CoMPaSS: Enhancing Spatial Understanding in Text-to-Image Diffusion Models
Paper • 2412.13195 • Published
import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", dtype=torch.bfloat16, device_map="cuda")
pipe.load_lora_weights("blurgy/CoMPaSS-FLUX.1")
prompt = "a photo of a laptop above a dog"
image = pipe(prompt).images[0]CoMPaSS-FLUX.1
[Project Page] [code] [arXiv]
- Prompt
- a photo of a laptop above a dog
- Prompt
- a photo of a bird below a skateboard
- Prompt
- a photo of a horse to the left of a bottle
Model description
CoMPaSS-FLUX.1
A LoRA adapter that enhances spatial understanding capabilities of the FLUX.1 text-to-image diffusion model. This model demonstrates significant improvements in generating images with specific spatial relationships between objects.
Model Details
- Base Model: FLUX.1-dev
- LoRA Rank: 16
- Training Data: SCOP dataset (curated from COCO)
- File Size: ~50MiB
- Framework: Diffusers
- License: Non-Commercial (see ./LICENSE)
ComfyUI Support
We provide a custom node with examples at comfyui-node-impl. Use the ComfyUI-compatible LoRA checkpoint comfyui-checkpoint to get started.
Intended Use
- Generating images with accurate spatial relationships between objects
- Creating compositions that require specific spatial arrangements
- Enhancing the base model's spatial understanding while maintaining its other capabilities
Performance
Key Improvements
- VISOR benchmark: +98% relative improvement
- T2I-CompBench Spatial: +67% relative improvement
- GenEval Position: +131% relative improvement
- Maintains or improves base model's image fidelity (lower FID and CMMD scores than base model)
Using the Model
See our GitHub repository to get started.
Effective Prompting
The model works well with:
- Clear spatial relationship descriptors (left, right, above, below)
- Pairs of distinct objects
- Explicit spatial relationships (e.g., "a photo of A to the right of B")
Training Details
Training Data
- Built using the SCOP (Spatial Constraints-Oriented Pairing) data engine
- ~28,000 curated object pairs from COCO
- Enforces criteria for:
- Visual significance
- Semantic distinction
- Spatial clarity
- Object relationships
- Visual balance
Training Process
- Trained for 24,000 steps
- Batch size of 4
- Learning rate: 1e-4
- Optimizer: AdamW with β₁=0.9, β₂=0.999
- Weight decay: 1e-2
Evaluation Results
| Metric | FLUX.1 | +CoMPaSS |
|---|---|---|
| VISOR uncond (⬆️) | 37.96% | 75.17% |
| T2I-CompBench Spatial (⬆️) | 0.18 | 0.30 |
| GenEval Position (⬆️) | 0.26 | 0.60 |
| FID (⬇️) | 27.96 | 26.40 |
| CMMD (⬇️) | 0.8737 | 0.6859 |
Citation
If you use this model in your research, please cite:
@inproceedings{zhang2025compass,
title={CoMPaSS: Enhancing Spatial Understanding in Text-to-Image Diffusion Models},
author={Zhang, Gaoyang and Fu, Bingtao and Fan, Qingnan and Zhang, Qi and Liu, Runxing and Gu, Hong and Zhang, Huaqi and Liu, Xinguo},
booktitle={ICCV},
year={2025}
}
Contact
For questions about the model, please contact blurgy@zju.edu.cn
Download model
Weights for this model are available in Safetensors format.
Download them in the Files & versions tab.
- Downloads last month
- 69
Model tree for blurgy/CoMPaSS-FLUX.1
Base model
black-forest-labs/FLUX.1-dev