FLUX.2 Klein 9B слитная модель с SNOFS v1.4
2 min read
This model is a merge of the base distilled FLUX.2 Klein (9B) AI model and the SNOFS v1.4 adapter, baked directly into the transformer weights. Since the LoKr (Kronecker low-rank adapters) format is not supported by standard loading methods in the diffusers library, the authors merged the weights manually on the CPU. The result is a ready-to-use image generation model with the SNOFS style already built-in. This is useful for anyone wanting to use this combination in code via diffusers without applying adapters manually.
Key features
- Built-in SNOFS v1.4 adapter with a strength of 1.0 directly in the transformer weights.
- Full compatibility with connecting additional LoRA adapters on top of this model.
- Precise merging of 144 target modules without errors or incorrect weights.
- The base FLUX.2 Klein 9B model features distillation and fast generation.
What you can do
- Generate high-quality images from a text prompt in the diffusers library.
- Use the model in text-to-image generation pipelines with a minimal number of steps.
- Set up generation with a fixed SNOFS style without needing to load the adapter weights separately.
Example
import torch
from diffusers import Flux2KleinPipeline
pipe = Flux2KleinPipeline.from_pretrained(
"feltrockni/flux2-klein-9b-snofs-v1-4-merged", torch_dtype=torch.bfloat16
).to("cuda")
image = pipe(
prompt="...",
num_inference_steps=8,
guidance_scale=1.0,
generator=torch.Generator(device="cpu").manual_seed(0),
).images[0]Read more
DetailsBase model: black-forest-labs/FLUX.2-klein-9B (9B parameters)
Library: diffusers (requires main branch version)
Features: guidance_scale is ignored due to distillation, 8 generation steps recommended










Comments