GeoCore-9B

Towards Geo-Aware Generative Foundation Models in Earth Observation

Jeonghyeok Do  ·  Munchurl Kim

Korea Advanced Institute of Science and Technology (KAIST)

A 9-billion-parameter generative foundation model for Earth Observation, trained from scratch exclusively on EO data. GeoCore-9B is a Flow Matching Diffusion Transformer that conditions generation on text and continuous geospatial metadata — ground sample distance (GSD), latitude and longitude.

Model details

Parameters	9.24 B
Weights	EMA, bfloat16, sharded safetensors
Architecture	Flow Matching DiT — 8 double-stream + 24 single-stream blocks, hidden 4096, 32 heads
Conditioning	CLIP + T5 text embeddings, GSD, latitude, longitude
Training data	Git-10M
Training	300K steps, global batch 1024, AdamW lr 1e-4, bf16, DeepSpeed ZeRO-2, 8x B200
Resolution	256x256

Training used a Geospatial Semantic Alignment (GSA) loss that aligns intermediate DiT features (block 8) with a frozen DINOv3-Sat teacher, weighted by mu = 0.5. GSA is training-only — the teacher and projection head are not needed for inference, and the exported weights add no overhead.

Usage

These weights use a custom Flux2 architecture, so load them with the model definition from the GitHub repository rather than a stock diffusers pipeline.

import torch
from huggingface_hub import snapshot_download

from models.flux2 import Flux2, GeoCore9BParams
from inference import load_state_dict

path = snapshot_download("JeonghyeokDo/GeoCore-9B")

model = Flux2(GeoCore9BParams()).to("cuda", torch.bfloat16)
model.load_state_dict(load_state_dict(path), strict=True)
model.eval()

Sampling, including the Euler flow-matching sampler and classifier-free guidance over both text and metadata, is provided by inference.py:

python inference.py \
    --ckpt /path/to/GeoCore-9B \
    --vae /path/to/ae.safetensors \
    --prompt "A satellite view of a highly dense urban city with towering skyscrapers" \
    --lon 126.97 --lat 37.56 --res 0.0 \
    --num-samples 4 --out samples/

Conditioning inputs

• res — resolution index, defined as 17 - z for Google XYZ tile zoom z. res = 0 is roughly 1.2 m/px at the equator; each +1 doubles the GSD.
• lon, lat — degrees.
• Any field set to -999.0 falls back to the model's learned null embedding for that field, so metadata is fully optional.

Additional requirements

The Flux.2 VAE (ae.safetensors, Black Forest Labs) is required to decode latents and is not included here. CLIP and T5 text encoders are downloaded from the Hub at runtime.

Limitations

• Trained on 256x256 RGB optical imagery; other resolutions and sensor modalities require adaptation.
• Git-10M coverage is uneven across the globe, so generation quality varies by region.
• Geospatial conditioning reflects correlations in the training corpus and is not a substitute for real observations of a location.

Citation

@article{do2026geocore,
  title   = {GeoCore-9B: Towards Geo-Aware Generative Foundation Models in Earth Observation},
  author  = {Do, Jeonghyeok and Kim, Munchurl},
  year    = {2026}
}