Microsoft Lens Turbo INT4 ConvRot
Repackaged model files for ComfyUI and others, with INT4 ConvRot converted for testing in Radiant Canvas for fast macOS inference: Radiant Canvas on the App Store.
The original model was published as microsoft/Lens-Turbo. The original
Microsoft Hugging Face repository is currently unavailable through the Hub
API. The preserved ComfyUI single-file BF16 transformer used for this
conversion came from
Comfy-Org/Lens.
File
lens-turbo-int4_convrot.safetensors: the diffusion transformer in ComfyUI's native ConvRot W4A4 format.
For ComfyUI, place the file in:
ComfyUI/models/diffusion_models/
Use a recent ComfyUI build with Lens and native INT4 ConvRot support. The GPT-OSS text encoder and FLUX.2 VAE are not included here. Get compatible versions of those components from the Comfy-Org repack.
Conversion details
- Comfy-Org repack revision:
393618fd7db3e370509f84390b9162b8fd839a25 - Converter revision:
b477b06ecc3afe8c3896a6cef02f5b1b8a550a0b - Source format: BF16 single-file transformer, 1,264 tensors
- Quantized layers: 288 image-side attention and MLP weights
- ConvRot group size: 256
- Quantization group size: 64
- Output size: 6,003,704,896 bytes
- SHA-256:
1a13bea9febbed68b3d6205295b3eeaf09a87b1fda1f212ecae7d8b11655dcd0
All text-side matrices are kept in BF16. This follows the Lens conversion profile and avoids a known empty-prompt failure. Sensitive input, output, modulation, normalization, and bias tensors are also kept in BF16.
This is a test conversion. Check output quality for your own prompts and workflows before relying on it.
License
The original Lens project and weights are released under the MIT License. A copy of the license is included in this repository.