This model is based on the Concept sliders LoRA model trained by FLUX.1D, which makes the characters in the generated pictures gradually change from unhappy to happy.

weight: -5.0 to 5.0 （Not limited to this interval, you can try other ranges.）

target_prompt: "photo of a person"
positive_prompt: "photo of a person, smiling, happy"
negative_prompt: "photo of a person, frowning"

Code used for training: https://github.com/SWHL/flux-sliders

Use this model:

from datetime import datetime
from pathlib import Path

import torch
from diffusers import FluxPipeline

lora_path = "flux-smiling_sliders_latest.safetensors"
pipe = FluxPipeline.from_pretrained("models/FLUX.1-dev", torch_dtype=torch.bfloat16)
pipe.to("cuda")
pipe.load_lora_weights(lora_path)

time_stamp = datetime.strftime(datetime.now(), "%Y-%m-%d-%H-%M-%S")
save_dir = Path("outputs") / time_stamp
save_dir.mkdir(parents=True, exist_ok=True)

scales = (-5, -2.5, 0, 2.5, 5)
prompt = "male person"

for scale in scales:
    out = pipe(
        prompt=prompt,
        guidance_scale=3.5,
        height=512,
        width=512,
        num_inference_steps=25,
        joint_attention_kwargs={"scale": scale * 1 / 16},
        generator=torch.Generator().manual_seed(42),
    ).images[0]

    save_img_path = save_dir / f"{time_stamp}_scale_{scale}.jpg"
    out.save(save_img_path)

Enjoy it!

⚠️ Notion:

• The current model does not support direct use of comfy.

• Since the model has been converted, you need to try the slider scale multiple times to select the range you need.

• ComfyUI will be supported later, so stay tuned.