WeVisOne-VL

Comfy-Org has repackaged the 2B and 4B sizes of Tencent's WeVisOne-VL into files directly loadable by ComfyUI. Each size is available in two precisions, bf16 and int8, and is placed in the models/text_encoders directory. The original model comes from tencent/WeVisOne-VL-4B, licensed under Apache 2.0. The local inpainting workflow in this issue's MOSS section uses it to locate editing regions based on text.

Comfy-Org has repackaged the 2B and 4B sizes of Tencent's WeVisOne-VL into files directly loadable by ComfyUI. Each size is available in two precisions, bf16 and int8, and is placed in the models/text_encoders directory. The original model comes from tencent/WeVisOne-VL-4B, licensed under Apache 2.0. The local inpainting workflow in this issue's MOSS section uses it to locate editing regions based on text.

Why it matters The official package includes both bf16 and int8 precisions, so you don't need to convert weights yourself when using it for text-guided editing workflows in ComfyUI.

Hug Face Likes 8 · Hug Face Upvotes 8 · Hug Face Downloads 0

License: Commercial use allowed · ComfyUI: Officially packaged · VRAM: ≈18 GB estimated

Reasons

  • Hugging Face Comfy-Org account
  • ComfyUI: Officially packaged by Comfy-Org; the model card documents ComfyUI usage
  • Score 60 (fresh 15 / source 15 / heat 0 / comfy 16 / vram 4 / demo 0 / license 10)

Usage

  • Download the desired size and precision, place them in ComfyUI/models/text_encoders/, and select them in the corresponding workflow

Risks

  • The model card only lists files, with no instructions on usage or supported workflows; the 4B bf16 version has high VRAM usage

Open

Hugging Face Comfy-Org 账号

Explore the latest AI Radar updates