Skip to content

12 models · Vision · Transformers.js

Computer vision in the browser

12 models handle vision in the browser today, from 11.3 MB (MODNet) to 948.7 MB (SigLIP2 Base Patch16-224), a 84.1 MB download at the midpoint. Every one runs with Transformers.js over WebGPU or WebAssembly: no server, no install, no account.

Models
12
Smallest
11.3 MB
Median
84.1 MB
Largest
948.7 MB

Every vision model, ranked by WebGPU size

Model WebGPU · WASM
Model WebGPU download WASM download
MODNet 6.5M params 11.3 MB q4f16 6.3 MB uint8
DINOv3 ViT-S/16 21.6M params 14.1 MB q4 20.9 MB q8
Depth Anything V2 Small 24.8M params 18.2 MB q4f16 26.0 MB uint8
SlimSAM-77 9.7M params 19.8 MB fp16 13.1 MB q8
SAM 2.1 Hiera Tiny 38.96M params 32.1 MB q4f16 59.1 MB uint8
ORMBG 84.0 MB q4f16 42.3 MB uint8
RMBG-1.4 44.1M params 84.1 MB fp16 42.3 MB q8
BiRefNet Lite 44.36M params 109.2 MB fp16 109.2 MB fp16
Grounding DINO Tiny 172.28M params 144.1 MB q4f16 194.4 MB uint8
BEN2 94.63M params 209.0 MB fp16 209.0 MB fp16
CLIP ViT-B/32 151.3M params 240.0 MB q4f16 291.3 MB uint8
SigLIP2 Base Patch16-224 375.19M params 948.7 MB q4f16 721.0 MB uint8

Sizes measured from the HuggingFace API file tree for each model's repo, not estimated. Sorted smallest to largest by the WebGPU headline pick.

Size ladder

The WebGPU download spans 84.3x here: 11.3 MB (MODNet) to 948.7 MB (SigLIP2 Base Patch16-224).

5 of 12 models download a smaller build over WebAssembly than WebGPU: the CPU-fallback quant compresses tighter than the GPU pick there.

7 of 12 models here fit under 100 MB over WebGPU.

Bytes-per-million-params ranges 3.9x within this group: 0.66 MB/M (DINOv3 ViT-S/16) to 2.53 MB/M (SigLIP2 Base Patch16-224).

Notable vision models

Other tasks

Or see every browser model grouped by task, or the main GGUF catalog for native runtimes outside the browser.

FAQ

Do all vision models here use the same WebGPU quant?

No. 7 use q4f16, 4 use fp16, 1 use q4. q4f16 is the most common pick.

What would it cost to download every vision model here?

1.87 GB total over WebGPU across all 12 models, if you tried every one back to back. Most projects only need the single model that fits the job, not the whole set.

Sizes measured 2026-08-01 from the HuggingFace API. Last validated 2026-09-14. See all browser models or the methodology.