AttriCtrl

📖 Introduction

Our method enables fine-grained control over the intensity of semantic attributes in diffusion models through a plug-and-play value encoder.
Unlike existing text encoders, which cannot interpret numeric intensity or continuous values, AttriCtrl bridges this gap and allows precise, interpretable adjustments of aesthetic attributes.

🎛️ Effect

Examples of controlling individual aesthetic attributes.

🔗 Applications

Demonstrations of seamless integration with other frameworks.

---

AttriCtrl is lightweight, model-agnostic, and achieves continuous controllability without modifying the underlying diffusion backbone.

Inference Code

git clone https://github.com/CD22104/AttriCtrl.git
cd AttriCtrl
pip install -e .

import torch
from diffsynth.pipelines.flux_image_new import FluxImagePipeline, ModelConfig


pipe = FluxImagePipeline.from_pretrained(
    torch_dtype=torch.bfloat16,
    device="cuda",
    model_configs=[
        ModelConfig(model_id="black-forest-labs/FLUX.1-dev", origin_file_pattern="flux1-dev.safetensors"),
        ModelConfig(model_id="black-forest-labs/FLUX.1-dev", origin_file_pattern="text_encoder/model.safetensors"),
        ModelConfig(model_id="black-forest-labs/FLUX.1-dev", origin_file_pattern="text_encoder_2/"),
        ModelConfig(model_id="black-forest-labs/FLUX.1-dev", origin_file_pattern="ae.safetensors"),
        ModelConfig(model_id="DiffSynth-Studio/AttriCtrl-FLUX.1-Dev", origin_file_pattern="models/detail.safetensors")
    ],
)

for i in [0.1, 0.3, 0.5, 0.7, 0.9]:
    image = pipe(prompt="a cat on the beach", seed=2, value_controller_inputs=[i])
    image.save(f"value_control_{i}.jpg")