System requirements
What LabelBlend needs to run well.
Requirements differ significantly between manual annotation and AI workflows. The table below separates them so you can plan hardware honestly.
Exact minimum and recommended figures are intentionally left as descriptive placeholders until verified on reference hardware. Replace the values below with measured numbers before publishing.
| Basic usage | AI usage | Video processing | Large batch processing | |
|---|---|---|---|---|
| Operating system | Windows | Windows | Windows | Windows |
| CPU | Modern multi-core CPU | Modern multi-core CPU | Higher core count helps frame decoding | Higher core count helps batch jobs |
| RAM | Sufficient for your image sizes | More RAM for larger images | Frame loading is RAM-bound; limits are configurable | Scales with batch size |
| GPU | Not required | Dedicated GPU recommended | Dedicated GPU recommended | Optional |
| VRAM | Not applicable | Depends on checkpoint size | Low VRAM mode available | Not applicable |
| Storage | Space for your dataset | Dataset plus model checkpoints | Dataset, frames and exports | Source plus output folders |
| Internet | Not required after install | Required to download models | Required to download models | Not required |
| AI model storage | None | Several GB depending on checkpoints | Several GB depending on checkpoints | None |
Planning notes
A few practical points that matter more than raw specifications.
Checkpoint size drives VRAM
Smaller SAM2 checkpoints run on modest GPUs; larger ones need more VRAM but generally produce better masks.
Frames are the memory cost
Video work is limited by how many frames you hold in RAM. Use the RAM limit or frame skip settings.
Storage adds up
Datasets, generated frames, synthetic images and model checkpoints all live on disk simultaneously.