KiwiMate-Image-Mini-1.0

KiwiMate-Image-Mini-1.0 is a LoRA fine-tune of black-forest-labs/FLUX.2-klein-4B, a compact 4B-parameter text-to-image diffusion model. It delivers fast, high-quality image generation while staying lightweight enough to run on consumer GPUs.

Files

File	Description
KiwiMate-Image-Mini-1.0.safetensors	Final LoRA weights (92 MB)
KiwiMate-Image-Mini-1.0_000000250 – _000002750.safetensors	Training step checkpoints
KiwiMate-Image-Mini-1.0_merged.safetensors	Full merged model (7.75 GB), no base model required

Usage

LoRA

Load the LoRA on top of the base FLUX.2-klein-4B model:

from diffusers import FluxKleinPipeline
import torch

pipe = FluxKleinPipeline.from_pretrained(
    "black-forest-labs/FLUX.2-klein-4B",
    torch_dtype=torch.bfloat16,
).to("cuda")

pipe.load_lora_weights("KiwiMate/KiwiMate-Image-Mini-1.0", weight_name="KiwiMate-Image-Mini-1.0.safetensors")

image = pipe("your prompt here").images[0]

Merged model

For a standalone setup without the base model, use the merged checkpoint:

pipe = FluxKleinPipeline.from_pretrained(
    "KiwiMate/KiwiMate-Image-Mini-1.0",
    torch_dtype=torch.bfloat16,
).to("cuda")

Sample Outputs

License

Apache 2.0