FLUX-Hyperscale-fused

FLUX-Hyperscale-fused (dev)

<Gallery />

This model is a fused model that mixes and fuses 5 high-quality, fine-tuned adapters in a certain ratio.

The code used for model fusing is as follows:

import torch
from diffusers import AutoPipelineForText2Image

pipe = AutoPipelineForText2Image.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16).to("cuda")

pipe.load_lora_weights(".", weight_name="ethereal-dark.safetensors", adapter_name="ethereal-dark")
pipe.load_lora_weights(".", weight_name="mythic-fantasy.safetensors", adapter_name="mythic-fantasy")
pipe.load_lora_weights(".", weight_name="mj-whisper.safetensors", adapter_name="mj-whisper")
pipe.load_lora_weights(".", weight_name="elden-ring.safetensors", adapter_name="elden-ring")
pipe.load_lora_weights(".", weight_name="retro-anime.safetensors", adapter_name="retro-anime")
pipe.set_adapters(["ethereal-dark", "mythic-fantasy", "mj-whisper", "elden-ring", "retro-anime"], adapter_weights=[0.3, 0.9, 0.5, 0.25, 0.6])
pipe.fuse_lora(adapter_names=["ethereal-dark", "mythic-fantasy", "mj-whisper", "elden-ring", "retro-anime"], lora_scale=1.0)

pipe.unload_lora_weights()

pipe.push_to_hub("minpeter/FLUX-Hyperscale-fused")

The original model can be found at the link below

Creating images using a diffusers

import torch
from diffusers import AutoPipelineForText2Image

pipe = AutoPipelineForText2Image.from_pretrained("minpeter/FLUX-Hyperscale-fused", torch_dtype=torch.bfloat16).to("cuda")

prompt = "orange tabby cat, curled up asleep, peaceful expression, lying in a warm patch of sunlight, on a cozy living room rug, warm lighting, soft focus background"

image = pipe(
    prompt=prompt,
    guidance_scale=3.5,
    height=1024,
    width=1024,
    num_inference_steps=30,
).images[0]

image.save("s2.png")