AlbedoBase XL(SFW 和 NSFW)不需要精炼机,含有VAE。
目标
Stable Diffusion XL 拥有 35 亿个参数(不包括 Refiner),比 SD v1.5 版本大约多出 3.6 倍。我相信这不只是一个数字,而是一个可以带来性能显著提升的数字。
我们已经意识到 SD v1.5 的整体性能已经提升到超乎想象的程度,这要归功于我们社区的大量贡献。因此,我正在努力完成这个 AlbedoBase XL 模型,以便在这个 XL 版本中也能最佳地重现 v1.5 中的性能改进。
我的目标是直接测试所有公开上传到 Civitai 的 Checkpoint 和 LoRA 的性能,并仅合并经过多次筛选后被判定为最佳的资源。这将超越 Midjourney 等公司的图像生成 AI 的性能。
截至目前,AlbedoBase XL v3.1 Large 已合并大约 200 个选定检查点和 251 个 LoRA。
v3.1-large
• 使用 V3 中采用的递归脚本合并了 50 多个选定的 SDXL 模型的最新版本。
作品参照
(masterpiece, best quality, score_9, score_8_up, ultra high resolution photo, girl has beautiful hand model hands:1.1),
A breathtakingly beautiful young woman gracefully floats alone before a colossal space station window, captured with unparalleled photographic realism using professional medium format precision. Her visage, embodying flawless symmetry and timeless elegance, exudes raw emotional vulnerability as perfectly crystalline tears drift weightlessly before her expressive eyes. These eyes, deep pools of unspoken longing, fixate intensely on Earth, a vivid blue orb suspended within the star-strewn cosmos beyond the transparent barrier, conveying profound inner turmoil. Her exquisitely pale hands press gently yet deliberately against the immaculate glass, fingers arranged in an eloquent, meticulously crafted gesture of yearning and connection. Silky hair flows effortlessly in zero gravity, framing her impeccable features with graceful precision. Cinematic lighting harmoniously blends Earth's ambient azure radiance with the subtle shimmer of distant starlight, meticulously sculpting every nuance of her serene yet troubled expression. The stark contrast between the cold, intricate mechanical surroundings and her warm, human presence amplifies the emotional resonance, while flawless photographic technique captures each minute detail with stunning lifelike clarity, from the delicate tremor in her softly parted lips to the intricate interplay of light across the suspended teardrops. Every element is meticulously refined, embodying the essence of craftsmanship through mathematical precision and artistic excellence. This portrait transcends digital artifice, achieving pure photographic authenticity and immortalizing a singular moment where intimate human emotion meets the vast, serene solitude of the cosmos. (hand model hands:1.25)
Negative prompt: (worst quality, normal quality, score_3, score_4, dirty_part:1.1), watermark_text, dirty_text, optical_illusion_photo, vague_fingertip
Steps: 72, CFG scale: 9.5, Sampler: DPM++ 2M SDE Heun, Seed: 9, Size: 720x1280, Model: albedobaseXL_v31Large, Version: f2.0.1v1.10.1-1.10.1, sag_scale: 0.5, sag_enabled: True, Schedule type: Exponential, sag_threshold: 1, sag_blur_sigma: 2
(realism:1.8), (Cel-shading:1.3), (in ARTIST style:1.4), (masterpiece:1.7),
(mechanical legs vending machine:2), solo, outdoors, sky, cloud, city, vending machine focus, scenery, (mechanical legs vending machine), (legged vending machine), (multiple legs),
(white neon), (orange neon), graffiti, Traction City, white paint, (paint scratches:2), (metal scratches:2),
(ultra sharp focus:2), (detailed windows:2), (thin lines:2), (reflections:2),
(best quality:1.8),
<lora:Apparatus_-_Traction_City_Walkers_Vehicles:0.8>
<lora:glowneon_xl_v1:1>
Negative prompt: (worst quality:1.8), (low quality:1.8), (normal quality:1.8), lowres, blurry, blur, fog:2, sepia, monochrome,
Steps: 40, CFG scale: 7, Sampler: Euler a, Seed: 2248831936, Size: 850x1350, Model: albedobaseXL_v31Large, Tiling: True, Version: f2.0.1v1.10.1-previous-564-g77464215, Model hash: bf3ab78614, Schedule type: Automatic, Variation seed: 3460687190, Variation seed strength: 0.54, Clip skip: 2
(32k uhd resolution, extremely intricate best detailed, masterpiece, best quality, score_9, score_8_up, architectural photography:1.1),
A breathtaking interior view of a minimalist luxury treehouse overlooking a vibrant solarpunk forest village. Massive panoramic windows frame the sunrise view from canopy height, with golden light filtering through distant tree crowns. The spacious interior features elegant earth-toned furnishings that complement the living plant walls. Gentle breeze animates falling leaves outside the windows, creating a dynamic natural scene. The design emphasizes clean lines and open space while incorporating organic elements and sustainable materials. Perfect morning light creates a hopeful, uplifting atmosphere. (Interior design, natural lighting, architectural detail, perfect perspective:1.1)
Negative prompt: (scuffed, worst quality, normal quality, score_3, score_4:1.1), watermark, text, cluttered space, artificial looking plants, poor lighting, flat perspective, unrealistic architecture
Steps: 80, CFG scale: 4, Sampler: DPM++ 3M SDE, Seed: 9, Size: 720x1280, Model: albedobaseXL_v31Large, Version: f2.0.1v1.10.1-1.10.1, freeu_b1: 1.3, freeu_b2: 1.4, freeu_s1: 0.9, freeu_s2: 0.2, freeu_end: 0.02, sag_scale: 0.25, freeu_start: 0, sag_enabled: True, Schedule type: Exponential, freeu_enabled: True, sag_threshold: 0.5, sag_blur_sigma: 1
作者: | albedobond |
最后更新: | 2024年2月5日 |
文件类型: | CHECKPOINT MERGE |
基本模型: | SDXL 1.0 |
tag标签: | 动漫 真人 3d 2d 游戏 角色 动物 幻想 |
模型版本: | v2.1 |
使用介绍
AlbedoBase XL (SFW&NSFW)
不需要refiner,包含VAE。
将负面提示留空通常会带来最好的质量。
目标
Stable Diffusion XL 拥有 66 亿个参数,比 SD v1.5 版本多出约 6.6 倍。我相信这不仅仅是一个数字,而是一个可以带来性能显着提升的数字。
我们已经有一段时间意识到,由于我们社区的爆炸性贡献,SD v1.5 的整体性能已经超出了想象。因此,我正在努力完成这个 AlbedoBase XL 模型,以便在此 XL 版本中最佳地再现 v1.5 中出现的性能改进。
我的目标是直接测试所有公开上传到 Civitai 的 Checkpoint 和 LoRA 的性能,并仅合并经过多次过滤器后判断为最佳的资源。这将超越Midjourney等公司的图像生成AI的性能。
到目前为止,AlbedoBase XL v1.3 已经合并了 141 个选定的检查点和 251 个 LoRA。
日志
v2.1
使用新的合并算法和公式重新合并并调整 v0.1 至 2.0。
作品参照
Stable Diffusion提示词
Imagine a breathtakingly cinematic photograph that captures the essence of the 70s: A young hippie girl, her brown hair cascading down in carefree waves, is adorned with a vibrant flower crown. She radiates joy and the spirit of the era, wearing colorful, flowy clothes that speak of freedom and rebellion. Her smile is infectious, and around her neck hangs a peace sign necklace, a symbol of the times. She stands on a bustling San Francisco street, the essence of the city's soul in the 70s palpable in the background. This image, taken with a 50mm lens, boasts a film-like quality, rich in bokeh, and exudes a professional, 4k level of high detail. The photograph captures not just a moment but the spirit of an age - a snapshot that is as vivid and detailed as memory itself.
Steps: 60, Size: 720x1280, Seed: 9, Model: albedobaseXL_v21, Version: v1.7.0, Sampler: DPM++ 2M SDE Heun Karras, CFG scale: 7.5, Model hash: 1718b5bb2d
作者: | albedobond |
最后更新: | 2023年11月13日 |
文件类型: | CHECKPOINT MERGE |
版本: | v1.2 |
基本模型: | SDXL 1.0 |
tag标签: | 游戏 动漫 真人 2D 3D 动物 电影 设计 摄影 肖像 多合一 |
触发词: | all in one photo 2d 3d anime game photorealistic cinematic design base |
使用介绍
不需要精炼机,包含VAE。
将否定提示留空是获得良好质量的方法。
目标
Stable Diffusion XL 拥有 66 亿个参数,比 SD v1.5 版本多出约 6.6 倍。我相信这不仅仅是一个数字,而是一个可以带来性能显着提升的数字。
我们已经有一段时间意识到,由于我们社区的爆炸性贡献,SD v1.5 的整体性能已经超出了想象。因此,我正在努力完成这个 AlbedoBase XL 模型,以便在此 XL 版本中最佳地再现 v1.5 中出现的性能改进。
我的目标是直接测试所有公开上传到 Civitai 的 Checkpoint 和 LoRA 的性能,并仅合并经过多次过滤器后判断为最佳的资源。这将超越Midjourney等公司的图像生成AI的性能。
截至目前,AlbedoBase XL v1.2 已合并了 96 个选定的检查点和 244 个 LoRA。
日志
v1.2
- 合并了22个最新的检查点。
作品参照
Stable Diffusion提示词
cinematic still frame, world's most beautiful cyborg Ayanami rei in Evangelion, street fashion model, cyberpunk 2077 style new york street
Negative prompt: worst quality, unrealistic details
Steps: 70, Size: 720x1280, Seed: 2920413160, Model: albedobaseXL_v12, Version: v1.6.0-2-g4afaaf8a, Sampler: DPM++ 2M SDE Karras, CFG scale: 4.1, Model hash: 8a8efa5ad2, Style Selector Style: base, Style Selector Enabled: True, Style Selector Randomize: False