel #HuggingFace #RunningHub #FluxVAE #GPTOSS #ComfyUISettings #AIModelTest #TextToImageWorkflow #AIDrawing #ComfyUIChina
————————————————————————————————————————
resource
model download：
https://huggingface.co/Comfy-Org/Lens
Runninghub Fan Benefit: Register to get 1,000 RHB
https://www.runninghub.ai/?utm_source=kol02-RH021
https://github.com/amao2001/ganloss-latent-space/blob/main/workflow/2026-05-28%20comfy_lens.json
————————————————————————————————————————
