BEN2
Will it run on your hardware?
Pick your GPU memory - see which quantizations fit, and the cheapest card for the rest
Need an exact figure for your context length? Use the VRAM calculator.
Run it locally
Copy-paste - running in under a minute
vllm serve PramaLLC/BEN2New to this? Start with Ollama · serve to many users with vLLM.
Deep dive
Notes, sources, and the full write-up
BEN2 is an open-weight image segmentation model published by PramaLLC on Hugging Face in January 2025, with 90M parameters under the mit license.
Among the 23 image segmentation models in this directory it ranks 14th by monthly Hugging Face downloads, and 14 of the others are larger.
HardwareLink to this section
Unquantized FP16 weights take about 1 GB; 8-bit brings that to ~1 GB and 4-bit (Q4_K_M) to ~1 GB, before activations and any context or image buffers. At this size it runs on almost any recent GPU, on Apple Silicon, and usable on a modern CPU. Check exact numbers for your card with the VRAM calculator.
Similar image segmentation modelsLink to this section
- SAM ViT-Base - 90M, 505,455 downloads/month
- clipseg-rd64-refined - 150M, 362,536 downloads/month
- sam2.1-hiera-small - 50M, 168,596 downloads/month
- face-parsing - 80M, 145,329 downloads/month
- fashn-human-parser - 60M, 76,198 downloads/month
PopularityLink to this section
BEN2 was downloaded 119,457 times on Hugging Face in the last month (1,224,278 all-time) and has 255 likes.
Run it locallyLink to this section
Weights and usage instructions are on the PramaLLC/BEN2 model card; it loads with ben2.
Local runtimes that fit this model:
- comfyui - node-based workflows, the usual way to run diffusion models locally
- invokeai - a canvas UI for image generation and editing
PaperLink to this section
From the community
What users are saying
“1. cloned the repo 2. created a venv named '.venv' 3. activate the venv 4. ran pip install -r requirements.txt 5. manually installed spaces because it's missing with pip install spaces 6. python app.py 7. Drag&Dropped an image into gradio UI 8. clicked on 'Submit' 9.”
“This is just an example I noticed that I can even resave the images or process them in another editor. This does not change the fact that they lose some of the color.”
“Hello, thank you for your comment. We are still preparing the paper for peer review, so outside perspectives are greatly appreciated. The range of the pred is 0 and 1 as it is the result of sigmoid(logits). This algorithm was used for BEN, and our BEN2 models employ a fixed range.”
Frequently asked
Quick answers to common questions
How much VRAM does BEN2 need?
BEN2 with 90M parameters needs approximately 1 GB at Q4_K_M quantization. Use our VRAM calculator for an exact estimate.
Is BEN2 better than other PramaLLC models?
BEN2 has 90M parameters - a strong choice for image-segmentation.
What license is BEN2 under?
BEN2 is released under the mit license, making it suitable for most commercial and personal projects.
What hardware runs BEN2 well?
With 90M parameters, BEN2 is small enough for most consumer GPUs, Apple Silicon Macs, and many CPUs. Check our hardware directory for specific recommendations.
What is the best quantization for BEN2?
Q4_K_M is the recommended sweet spot - ~98% of FP16 quality at ~27% of the size. Q5_K_M (~1 GB) is an option if you have spare VRAM. Use our VRAM calculator to compare.
What models compete with BEN2?
BEN2 competes with other models in its class. Browse our model directory for comparisons, benchmarks, and community reviews to find the best fit.
Nearby options
Similar models and compatible hardware by spec