FLUX.1-Turbo-Alpha Open-Source Image Generation Model - Accelerate the Conversion of Text into Exquisite Images

FLUX.1 Turbo Alpha

Developed by alimama-creative

8-step distilled LoRa based on FLUX.1-dev model for accelerating text-to-image generation

Image Generation EnglishOpen Source License:Other #8-step distillation acceleration #Text-to-image generation #Inpainting control network adaptation

Downloads 37.49k

Release Time : 10/12/2024

Model Overview

This is an 8-step distilled LoRa trained on the FLUX.1-dev model, using multi-head discriminators to improve distillation quality, suitable for tasks like text-to-image generation and inpainting control networks

Model Features

8-step distillation acceleration

Significantly reduces inference steps through 8-step distillation, improving generation speed

Multi-head discriminator

Uses multi-head discriminator technology to enhance distillation quality

Compatible with control networks

Well-adapted for inpainting control networks while maintaining original output quality

Model Capabilities

Text-to-image generation

Image inpainting

Fast inference

Use Cases

Creative design

Ad creative generation

Quickly generate high-quality ad creative images

Example image showing a combination of vehicle and sloth

Image inpainting

Perform image inpainting and editing with control networks

Example image showing inpainting effects

🚀 FLUX.1-Turbo-Alpha

This repository offers an 8-step distilled LoRA for the FLUX.1-dev model released by the Alimama Creative Team. It can be applied to text-to-image generation, inpainting with ControlNet, and other FLUX-related models.

Chinese Version Readme

✨ Features

Text-to-Image: Capable of generating high - quality images from text prompts.
Inpainting with ControlNet: Works well with alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Beta. The distilled LoRA can adapt to the inpainting ControlNet, and the accelerated generation effect closely follows the original output.

📦 Installation

No specific installation steps are provided in the original document.

💻 Usage Examples

Basic Usage

This model can be used directly with diffusers.

import torch
from diffusers.pipelines import FluxPipeline

model_id = "black-forest-labs/FLUX.1-dev"
adapter_id = "alimama-creative/FLUX.1-Turbo-Alpha"

pipe = FluxPipeline.from_pretrained(
  model_id,
  torch_dtype=torch.bfloat16
)
pipe.to("cuda")

pipe.load_lora_weights(adapter_id)
pipe.fuse_lora()

prompt = "A DSLR photo of a shiny VW van that has a cityscape painted on it. A smiling sloth stands on grass in front of the van and is wearing a leather jacket, a cowboy hat, a kilt and a bowtie. The sloth is holding a quarterstaff and a big book."
image = pipe(
            prompt=prompt,
            guidance_scale=3.5,
            height=1024,
            width=1024,
            num_inference_steps=8,
            max_sequence_length=512).images[0]

Advanced Usage

ComfyUI

T2I turbo workflow: click here
Inpainting controlnet turbo workflow: click here

🔧 Technical Details

The model is trained on 1M open - source and internal - sourced images, with an aesthetic score of 6.3+ and a resolution greater than 800. Adversarial training is used to improve the quality. The method fixes the original FLUX.1 - dev transformer as the discriminator backbone and adds multi - heads to every transformer layer. During training, the guidance scale is fixed at 3.5, and the time shift is set to 3.

Property	Details
Mixed precision	bf16
Learning rate	2e - 5
Batch size	64
Image size	1024x1024

📄 License

License: other
License Name: flux-1-dev-non-commercial-license
License Link: https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご