【AI绘画】幼绘&绘#1

恋染秋山，绘若星繁

        前段时间学习了很多新内容，尝试着画了想画出来的绘，但其实大部分内容还是纯抽卡，技术含量并不是很高。AI修图能够让画的质量提高，但我暂时没有这个时间来精修。如果你有兴趣可以尝试一下。不过这些图片在某老鸽那里意外的很受欢迎，我也就趁机发个专栏好了。有关ai绘画相关基础知识还请自行寻找其他up主发布的视频。

——Base on Stable Diffusion WebUI——

备注：这组图片使用了同一组tag，区别在于大模型和seed，Sakuramochi的模型能够较好的得到一些幼和柔和的效果，我非常喜欢。详细参数如下。

Prompt：1girl, qiuhui, original, (illustration:1.1), (best quality), (masterpiece:1.1), (extremely detailed CG unity 8k wallpaper:1.1), (full body), (solo:1.2), loli, beautyfull detailed face and eyes, beautyfull intricacy clothing decorative pattern details, tang style outfits, long upper shan, chest pleated skirt, (Hanfu:1.1), blush, light smile, chinese, china, orange eyes, orange hair, <lora:qiuhui:0.6:OUTALL>, <lora:oniharigusha:1>

Negative Prompt：EasyNegativeV2

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 4.5, Size: 512x768, Denoising strength: 0.43, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires steps: 7, Hires upscaler: R-ESRGAN 4x+ Anime6B

备注：这组图片使用的大模型均是meinahentai_v21，虽然它本身是一个nsfw模型，但意外的获得了很好的效果。（v21版本相对来说手还是比较容易出问题，之后的v3版好了许多。）此外，这组图片本身色调偏粉，已经不像是绘了，这是因为我在tag中加入了pink hair，有时，这使得图片看起来更可爱一些。但其实我一开始想要得到的是小护士绘的效果，另外，双人图也是roll出来的，并没有使用controlnet。详细参数如下。

Prompt：1girl, solo, qiuhui, masterpiece, best quality, (loli:1.2), nurse cap, hat, pink hair, halo, nurse, blush, full body, looking at viewer, standing, short sleeves, breasts, medium breasts, open clothes, arms behind back, closed mouth, pink eyes, bangs, one side up, holding, clipboard, white background, collarbone, hair between eyes, simple background, sweat, tiles, holding clipboard, ribbon, white footwear, boots, medium hair, hair ribbon, trembling, <lora:qiuhui:0.6:OUTALL>, <lora:oniharigusha:1>

Negative Prompt：EasyNegativeV2, (low quality, worst quality:1.4), (bad anatomy:1.2), (inaccurate limb:1.2), bad composition, inaccurate eyes, (extra fingers:1.2), (fewer fingers:1.2), inaccurate fingers, extra digits, fewer digits, (bad feet:1.2), (bad hand:1.4)

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 4.5, Size: 512x768, Denoising strength: 0.43, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires steps: 9, Hires upscaler: R-ESRGAN 4x+ Anime6B

备注：在这组图片中，我试着修改了上一组图片的tag，使之呈现出手持注射器的效果，但实际上ai并没有理解我的意思，也有可能是大模型的原因，很遗憾。详细参数如下。

Prompt：1girl, solo, qiuhui, loli, full body, nurse cap, nurse, (holding syringe, injector syringe on right hand, holding injector, pink injector, pink syringe, injector, syringe:1.3), halo, pink hair, twintails, blush, light smile, looking at viewer, standing, white pantyhose, lolita shoes, white background, simple background, short sleeves, medium breasts, open clothes, closed mouth, pink eyes, bangs, one side up, collarbone, hair between eyes, sweat, tiles, ribbon, trembling, masterpiece, best quality, <lora:qiuhui:0.6:OUTALL>, <lora:oniharigusha:1>

备注：我修改了发色，让人物更接近原有的效果。此外，我其实尝试过手持注射器的lora模型，但并没有得到好的效果。详细参数如下。

Prompt：1girl, solo, qiuhui, loli, full body, nurse cap, nurse, ([holding syringe:1.3|holding injector:1.3], injector syringe on right hand:1.3, [injector|syringe]), halo, [orange hair|pink hair], twintails, blush, light smile, looking at viewer, standing, white pantyhose, lolita shoes, white background, simple background, short sleeves, closed mouth, one side up, collarbone, hair between eyes, sweat, ribbon, <lora:qiuhui:0.6:OUTALL>, <lora:oniharigusha:1>

Negative Prompt：EasyNegativeV2

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Size: 512x768, Denoising strength: 0.43, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires steps: 10, Hires upscaler: R-ESRGAN 4x+ Anime6B

备注：试图画出的天使绘，实际出图效果并不好。详细参数如下。

Prompt：1girl, solo, qiuhui, masterpiece, best quality, (loli:1.2), wrist scrunchie, pink hair, halo, scrunchie, hair ornament, skirt, wings, hair bun, shirt, white shirt, black socks, squatting, school uniform, bag, loafers, shoes, flower, hair flower, looking at viewer, smile, brown footwear, socks, simple background, short sleeves, heart, bow, bangs, low wings, blush, crescent, school bag, full body, single side bun, double bun, open mouth, angel wings, sidelocks, grey background, blue bow, white wings, <lora:qiuhui:0.6:OUTALL>, <lora:oniharigusha:1>

Negative Prompt：EasyNegativeV2, (low quality, worst quality:1.4), (bad anatomy:1.2), (inaccurate limb:1.2), bad composition, inaccurate eyes, (extra fingers:1.2), (fewer fingers:1.2), inaccurate fingers, extra digits, fewer digits, (bad feet:1.2), (bad hand:1.4)

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3395496700, Size: 512x768, Denoising strength: 0.5, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent

备注：这是两张意外的产物，这组tag的出图效果并不好，经常崩。详细参数如下。

Prompt：1girl, qiuhui, masterpiece, silhouette, stats, wide shot, simple background, loli, reclining, hair over one eye, eye contact, striped neckwear, skirt basket, no shoes, white legwear, cane, naughty, toeless boots, orange hair, <lora:qiuhui:0.6:OUTALL>, <lora:oniharigusha:1>

Negative Prompt：EasyNegativeV2

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 4.5, Size: 512x768, Denoising strength: 0.43, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires steps: 9, Hires upscaler: R-ESRGAN 4x+ Anime6B

备注：试验tag的产物。详细参数如下。

Prompt：1girl, qiuhui, (loli), (petite), (big breasts), looking at viewer, full body, star-shaped pupils, sailor dress, ponytail, fur collar, =_=, hand in bra, bodystocking, horse tail, cuffs, (wisteria), <lora:qiuhui:0.6:OUTALL>, <lora:oniharigusha:1>

Negative Prompt：EasyNegativeV2

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 4.5, Size: 512x768, Denoising strength: 0.43, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires steps: 10, Hires upscaler: R-ESRGAN 4x+ Anime6B

备注：我非常喜欢的设计图tag，在这组图片中我尝试了各种大模型。Anythingv5和cfv30的表现中规中矩，dorayakimix本身是个幼态模型，但看起来和Anythingv5有类似的出图效果，daintymix的皮肤质感其实很好，但似乎看不太出来。这组图片也是我试验tag的产物，看起来有些和绘并不相像。详细参数如下。

Prompt：((masterpiece:1.4, best quality))+, (ultra detailed)+, full body, (infographic:1.2), pajamas, patent drawings, physical measurement, (all clothes configuration:1.15), stationery, (solo), standing, cohesive background, (character sheet:1.2), detailed background,(by exquisite colors block),(by exquisite colors block), 1girl, long hair, headdress, orange hair, light smile, hair ornament, white thighhighs, dress, <lora:qiuhui:0.6:OUTALL>, <lora:add_detail:0.5>

Negative Prompt： EasyNegativeV2

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 6, Size: 512x768, Denoising strength: 0.36, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires steps: 8, Hires upscaler: R-ESRGAN 4x+ Anime6B

备注：有时，add_details的lora能够得到一些较好的效果，对于修手也是。详细参数如下。

Prompt：((masterpiece:1.4, best quality))+, (ultra detailed)+, full body, (infographic:1.2), pajamas, patent drawings, physical measurement, (all clothes configuration:1.15), stationery, (solo), standing, cohesive background, (character sheet:1.2), detailed background,(by exquisite colors block),(by exquisite colors block), 1girl, loli, long hair, headdress, orange hair, light smile, hair ornament, white thighhighs, dress, <lora:qiuhui:0.6:OUTALL>, <lora:oniharigusha:1>

Negative Prompt： EasyNegativeV2

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Size: 768x512, Denoising strength: 0.36, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires steps: 8, Hires upscaler: R-ESRGAN 4x+ Anime6B

备注：以前试验tag的产物，除了光影效果较好以外，手和其他部分都不是很好，残念。因为是以前的试验品，tags显得不是很好，而且也没有使用分层以减少过拟合，请谅解。

Prompt：(masterpiece, best quality:1.4), 1girl, solo, black skirt, orange eyes, electric guitar, guitar, headphones, holding, holding plectrum, instrument, long hair, music, one side up, orange hair, playing guiter, skinny, very young girl, (plunging neckline), wedding dress, deep V sling, bare chest, (bare shoulders, collarbone, strapless:1.2), (cleavage), cleavage cutout, (areola slip), (pale skin:1.05), light smile, <lora:qiuhui:0.5>

Negative Prompt： EasyNegative, bad anatomy, bad feet, bad hands, bad fingers, bad mouth, (big breasts:1.2), (worst quality, low quality:1.3), (depth of field, blurry:1.2), (greyscale, monochrome:1.1), 3D face, nose, cropped, lowres, text, jpeg artifacts, signature, watermark, username, blurry, artist name, trademark, watermark, title, multiple view, reference sheet

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 6, Size: 768x512, Denoising strength: 0.3, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires steps: 7, Hires upscaler: R-ESRGAN 4x+ Anime6B

备注：尝试的弹吉他动作。WD4模型也是我非常喜欢的一个模型，虽然有时候显得画面比较黑，此外也是这时候我开始使用meina的模型。

Prompt：1girl, solo, qiuhui, idol singing costume, masterpiece, best quality,

electric guitar, headphones, holding plectrum, instrument, music, playing guiter,

long hair, very young girl, light smile, one side up, skinny, pale skin, bare thighs, bare chest, bare shoulders, collarbone, strapless, plunging neckline, wedding dress, deep V sling, cleavage, cleavage cutout, areola slip, <lora:qiuhui:0.6:OUTALL>, <lora:oniharigusha:0.6>, <lora:idolSingingCostume_v11:0.6>

Negative Prompt：EasyNegativeV2, (low quality, worst quality:1.4), (bad anatomy:1.2), (inaccurate limb:1.2), bad composition, inaccurate eyes, (extra fingers:1.2), (fewer fingers:1.2), inaccurate fingers, extra digits, fewer digits, (bad feet:1.2), (bad hand:1.4)

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 4.5, Size: 512x768, Denoising strength: 0.43, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires steps: 10, Hires upscaler: R-ESRGAN 4x+ Anime6B

备注：早期尝试的手办风。这些tag显得不是很美，遗憾。

Prompt：extremely delicate and beautiful, depth of field, amazing, masterpiece,growth,visual impact, ultra-detailed, highly, Amazing finely detail, angel_wings, feathered_wings, wings, white_wings, angel, 1girl, long_hair, multiple_wings, silver_hair, halo, solo, thighhighs, spread_wings, black_legwear, low_wings, smile, full body, detached_wings, very_long_hair, single_wing, (perfect body), (fine and beautiful detailed skin:1.2), (shiny skin:1.2), (seductive), full body, high-key and low-variance brightness scale, soft light, light and dark interlaced, <lora:qiuhui:0.7>, <lora:mignon374StyleLora_v10:0.5> 

Negative Prompt：(umbrella:1.9),(worst quality, low quality:2),EasyNegative, lanjiao3 lanjiao4 lanjiao5,(mutated hands and fingers:1.5), (long body :1.3), (mutation, poorly drawn:1.2), (bad and mutated hands:1.3), (bad hands), (missing fingers), multiple limbs, bad anatomy, (interlocked fingers:1.2), Ugly Fingers, (extra digit and hands and fingers and legs and arms:1.4),(deformed fingers:1.2), (long fingers:1.2),normal quality, letterbox,deformed,out of frame,60s,70s,80s,90s,00s,ugly,morbid, mutation, arthropod girl,arthropod limbs, disfigured,extra_arms,extra_ears,extra_breasts,(extra_legs:1.2)

Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Size: 512x768, Denoising strength: 0.43, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent

备注：我想特别展示一下这张画，是我挺早期尝试的幼绘，当时也给了我关于ai绘画的很大的震撼，这也是我坚持学习不断更新的新技术新领域的一个动力。（虽然配置仍然限制着我，残念。）当然，本身这张图并没有很高的水平，请理解。此外，这张画的手动修改图也让亲爱的云殇看到了，期待它的另一个新生。

Prompt：1girl, (from below), long hair, maple eyes, ((loli, little girl)), (white pantyhose), white legwear, white thighhighs, cute face, light smile, full body, <lora:incomingHugKiss_v12:0.6>, <lora:qiuhui:0.7>, <lora:oniharigusha_v1:0.7>

Negative Prompt：(extra navel:1.4), extra fingers, fewer fingers, extra legs, fewer legs, (low quality, worst quality:1.4), (bad anatomy), (inaccurate limb:1.2), bad composition, inaccurate eyes, extra digit, fewer digits, (extra arms:1.2), EasyNegative

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 10, Size: 488x728, Denoising strength: 0.2, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires steps: 10, Hires upscaler: R-ESRGAN 4x+ Anime6B

备注：封面图，这个比例真的很不常见，所以我特地画了一张。

Prompt：1girl, qiuhui, loli, original,(illustration:1.1),(best quality),(masterpiece:1.1),(extremely detailed CG unity 8k wallpaper:1.1), (solo:1.2), beautyfull detailed face and eyes, beautyfull intricacy clothing decorative pattern details, tang style outfits,long upper shan, chest pleated skirt, (Hanfu:1.1), blush, light smile, chinese, china, orange eyes, orange hair, hair ornament, maple leaf, lying, on side, upper body, <lora:qiuhui:0.6:OUTALL>, <lora:oniharigusha:1>

Negative Prompt：EasyNegativeV2

Steps: 20, Sampler: DPM++ 2M SDE Karras, CFG scale: 4.5, Size: 896x256, Denoising strength: 0.3, Clip skip: 2, Hires upscale: 2, Hires steps: 10, Hires upscaler: R-ESRGAN 4x+ Anime6B

        感谢您的认真观看，如果能够得到您的喜欢就再好不过了。

        需要注意的一点是，实际上出图的tag并非文章中展示的那样，因为许多tag组是我复制或反推字别人的图片再加以修改而成。为了放到文中，一些不合适的tag被我删除，以及tag的排版也被我忽略了。实际上我还得到了一些其他的绘画，但毕竟不方便在专栏展示，敬请谅解。如果您需要相关绘画的原图图包，请告诉我，我会在后续放到网盘链接中。

        此外由于各种各样的原因，文章中使用的某两个lora模型并不会公开分享，用ai绘画圈的话来说就是，是私模.jpg。当然，如果身为绘星的您也想画出特别的绘，请私信我。虽然SD本身是开源的，但ai绘画圈仍旧处于较为混乱的状态，如果不是有人需要，我应该不会发布这篇专栏。如果您也想得到相似的绘画效果，请自行学习lora模型的训练或controlnet的reference模型相关知识，期待看到您的作品。

        如果有机会的话，我也会继续分享ai绘画的相关成果，虽然编辑专栏真的很累。这里是从17年开始就单推绘总的绘星哀与默，再次感谢您的喜欢。

2023年6月11日