import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("black-forest-labs/FLUX.2-klein-base-9B", dtype=torch.bfloat16, device_map="cuda")
pipe.load_lora_weights("sololo-xyz/VG20-Rebecca")
prompt = "-"
image = pipe(prompt).images[0]VG20 Rebecca SoloLoRA F2K9v1
- Prompt
- -
- Prompt
- -
- Prompt
- -
- Prompt
- -
- Prompt
- -
- Prompt
- -
- Prompt
- -
- Prompt
- -
Original link: https://sololo.xyz/model?id=586
Model description
Rebecca, a virtual woman.
Experimental, almost there, but still a long way to go.
Both base 9B and 9B are supported.
This is my first klein 9B LoRA. I tried training on 4B several times before but ended up putting it aside for now. From a LoRA training perspective, 9B works much better than 4B.
What surprised me was that it turned out to be even harder to train than ZIT. I didn’t expect that, especially since one is a base model and the other is a turbo model.
That actually makes me even more excited about Z Image. Hopefully it won’t be a letdown when it finally arrives.
The main issues with this LoRA at the moment are:
- Some facial distortion.
- The face likeness isn’t quite where I want it to be yet.
- Facial consistency is still not very stable. Also, I’m still getting familiar with how klein behaves when generating images, and I haven't fully figured out what kind of prompts work best yet.
My rough impression is that klein goes further than ZIT in terms of realism, but it seems to fall a bit short when it comes to aesthetics.
That’s it for now. I’ll keep experimenting.
Download model
Download them in the Files & versions tab.
- Downloads last month
- 12
Model tree for sololo-xyz/VG20-Rebecca
Base model
black-forest-labs/FLUX.2-klein-base-9B