Model Zoo

To see all available presets:

from slide2vec import list_models

list_models()        # all presets
list_models("tile")  # tile-level only
list_models("slide") # slide-level only

Tile-level encoders

Preset	Model	Output dim	Spacing (um)	Notes
lunit	Lunit ViT-S/8	384	0.5	Kang et al. (2023)
prost40m	Prost40M	384	0.5	Grisi et al. (2026)
conch	CONCH	512	0.5	Lu et al. (2024)
phikon	Phikon	768	0.5	Filiot et al. (2023)
conchv15	CONCHv1.5	768	0.5	Lu et al. (2024)
hibou-b	Hibou-B	768	0.5	Nechaev et al. (2024)
h0-mini	H0-mini	768 / 1536	0.5	Filiot et al. (2024)
phikonv2	Phikon-v2	1024	0.5	Filiot et al. (2024)
hibou-l	Hibou-L	1024	0.5	Nechaev et al. (2024)
uni	UNI	1024	0.5	Chen et al. (2024)
musk	MUSK	1024 / 2048	0.25, 0.5, 1.0	Xiang et al. (2024)
virchow	Virchow	1280 / 2560	0.5	Vorontsov et al. (2024)
virchow2	Virchow2	1280 / 2560	0.5, 1.0, 2.0	Zimmermann et al. (2024)
uni2	UNI2	1536	0.5	Chen et al. (2024)
gigapath	GigaPath	1536	0.5	Xu et al. (2024)
h-optimus-0	H-Optimus-0	1536	0.5	Saillard et al. (2024)
h-optimus-1	H-Optimus-1	1536	0.5	Saillard et al. (2024)
midnight	Midnight	3072	0.25, 0.5, 1.0, 2.0	Karasikov et al. (2025)

Slide-level encoders

Preset	Model	Tile encoder	Spacing (um)	Output dim	Notes
gigapath-slide	GigaPath	gigapath	0.5	768	Xu et al. (2024)
titan	TITAN	conchv15	0.5	768	Ding et al. (2024)
prism	PRISM	virchow	0.5	1280	Shaikovski et al. (2024)
moozy-slide	MOOZY	lunit	0.5	768	Kotp et al. (2026)

Patient-level encoders

Patient-level encoders aggregate multiple slide embeddings for the same patient into a single patient-level embedding. They require a patient_id column in the input manifest csv (or patient_id keys in each slide dict when using the Python API).

Preset	Model	Tile encoder	Spacing (um)	Output dim	Notes
moozy	MOOZY	lunit	0.5	768	Kotp et al. (2026)

Custom registry-backed encoders

If you want to use a model that is not shipped with slide2vec, wrap it in an encoder class and register it under a new preset name.

Tile encoder example

import torch
from torch import Tensor

from slide2vec.encoders import TileEncoder
from slide2vec.encoders import register_encoder, resolve_requested_output_variant


@register_encoder(
    "my-tile-model",
    output_variants={"default": {"encode_dim": 768}},
    default_output_variant="default",
    input_size=224,
    supported_spacing_um=0.5,
    precision="fp16",
    source="my-org/my-tile-model",
)
class MyTileModel(TileEncoder):
    def __init__(self, *, output_variant: str | None = None):
        self._output_variant = resolve_requested_output_variant(output_variant)
        self._device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        self._model = self._load_model().eval()

    def _load_model(self):
        ...

    def get_transform(self):
        ...

    def encode_tiles(self, batch: Tensor) -> Tensor:
        return self._model(batch)

    @property
    def encode_dim(self) -> int:
        return 768

    @property
    def device(self) -> torch.device:
        return self._device

    def to(self, device: torch.device | str):
        self._device = torch.device(device)
        self._model = self._model.to(self._device)
        return self

Once the module is imported, the preset is available through the existing API:

from slide2vec import Model

model = Model.from_preset("my-tile-model")

Slide encoder example

import torch
from torch import Tensor

from slide2vec.encoders import SlideEncoder
from slide2vec.encoders import register_encoder, resolve_requested_output_variant


@register_encoder(
    "my-slide-model",
    level="slide",
    tile_encoder="my-tile-model",
    tile_encoder_output_variant="default",
    output_variants={"default": {"encode_dim": 512}},
    default_output_variant="default",
    supported_spacing_um=0.5,
    precision="fp16",
    source="my-org/my-slide-model",
)
class MySlideModel(SlideEncoder):
    def __init__(self, *, output_variant: str | None = None):
        self._output_variant = resolve_requested_output_variant(output_variant)
        self._device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        self._model = self._load_model().eval()

    def _load_model(self):
        ...

    @property
    def encode_dim(self) -> int:
        return 512

    @property
    def device(self) -> torch.device:
        return self._device

    def to(self, device: torch.device | str):
        self._device = torch.device(device)
        self._model = self._model.to(self._device)
        return self

    def encode_slide(
        self,
        tile_features: Tensor,
        coordinates: Tensor | None = None,
        *,
        tile_size_lv0: int | None = None,
    ) -> Tensor:
        return self._model(tile_features)