Instructions to use Alpha-VLLM/Lumina-Image-2.0 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use Alpha-VLLM/Lumina-Image-2.0 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("Alpha-VLLM/Lumina-Image-2.0", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Inference
- Notebooks
- Google Colab
- Kaggle
- Local Apps
- Draw Things
- DiffusionBee
Switch to FLUX.1-schnell vae files to avoid FLUX.1-dev license restrictions
FLUX.1-dev is under a strictly non-commercial license, whereas FLUX.1-schnell is Apache 2.0 open source.
As far as I'm aware, the actual VAE weights is the same between both versions, but for some reason Lumina-Image-2.0 is using the FLUX.1-dev copy of the VAE files, meaning it is under a non-commercial license. You can see in the config file it is using the FLUX.1-dev version of the VAE files: https://huggingface.co/Alpha-VLLM/Lumina-Image-2.0/blob/main/vae/config.json -> "_name_or_path": "black-forest-labs/FLUX.1-dev"
Can the VAE files get switched over to the FLUX.1-schnell version so it resolves any potential licensing issues? It shouldn't be a complicated switch since the VAE itself has the same weights as far as I'm aware, it's just the config file that differs. For good measure I would also recommend updating the link on the GitHub page to the FLUX.1-schnell VAE files as that also currently links to FLUX.1-dev.
I feel it is best to avoid any references to or usage of FLUX.1-dev files given its non-commercial license restrictions.
Good idea! We will seriously consider your recommendation and update it as soon as possible. Stay tuned!
Excellent idea, looking forward for update.