使用介绍
大家好,你们可能已经注意到,我的小世界里现在很忙碌。
我正在学习使用 XL(我不知道您是否见过AniVerse XL的推出)、AniVerse Pony XL 进行训练,现在我想学习如何为 Pixart Sigma 创建模型,只要我的视频卡允许。
出于这个原因,我每周一发布的新模型可能会被跳过。
因此,此版本的 AniVerse 只是一个合并,旨在不让所有使用 Stable Diffusion 1.5 的人“孤单”(包括我)。
我特别喜欢它的色彩、细节和动感。可惜的是它的手,根据我的测试,它并不是最好的。
让我知道你的想法。
PS 一有时间,我就会为那些拥有 3/4GB 显卡的人写一篇文章,介绍如何在这些显卡中使用 XL。
正面提示:
(best quality, masterpiece, colorful, highest detailed) YOUR PROMPT (ultra-detailed body), (light smile:0.3), moonlight passing through hair, (night beautiful background:1.3), (intricate details), (dynamic angle)
负面提示:
1) Easy Negative, moles, freckles,
2) Bad-Images-39000, moles, freckles
PS:下周将推出精简版
推荐设置:
- 过多的负面提示会让你的创作变得更糟,所以请遵循我下面的建议!
- 在应用 LoRA 制作您最喜欢的角色之前,请先尝试一下。您可能会对这个模型的功能感到惊讶!
如果您使用我的型号,请为您的 A1111安装ADetailer扩展。
导航到Stable Diffusion的“Extensions”选项卡。
- 转到“Install from URL”小节。
- 粘贴以下 URL:https://github.com/Bing-su/adetailer
- 单击“Install”按钮安装扩展
- 重启你的Stable Diffusion
- VAE: 包含VAE(但通常我仍然使用840000 ema pruned)
- Clip skip: 2
- Upscaler: 4x-Ultrasharp或4X NMKD Superscale
我最喜欢的提示:
(masterpiece, best quality, highres:1.2), (photorealistic:1.2), (intricate and beautiful:1.2), (detailed light:1.2), (colorful, dynamic angle), RAW photo, upper body shot, fashion photography, YOUR PROMPT, (highres textures), dynamic pose, bokeh, soft light passing through hair, (abstract background:1.3), (sharp), exposure blend, bokeh, (hdr:1.4), high contrast, (cinematic), (muted colors, dim colors, soothing tones:1.3), morbid
负面提示:
(worst quality, low quality), negative_hand-neg, bad-hands-5, naked, nude, braless, cross, sepia, black&white, B&W, painting, drawing, illustration
否定提示词:
您还可以使用这些负嵌入:
- 1) Easy Negative , negative_hand-neg , bad-hands-5
- 2) Bad-Images-39000、negative_hand-neg、bad-hands-5
- 3) ng_deepnegative_v1_75t , bad-hands-5
- 4) FastNegativeV2 , negative_hand-neg , bad-hands-5
- 5)对于男性图像:girl, woman, female, tits,BadImage_v2-39000、negative_hand-neg、bad-hands 5
该模型免费供个人 使用,免费供个人合并( * )。
如需商业用途,请务必联系我(Ko-fi)或通过电子邮件:samuele[dot]bonzio[at]gmail[dot]com
作品参照
(best quality, masterpiece, colorful, highest detailed) upper body photo, fashion photography of cute (Hatsune Miku), very long turquoise pigtails and a school uniform-like outfit. She has teal eyes and very long pigtails held with black and red square-shaped ribbons that have become a signature of her design, (ultra-detailed body), (light smile:0.3), moonlight passing through hair, (silver lake background:1.3), (intricate details), (dynamic angle)
Negative prompt: bad-image-v2-39000, moles, (grayscale:1.4), fat, ugly, freckles
Steps: 30, RNG: NV, Size: 576x1024, Seed: 316717321, Model: AniVerse v5.0-Full, Version: v1.9.4, Sampler: DPM++ 2M, CFG scale: 6, Clip skip: 2, TI hashes: [object Object], Model hash: e56b2bb810, Schedule type: Karras, ADetailer model: face_yolov8n.pt, ADetailer version: 24.6.0, ADetailer mask blur: 4, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer inpaint padding: 32, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True
(Tifa Lockhart), (best quality, masterpiece, colorful, highest detailed) upper body photo, fashion photography of cute (Tifa Lockhart), long brown hair, deep reddish-brown eyes, long straight tied hair, tied at the end, hair partially covering her right eye. She wears a white tank top, exposed stomach, black leather mini-skirt, black belt and black suspenders, red boots, black socks, white drop earrings. Metal reinforcement is on her left shoe, studded reinforcement on her left elbow, (ultra-detailed body), (light smile:0.3), moonlight passing through hair, (colorful background:1.3), (intricate details), (dynamic angle)
Negative prompt: CyberRealistic_Negative_Anime-neg, moles, (grayscale:1.4), fat, ugly, freckles
Steps: 30, RNG: NV, Size: 576x1024, Seed: 2281244268, Model: AniVerse v5.0-Full, Version: v1.9.4, Sampler: DPM++ 2M, CFG scale: 6, Clip skip: 2, TI hashes: [object Object], Model hash: e56b2bb810, Schedule type: Karras, ADetailer model: face_yolov8n.pt, ADetailer version: 24.6.0, ADetailer mask blur: 4, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer inpaint padding: 32, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True
(best quality, masterpiece, colorful, highest detailed) upper body photo, fashion photography of cute (1 cute cat-girl) cat ears, cat's tail, high detailed sailor suit, navel, pink hair ribbon:1.2, hairband, pink thighhighs, pink shoes, wrist cuffs, twintails, gem, frills, (ultra-detailed body), (light smile:0.3), softlight, (night beautiful background:1.3), (intricate details), (dynamic angle)
Negative prompt: CyberRealistic_Negative_Anime-neg, moles, (grayscale:1.4), fat, ugly, freckles
Steps: 30, RNG: NV, Size: 576x1024, Seed: 2908515483, Model: AniVerse v5.0-Full, Version: v1.9.4, Sampler: Euler_Max, CFG scale: 6, Clip skip: 2, TI hashes: [object Object], Model hash: e56b2bb810, Schedule type: Automatic, ADetailer model: face_yolov8n.pt, ADetailer version: 24.6.0, ADetailer mask blur: 4, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer inpaint padding: 32, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True
使用介绍
这个版本的 AniVerse 是上次灾难性训练中出名的版本。幸运的是,我设法保留了其中的好东西。
一般来说,它往往能更好地遵循提示,并且在相同的步骤下,会提供更多细节。
另一件积极的事情是,灰度或黑白的随机生成已经大幅减少!
我希望你喜欢 :)
2024/6/3:下载地址已修复
正面提示:
(best quality, masterpiece, colorful, highest detailed) YOUR PROMPT (ultra-detailed body), (light smile:0.3), moonlight passing through hair, (night beautiful background:1.3), (intricate details), (dynamic angle)
负面提示:
1) Easy Negative, moles, freckles,
2) Bad-Images-39000, moles, freckles
PS:下周将推出精简版
推荐设置:
- 过多的负面提示会让你的创作变得更糟,所以请遵循我下面的建议!
- 在应用 LoRA 制作您最喜欢的角色之前,请先尝试一下。您可能会对这个模型的功能感到惊讶!
如果您使用我的型号,请为您的 A1111安装ADetailer扩展。
导航到Stable Diffusion的“Extensions”选项卡。
- 转到“Install from URL”小节。
- 粘贴以下 URL:https://github.com/Bing-su/adetailer
- 单击“Install”按钮安装扩展
- 重启你的Stable Diffusion
- VAE: 包含VAE(但通常我仍然使用840000 ema pruned)
- Clip skip: 2
- Upscaler: 4x-Ultrasharp或4X NMKD Superscale
我最喜欢的提示:
(masterpiece, best quality, highres:1.2), (photorealistic:1.2), (intricate and beautiful:1.2), (detailed light:1.2), (colorful, dynamic angle), RAW photo, upper body shot, fashion photography, YOUR PROMPT, (highres textures), dynamic pose, bokeh, soft light passing through hair, (abstract background:1.3), (sharp), exposure blend, bokeh, (hdr:1.4), high contrast, (cinematic), (muted colors, dim colors, soothing tones:1.3), morbid
负面提示:
(worst quality, low quality), negative_hand-neg, bad-hands-5, naked, nude, braless, cross, sepia, black&white, B&W, painting, drawing, illustration
否定提示词:
您还可以使用这些负嵌入:
- 1) Easy Negative , negative_hand-neg , bad-hands-5
- 2) Bad-Images-39000、negative_hand-neg、bad-hands-5
- 3) ng_deepnegative_v1_75t , bad-hands-5
- 4) FastNegativeV2 , negative_hand-neg , bad-hands-5
- 5)对于男性图像:girl, woman, female, tits,BadImage_v2-39000、negative_hand-neg、bad-hands 5
该模型免费供个人 使用,免费供个人合并( * )。
如需商业用途,请务必联系我(Ko-fi)或通过电子邮件:samuele[dot]bonzio[at]gmail[dot]com
作品参照
Stable Diffusion提示词
(best quality, masterpiece, colorful, highest detailed) upper body photo, fashion photography of cute (Asuka Langley), in high detailed textured Evangelion red plugsuit, (ultra-detailed body), (light smile:0.3), moonlight passing through hair, (colorful background:1.3), (intricate details), (dynamic angle)
Negative prompt: EasyNegative, moles, freckles
Steps: 30, RNG: NV, VAE: vae-ft-mse-840000-ema-pruned.safetensors, Size: 576x1024, Seed: 2977461426, Model: AniVerse v4.0-Full, Version: v1.8.0, Sampler: DPM++ 2M Karras, CFG scale: 7, Clip skip: 2, Mask blur: 4, TI hashes: [object Object], Model hash: b5bc30f560, Inpaint area: Only masked, ADetailer model: face_yolov8n.pt, ADetailer version: 24.1.2, Denoising strength: 0.4, ADetailer mask blur: 4, Masked area padding: 32, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer inpaint padding: 32, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True
(best quality, masterpiece, colorful, dynamic angle, realistic)\Lum\, upper body photo, fashion photography of cute, intense long green hair, orange eyes, tiger print, strapless bikini, yellow boots, yellow little horns, small horns, \Lum\, <lora:lum4:0.6> (ultrahigh resolution textures), in dynamic pose, bokeh, (intricate details, hyperdetailed:1.15), moonlight passing through hair, perfect night, (fantasy background), (official art, extreme detailed, highest detailed), HDR+
Negative prompt: EasyNegative, moles, freckles
Steps: 30, RNG: NV, VAE: vae-ft-mse-840000-ema-pruned.safetensors, Size: 576x1024, Seed: 2977461429, Model: AniVerse v4.0-Full, Version: v1.8.0, Sampler: DPM++ 2M Karras, CFG scale: 7, Clip skip: 2, Mask blur: 4, TI hashes: [object Object], Model hash: b5bc30f560, Inpaint area: Only masked, ADetailer model: face_yolov8n.pt, ADetailer version: 24.1.2, Denoising strength: 0.4, ADetailer mask blur: 4, Masked area padding: 32, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer inpaint padding: 32, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True
masterpiece, best quality, 1 girl, (ultra-detailed body), Azure hair, long hair, detailed eyes, forest, bare shoulders, hanfu, lakes, pure, soft (light smile:0.3), bamboo, official art, 8k wallpaper, highly detailed,
Negative prompt: EasyNegative, moles, freckles
Steps: 30, RNG: NV, VAE: vae-ft-mse-840000-ema-pruned.safetensors, Size: 576x1024, Seed: 2977461426, Model: AniVerse v4.0-Full, Version: v1.8.0, Sampler: DPM++ 2M Karras, CFG scale: 7, Clip skip: 2, Mask blur: 4, TI hashes: [object Object], Model hash: b5bc30f560, Inpaint area: Only masked, ADetailer model: face_yolov8n.pt, ADetailer version: 24.1.2, Denoising strength: 0.4, ADetailer mask blur: 4, Masked area padding: 32, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer inpaint padding: 32, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True
作者: | Samael1976 |
最后更新: | 2024年4月4日 |
基本模型: | SD 1.5 |
模型版本: | v3.0 |
使用提示: | CLIP SKIP: 2 |
使用介绍
这是一个长期项目,我想在每次更新时实现一些新的东西!
首先,这个名字是两个词的合并:Animation 和 Universe(还有一个双关语:Any+Universe -> Anyverse -> Aniverse)
推荐设置:
- 过多的负面提示会让你的创作变得更糟,所以请遵循我下面的建议!
- 在应用 LoRA 制作您最喜欢的角色之前,请先尝试一下。您可能会对这个模型的功能感到惊讶!
如果您使用我的型号,请为您的 A1111安装ADetailer扩展。
导航到Stable Diffusion的“Extensions”选项卡。
- 转到“Install from URL”小节。
- 粘贴以下 URL:https://github.com/Bing-su/adetailer
- 单击“Install”按钮安装扩展
- 重启你的Stable Diffusion
- VAE: 包含VAE(但通常我仍然使用840000 ema pruned)
- Clip skip: 2
- Upscaler: 4x-Ultrasharp或4X NMKD Superscale
我最喜欢的提示:
(masterpiece, best quality, highres:1.2), (photorealistic:1.2), (intricate and beautiful:1.2), (detailed light:1.2), (colorful, dynamic angle), RAW photo, upper body shot, fashion photography, YOUR PROMPT, (highres textures), dynamic pose, bokeh, soft light passing through hair, (abstract background:1.3), (sharp), exposure blend, bokeh, (hdr:1.4), high contrast, (cinematic), (muted colors, dim colors, soothing tones:1.3), morbid
负面提示:
(worst quality, low quality), negative_hand-neg, bad-hands-5, naked, nude, braless, cross, sepia, black&white, B&W, painting, drawing, illustration
否定提示词:
您还可以使用这些负嵌入:
- 1) Easy Negative , negative_hand-neg , bad-hands-5
- 2) Bad-Images-39000、negative_hand-neg、bad-hands-5
- 3) ng_deepnegative_v1_75t , bad-hands-5
- 4) FastNegativeV2 , negative_hand-neg , bad-hands-5
- 5)对于男性图像:girl, woman, female, tits,BadImage_v2-39000、negative_hand-neg、bad-hands 5
该模型免费供个人 使用,免费供个人合并( * )。
如需商业用途,请务必联系我(Ko-fi)或通过电子邮件:samuele[dot]bonzio[at]gmail[dot]com
作品参照
Stable Diffusion提示词
(best quality, masterpiece, colorful, highest detailed) upper body photo, fashion photography of cute (1girl), green eyes, blonde, hair ornament, short hair, blush, lipstick, slim, slim body, perfect detailed face, naughty face, detailed eyes, ((fantasy forest)), ((small flowers around)), detailed legs, barefoot, ((dark night)), ((darkness)), elf ears, ideal bobtail, magic flowers, ((small moon)), dim colors
Negative prompt: EasyNegative, moles, freckles
Steps: 30, RNG: NV, VAE: vae-ft-mse-840000-ema-pruned.ckpt, Size: 576x1024, Seed: 2534949469, Model: AniVerse v3.0-Full, Version: v1.8.0, Sampler: DPM++ 2M Karras, CFG scale: 7, Clip skip: 2, Mask blur: 4, TI hashes: [object Object], Model hash: ece886993c, Inpaint area: Only masked, ADetailer model: face_yolov8n.pt, ADetailer version: 24.3.1, Denoising strength: 0.4, ADetailer mask blur: 4, Masked area padding: 32, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer inpaint padding: 32, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True
作者: | Samael1976 |
最后更新: | 2023年12月20日 |
文件类型: | CHECKPOINT TRAINED |
基本模型: | SD 1.5 |
tag标签: | 动漫 肖像 女性 2.5D 游戏 角色 科幻 新世纪福音战士 刀剑神域 |
模型版本: | v2.0(HD)/v2.0(HD)- Pruned |
使用提示: | CLIP SKIP: 2 |
使用介绍
这个 AniVerse 版本实际上是我制作过的最难的模型,但我也认为它也是所有发布的版本中最好的版本。
我有很多东西要写,但我首先会在简短的括号之后向您提供创作的详细信息。
为什么用HD和2.0这个词?我想你一定想知道。
HD:因为这个版本的 AniVerse 是(几乎)完全在 1024X1024px 下训练的。
2.0:因为与之前的版本相比,真实感略有提高
/////////////////////////////////////////////////////////// ///////////////////////////////////////////////////////////
该版本为您提供了更多的使用范围。
您可以照常使用它,以半写实、卡通或更写实的风格创建 2.5D 图像。
这完全取决于您将使用的积极提示和消极提示。
您可以找到我使用的提示示例。
总之我可以告诉你:
1) Bad-Images-39000 = 更多卡通图像
2) Easy Negative和FastNegativeV2 = 通常的 Aniverse 2.5D 风格图像
3) ng_deepnegative_v1_75t和EasyNegativeV2 = 类似经典 Aniverse 的图像,带有一点更大的真实感(不多)
/////////////////////////////////////////////////////////// ///////////////////////////////////////////////////////////
但是 1024px 的训练意味着什么呢?
您可以安心地将您的创作推至 1024×576(或 1024×768)分辨率,而无需使用 HiRes.fix。我甚至更进一步,高达 1280×720,误差幅度约为 5%(如果您遵循我的建议):
- VAE: VAE is included (但通常我仍然使用 840000 ema pruned)
- Clip skip: 2
- Upscaler: 4x-Ultrasharp or 4X NMKD Superscale
- Sampling method: DPM++ 2M SDE Karras
- Width: 576 (o 768)
- Height: 1024
- CFG Scale: 3 -> Steps: 15
CFG Scale: 4 -> Steps: 20
CFG Scale: 5 -> Steps: 25
CFG Scale: 6 -> Steps: 30
- 等等…
我最喜欢的提示:
- (masterpiece, best quality, highres:1.2), (photorealistic:1.2), (intricate and beautiful:1.2), (detailed light:1.2), (colorful, dynamic angle), RAW photo, upper body shot, fashion photography, YOUR PROMPT, (highres textures), dynamic pose, bokeh, soft light passing through hair, (abstract background:1.3), (sharp), exposure blend, bokeh, (hdr:1.4), high contrast, (cinematic), (muted colors, dim colors, soothing tones:1.3), morbid
- 负面提示:
- (worst quality, low quality), negative_hand-neg, bad-hands-5, naked, nude, braless, cross, sepia, black&white, B&W, painting, drawing, illustration
您还可以使用这些负嵌入:
- 1) Easy Negative , negative_hand-neg , bad-hands-5
- 2) Bad-Images-39000、negative_hand-neg、bad-hands-5
- 3) ng_deepnegative_v1_75t , bad-hands-5
- 4) FastNegativeV2 , negative_hand-neg , bad-hands-5
- 5)对于男性图像:女孩、女人、女性、tits,BadImage_v2-39000、negative_hand-neg、bad-hands 5
可以肯定的是,我创建了 1000 多张图像,当我谈到大约 5% 的错误率时,我自然指的是由于分辨率太高而导致的经典拉长或双体。
有时可能会出现三只手臂或三只腿,但这是另一种类型的误差,与分辨率无关,所以我将其排除在误差计算之外。
如果你想避免这种错误,只需将分辨率降低到1024X576px,以返回到某种安全区域。
手!!!我的手终于变好了!!!
我说改进,但它们还不够完美,我对创作中所实现的细节感到非常满意。
我想在下一个版本中改进其他一些事情(或者更确切地说,我已经在努力消除有时发生在脸上的任何割伤/伪影)
我还想改善阴影和其他小东西(如牙齿)的角度。
但我可以说我对这个版本绝对满意。
现在闲聊一下,告诉你这个版本有多汗。
我正在训练的电脑时不时地(我仍在试图确定原因)决定崩溃,所以我必须将其关闭并重新开始训练(有时我不在家并失去全部数据)天)。
培训持续约490小时,相当于连续培训约21天。
为了找到手部的修复方案,在完成最终模型后,我又花了两周时间进行研究。
我再说一遍,这很艰难,很漫长,也很艰难,不止一次我想过扔掉一切,但最后,也感谢一点运气,我成功地创建了这个版本的 AniVerse,这真的让我很满意。
我希望它能给你带来和我一样的满足感。
我们很快就会再次讨论……一个新项目(该项目已经进行了一段时间,但在撰写本文时,我完全不知道它是否会带来预期的结果)。
作品参照
Stable Diffusion提示词
(best quality, masterpiece, colorful, dynamic angle, highest detailed)(Asuka Langley), upper body shot, fashion photography of cute, intense long hair, (Asuka Langley), dressing high detailed Evangelion red suit in dynamic pose, bokeh, light passing through hair, (abstract background:1.3), (official art)
Negative prompt: EasyNegative, moles, freckles, skindentation, cutoffs, shiny skin, bad-hands-5, scars on face,
Steps: 50, RNG: NV, VAE: vae-ft-mse-840000-ema-pruned.ckpt, Size: 720x1280, Seed: 1754034078, Model: Aniverse V2.0-Full, Version: v1.6.0, Sampler: DPM++ 2M SDE Karras, VAE hash: c6a580b13a, CFG scale: 6.5, Clip skip: 2, Model hash: 5473581026, bad-hands-5: aa7651be154c", "EasyNegative: 66a7279a88dd, ADetailer model: face_yolov8n.pt, ADetailer version: 23.11.0, ADetailer mask blur: 4, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer inpaint padding: 32, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True