<a target="_blank" rel="ugc" href="https://huggingface.co/alimama-creative/FLUX.1-Turbo-Alpha/blob/main/README_ZH.md">中文版Readme</a>This repository provides a 8-step distilled lora for <a target="_blank" rel="ugc" href="https://huggingface.co/black-forest-labs/FLUX.1-dev">FLUX.1-dev</a> model released by AlimamaCreative Team.<h1 id="description-ml2k811zd">Description</h1>This checkpoint is a 8-step distilled Lora, trained based on FLUX.1-dev model. We use a multi-head discriminator to improve the distill quality. Our model can be used for T2I, inpainting controlnet and other FLUX related models. The recommended guidance_scale=3.5 and lora_scale=1. Our Lower steps version will release later.<ul><li>Text-to-Image.</li></ul><a target="_blank" rel="ugc" href="https://huggingface.co/alimama-creative/FLUX.1-Turbo-Alpha/blob/main/images/T2I.png"><img src="https://huggingface.co/alimama-creative/FLUX.1-Turbo-Alpha/resolve/main/images/T2I.png" alt /></a><ul><li>With <a target="_blank" rel="ugc" href="https://huggingface.co/alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Beta">alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Beta</a>. Our distilled lora can be well adapted to the Inpainting controlnet, and the accelerated generated effect can follow the original output well.</li></ul><a target="_blank" rel="ugc" href="https://huggingface.co/alimama-creative/FLUX.1-Turbo-Alpha/blob/main/images/inpaint.png"><img src="https://huggingface.co/alimama-creative/FLUX.1-Turbo-Alpha/resolve/main/images/inpaint.png" alt /></a><h1 id="how-to-use-gdtlo98h5">How to use</h1><h2 id="diffusers-1ijki2nbp">diffusers</h2>This model can be used ditrectly with diffusers<pre><code>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]
</code></pre><h2 id="comfyui-n0tdodyng">comfyui</h2><ul><li>T2I turbo workflow: <a target="_blank" rel="ugc" href="https://huggingface.co/alimama-creative/FLUX.1-Turbo-Alpha/blob/main/workflows/t2I_flux_turbo.json">click here</a></li><li>Inpainting controlnet turbo workflow: <a target="_blank" rel="ugc" href="https://huggingface.co/alimama-creative/FLUX.1-Turbo-Alpha/blob/main/workflows/alimama_flux_inpainting_turbo_8step.json">click here</a></li></ul><h1 id="training-details-2hcvk75iv">Training Details</h1>The model is trained on 1M open source and internal sources images, with the aesthetic 6.3+ and resolution greater than 800. We use adversarial training to improve the quality. Our method fix the original FLUX.1-dev transformer as the discriminator backbone, and add multi heads to every transformer layer. We fix the guidance scale as 3.5 during training, and use the time shift as 3.Mixed precision: bf16Learning rate: 2e-5Batch size: 64Image size: 1024x1024

alimama-creative / FLUX.1-Turbo-Alpha

diffusion_pytorch_model.safetensors

v1.0

8 stepsCách sử dụng:<ul><li>•Sampling method: Euler</li><li>•Schedule type: Simple</li><li>•Sampling steps: 30</li><li>Weight: 0.8 - 1.2. Best: 0.8</li></ul>Instructions for use:<ul><li>•Sampling method: Euler a</li><li>•Schedule type: Simple</li><li>•Sampling steps: 30</li><li>Weight: 0.8 - 1.2. Best: 0.8</li></ul>In the prompt, you should use the word: Woman using the word Girl will create many body anatomy errors.<s>0-0-0-0-0-0-0-0-0-0-0-0</s>Actually, you will not need the activation keyword for it to work, but you can add it to make Flux understand faster and give better results. ⚡Note: It works well with FLUX.1-Turbo-Alpha, LORA human face. 👤✨<img src="https://models-online-persist.shakker.cloud/img/c7d573e4c3d24f97bc7b6c9b5c950091/91665a2d7674a2db4d6181122f9ffde291daf229d8d776e49d4584895d189263.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" alt />Useful and FREE resources:❤️Free server to make art with Flux: <a target="_blank" rel="ugc" href="https://www.shakker.ai/userpage/c7d573e4c3d24f97bc7b6c9b5c950091/publish?sourceId=002720">Shakker</a>✨ More FLUX LORA? List and detailed description of each LORA I implement here: <a target="_blank" rel="ugc" href="https://maitruclam.com/lora">https://maitruclam.com/lora</a>🆕 First time using FLUX? Explanation and tutorial with A1111 forge offline and Comfy UI here: <a target="_blank" rel="ugc" href="https://maitruclam.com/flux-ai-la-gi/">https://maitruclam.com/flux-ai-la-gi/</a>🛠️ How to train your LORA with Flux? My detailed instructions are here: <a target="_blank" rel="ugc" href="https://maitruclam.com/training-flux/">https://maitruclam.com/training-flux/</a>❤️ Donate me (I would be really surprised if you did that! 😄): <a target="_blank" rel="ugc" href="https://maitruclam.com/donate">https://maitruclam.com/donate</a>Find me / Contact for work on:📱 Facebook: @maitruclam4real💬 Discord: @maitruclam🌐 Web: <a target="_blank" rel="ugc" href="http://maitruclam.com">maitruclam.com</a>

FLUX.1-Turbo-Alpha

FLUX.1-Turbo-Alpha.safetensors

This checkpoint is a 8-step distilled Lora, trained based on FLUX.1-dev model.It use a multi-head discriminator to improve the distill quality. the model can be used for T2I, inpainting controlnet and other FLUX related models. The recommended guidance_scale=3.5 and lora_scale=1, Euler = Simple.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/6c8faa1f-2270-4127-b8df-1bcc230c6e1d/width=525/6c8faa1f-2270-4127-b8df-1bcc230c6e1d.jpeg" /><h3 id="key-features:-ogdjw8l6d">Key Features:</h3><ul><li>Rapid Image Generation: FLUX.1-Turbo-Alpha allows users to create images quickly, making it ideal for projects requiring fast turnaround times</li><li>Versatility in Applications: This model supports text-to-image generation and inpainting, providing users with the flexibility to create varied and intricate scenes easily.</li><li>Optimized Settings: For optimal performance, it is recommended to use a guidance scale of 3.5 and a Lora scale of 1, which allows for effective control over the output quality and style.</li><li>Future-Proofing: Upcoming versions of FLUX.1-Turbo-Alpha will include configurations that utilize fewer steps, further enhancing speed without sacrificing quality.</li></ul><h3 id="ideal-use-cases:-su6c344mt">Ideal Use Cases:</h3>FLUX.1-Turbo-Alpha is perfect for artists, designers, and content creators who need quick and high-quality image outputs. Its adaptability to various creative applications makes it a valuable addition to any toolkit, whether for professional projects or personal exploration.With FLUX.1-Turbo-Alpha, you can harness the power of AI to elevate your creative projects, transforming ideas into stunning visuals in mere moments. Thanks to the original creator : alimama-creative Original HuggingFace Repo : <a target="_blank" rel="ugc" href="https://huggingface.co/alimama-creative/FLUX.1-Turbo-Alpha">https://huggingface.co/alimama-creative/FLUX.1-Turbo-Alpha</a>

This repository provides a 8-step distilled lora for <a target="_blank" rel="ugc noopener noreferrer nofollow" href="https://huggingface.co/black-forest-labs/FLUX.1-dev">FLUX.1-dev</a> <a target="_blank" rel="ugc noopener noreferrer nofollow" href="https://huggingface.co/alimama-creative/FLUX.1-Turbo-Alpha">model</a> released by AlimamaCreative Team.This checkpoint is a 8-step distilled Lora, trained based on FLUX.1-dev model. We use a multi-head discriminator to improve the distill quality. Our model can be used for T2I, inpainting controlnet and other FLUX related models. The recommended guidance_scale=3.5 and lora_scale=1. Our Lower steps version will release later.

Files with SHA256: 77f7523a5e9c3da6cfc730c6b07461129fa52997ea06168e9ed5312228aa0bff