Instructions to use blurgy/CoMPaSS-FLUX.1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use blurgy/CoMPaSS-FLUX.1 with Diffusers:
pip install -U diffusers transformers accelerate
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] - Inference
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
- DiffusionBee
Is this improving the quality of flux altogether?
I noticed that there's a level of quality this lora is ameliorating in flux dev beyond just spatial. Is this something anyone else is noticing?
Hi @omarei , thanks you for the insightful comment!
We also observed that the improvements extend beyond spatial understanding, enhancing aspects like image fidelity and text-image alignment. This is something we quantitatively analyze in our paper (Table 2). We offered a preliminary conjecture for this phenomenon in Section 4.2:
We conjecture that in base models, spatial terms are often entangled with unrelated semantics due to flawed data. By disentangling these terms, CoMPaSS may also help the model better understand other aspects of language, resulting in these broader improvements.