FLUX12B震撼发布:SD创始团队,23G迄今最大开源文生图模型

FLUX模型简介

8月1这天开源文生图模型迈入了有一个大里程碑,黑森林实验室(一家已完成3100万美元的种子轮系列融资)的12B文生图大模型:FLUX震撼发布。这是迄今为止最大的文生图开源模型。这也是目前高质量的文生图模型,FLUX.1文生图模型套件,为文本到图像合成定义了新的最先进水平,在图像细节、提示遵循、风格多样性和场景复杂性方面树立了新的基准标准。在可访问性和模型能力之间取得平衡,FLUX.1一共有三个变体:FLUX.1 [pro]、FLUX.1 [dev] 和 FLUX.1 [schnell]:

  • FLUX.1 [dev] :基础模型,开源且拥有非商业许可,供社区在此基础上进行构建。https://huggingface.co/black-forest-labs/FLUX.1-dev/tree/main
  • FLUX.1 [schnell] :基本模型的精简版本,蒸馏版本,23.8 GB,运行速度最高可提高 10 倍。Apache 2 许可。https://huggingface.co/black-forest-labs/FLUX.1-schnell/blob/main/flux1-schnell.sft
  • FLUX.1 [pro] :官方闭源版本,可以通过 API提供服务使用。API地址:https://replicate.com/black-forest-labs/flux-pro、https://docs.bfl.ml/
FLUX12B震撼发布:SD创始团队,23G迄今最大开源文生图模型
FLUX12B震撼发布:SD创始团队,23G迄今最大开源文生图模型

FLUX模型评估

FLUX.1 [pro] 和 [dev] 在以下每个方面超越了流行的模型,如 Midjourney v6.0DALL·E 3(HD)SD3-Ultra:视觉质量、提示遵循、尺寸/方面变化、排版和输出多样性。FLUX.1 [schnell] 是迄今为止最先进的几步模型,不仅超越了其同类竞争对手,还超越了像 Midjourney v6.0 和 DALL·E 3(HD)这样的强大非蒸馏模型。FLux模型专门针对预训练中保持整个输出多样性进行了微调。与当前的最先进技术相比,它们提供了显著改进的可能性,如下所示。

FLUX12B震撼发布:SD创始团队,23G迄今最大开源文生图模型

所有FLUX.1模型变体都支持 0.1 和 2.0 百万像素的多种纵横比和分辨率,如下例所示。

FLUX12B震撼发布:SD创始团队,23G迄今最大开源文生图模型

FLUX.1 支持的纵横比和分辨率示例

另外黑森林实验室团队提到在Flux文生图模型之后,计划会进一步进军文生视频领域。不久的将来我们可能会在文生视频领域的重大里程碑震撼发布。

Flux模型体验

Flux ComfyUI安装

ComfyUI的最新版本已支持Flux模型的运行,仅需将ComfyUI更新到最新版本即可。ComfyUI官方文档:https://comfyanonymous.github.io/ComfyUI_examples/flux/。

  • • 需要下载模型Flux1-schnell.sft放置到目录ComfyUI/models/unet/ 下。下载地址:https://huggingface.co/black-forest-labs/FLUX.1-schnell/blob/main/flux1-schnell.sft。显存不足24,尽量采用8为量化版本:https://hf-mirror.com/maximsobolev275/flux-fp8-schnell/resolve/main/flux1-schnell_fp8_unet.safetensors?download=true
  • • 需要下载VAE模型ae.sft放置到目录ComfyUI/models/vae 下。下载地址:https://huggingface.co/black-forest-labs/FLUX.1-schnell/blob/main/ae.sft
  • • 需要下载T5文本编码模型t5xxl_fp8_e4m3fn.safetensors放置到目录ComfyUI/models/clip 下。下载地址:https://huggingface.co/comfyanonymous/flux_text_encoders/tree/main

工作流界面

该工作流已发布到LIBLIBAI上可自由下载使用:https://www.liblib.art/modelinfo/4e5daf0cf50542199e5bd3b5174b168e。

FLUX12B震撼发布:SD创始团队,23G迄今最大开源文生图模型

01. 牛魔女

A Korean beautiful idol with horns and beautiful face, in a black outfit in the style of James Jean, against a flat red background, with cinematic lighting, in a minimalistic design, with dark contrast
FLUX12B震撼发布:SD创始团队,23G迄今最大开源文生图模型

02. 光影

masterpiece, best quality, 1girl ((pure gradient background, )), long hair, floating hair, blush, looking at viewers, happy, ((front)),(upper body), (studio light), soft light, dark style, night style
FLUX12B震撼发布:SD创始团队,23G迄今最大开源文生图模型

03. 旗袍

(look at viewer:2), 1 girl, solo,chinese dress, cheongsam, earrings, Chinese roll, smile, lips, light green background,sfw,8k high definition, 35 mm film photography, photo realistic, insanely detailed, intricate, elegant,  best quality, ultra-detailed, masterpiece, finely detail, highres, 8k wallpaper
FLUX12B震撼发布:SD创始团队,23G迄今最大开源文生图模型

04. 动漫

(Animation style:1.3),a female character with long, flowing hair that appears to be made of ethereal, swirling patterns resembling the Northern Lights or Aurora Borealis. The background is dominated by deep blues and purples, creating a mysterious and dramatic atmosphere. The character's face is serene, with pale skin and striking features. She wears a dark-colored outfit with subtle patterns. The overall style of the artwork is reminiscent of fantasy or supernatural genres
FLUX12B震撼发布:SD创始团队,23G迄今最大开源文生图模型

05. 摄影构图-全身镜头+正面视角

front view, full body shot),1girl, solo, realistic, chinese girl,(cowboy shot:1.2), real life location,tiny pink shirts, midriff, short skirt, smile, cute,, thin short waist, large pelvis,, (high quality:1.4), (photorealistic:1.6), 8k, uhd, highres, absurdres, professional photo, highly detailed, detailed skin
FLUX12B震撼发布:SD创始团队,23G迄今最大开源文生图模型

06. 莲花

A lotus flower, close to the sun, triple exposure, fantastic illustrations
FLUX12B震撼发布:SD创始团队,23G迄今最大开源文生图模型

07. 小黄人

a group of six minions in a yellow boat on a river in Paris, France. The boat is floating on the water, with the Eiffel Tower in the background. The minions are all facing the same direction and appear to be happy and excited. They are all wearing blue overalls and have big smiles on their faces. The river is lined with buildings on both sides, and there are pink and purple flowers floating in the water. The sky is blue and the overall mood of the image is cheerful and playful.
FLUX12B震撼发布:SD创始团队,23G迄今最大开源文生图模型

08. 大猫

hug cat,1girl,solo,(1 orange Giant cat:1.3),red dress,indoor,the cat stands, the girl is next to the cat, 8k high definition, 35 mm film photography
FLUX12B震撼发布:SD创始团队,23G迄今最大开源文生图模型

09. 冰雕-文学讲坛

three statues of Chinese characters standing on a stage with a blue background. The statues are made of ice and are intricately carved with detailed features. The characters are dressed in traditional Chinese clothing, with long robes and hats. They are standing in a line, facing towards the left side of the image. The figure on the left is holding a book in his hands, while the figure in the middle is standing with his hands clasped in front of him. All three figures have a serious expression on their faces and appear to be in a contemplative pose. The background is a gradient of blue and white, with rays of light shining down on the figures
FLUX12B震撼发布:SD创始团队,23G迄今最大开源文生图模型

10. 沙滩

Beautiful woman in white summer dress standing at the beach, dress and hair flying in the wind, sun glasses, summer day, sunny, detailed sand textures, piercing eyes, perfect delicate face, perfect lips, Oil Painting, expressive brushwork, luminous color palette, and delicate details, Miki Asai Macro photography, close-up, hyper detailed, trending on artstation, sharp focus, studio photo, intricate details, highly detailed, by greg rutkowski

FLUX12B震撼发布:SD创始团队,23G迄今最大开源文生图模型

 

声明:内容均采集自公开的网站等各类媒体平台,若收录的内容侵犯了您的权益,请联系邮箱,本站将第一时间处理。
百科

24小时虚拟人直播,盘点7款国内3D数字人定制和24小时直播平台

2024-8-2 11:02:12

百科

AI图像生成平台,LibLib AI使用方法教程与免费试用入口

2024-8-3 10:23:06

搜索