前言

不知道大家上学的时候有没有试过出黑板报呢？反正我是没试过。但背着老师,给全班整点活儿的想法恐怕大伙儿都有。当年上学的时候没现在胆子大错过了，好在现在有AI绘画，可以弥补这个童年遗憾。

这次要测试的lora是由kozue大佬带来的黑板绘lora，主要作用是在教室黑板的环境种生成带有粉笔笔触的画面。配合一些简单的操作，应该可以整出很多有趣的活。这个lora目前的上传时间是5月12日，按照作者所说，发布后一个月内禁止使用它参加AI绘画比赛，希望大家在玩的时候也能尊重作者的意愿，不要用来参与相关赛事。

文档补充

↓因为b站文档排版太麻烦，阅读体验差，而且只能更新三次。主要内容与后续更新都放在下面这个链接↓

https://zhuanlan.zhihu.com/p/629606204

下载链接

图包链接：

https://pan.baidu.com/s/1lxH9WEl_y47Pvb0s_nFQmg?pwd=khb0

提取码：khb0

大模型链接：

https://pan.baidu.com/s/15KGEGQGBRHh8Q-Yy0AGTQw?pwd=ozkr

提取码：ozkr

lora链接：

https://pan.baidu.com/s/173IOCHJmIuSZn9NYw0UqBw?pwd=1ccd

提取码：1ccd

交流群

qq群：697080799

微信群:

测试中使用的模型

CKPT大模型

abyssorangemix

anthing-v45

realisticVisionV20

revAnimated_V11

chilloutmix

bl5（官方使用）

实际使用案例

大模型：revAnimated_v11

lora：blackboard 权重：0.8

lora：kamisatoAyakaGenshin 权重：1

正：

masterpiece, best quality,

chalkboard, traditional media, (chalk outline:0.7), (monochrome:0.4),

facing to the side,

(1girl,upper body, solo),

arm guards, armored dress, breastplate, japanese armor, hand fan, fingerless gloves,kamisato ayaka,

lora:kamisatoAyakaGenshin_kamisatoAyaka:0.8

负：

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, lowres, text, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, text, signature, watermark, simple background, toony, dated, low res, line art, flat colors

Steps: 25, Sampler: Euler a, CFG scale: 7, Seed: 1483223542, Size: 768x512

大模型：revAnimated_v11

lora：blackboard 权重：0.8

正：

masterpiece, best quality,

chalkboard, traditional media, (chalk outline:0.7), (monochrome:0.6),

close-up, facing to the side,

1girl,floating hair,bespectacled, short hair with long locks,

flower,falling petals,

负：

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, lowres, text, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, text, signature, watermark, simple background, toony, dated, low res, line art, flat colors

Steps: 30, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 2712620177, Size: 768x512

大模型：revAnimated_v11

lora：blackboard 权重：0.8

lora：kleeGenshinImpact 权重：0.8

正：

masterpiece, best quality,

chalkboard, traditional media, (chalk outline:0.9), (monochrome:0.4),

(1girl,upper body, solo),smile,

backpack pendant, hat ,stuffed animal, hat ,klee \(genshin impact\) \(cosplay\), (little girl:1.2)

lora:kleeGenshinImpact_kleeGenshinImpact:0.8

负：

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, lowres, text, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, text, signature, watermark, simple background, toony, dated, low res, line art, flat colors

Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 1408063099, Size: 768x512

大模型：revAnimated_v11

lora：blackboard 权重：0.8

lora：dva 权重：0.8

正：

masterpiece, best quality,

chalkboard, traditional media, (chalk outline:0.9), (monochrome:0.4),

close-up, facing to the side,

1girl, d.va (overwatch), solo, long hair, lora:dva:0.8

负：

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, lowres, text, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, text, signature, watermark, simple background, toony, dated, low res, line art, flat colors

Steps: 25, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 1073074156, Size: 768x512

大模型：revAnimated_v11

lora：blackboard 权重：0.8

正：

masterpiece, best quality,

chalkboard, traditional media, (chalk outline:1), (monochrome:0.7),

1dog,solo,

(golden retriever), (close-up:1.3),floating leaf,

负：

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, lowres, text, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, text, signature, watermark, simple background, toony, dated, low res, line art, flat colors

Steps: 25, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 1788228922, Size: 768x512

大模型：revAnimated_v11

lora：blackboard 权重：0.8

正：

masterpiece, best quality,

chalkboard, traditional media, (chalk outline:1), (monochrome:0.7),

1cat,solo,

负：

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, lowres, text, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, text, signature, watermark, simple background, toony, dated, low res, line art, flat colors

Steps: 25, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 3137366090, Size: 768x512

提示词对照表

基础信息

大模型：revAnimated_v11

lora：blackboard 权重：0.7

正：

masterpiece, best quality,

soft light,

upper body, looking at viewer,

(1girl,solo),short hair,jacket,beret, neckerchief,cute,(smile:0.6),floating scarf, floating hair,

(park, garden),maple tree, flying leaf,

负：

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, lowres, text, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, text, signature, watermark, simple background, toony, dated, low res, line art, flat colors

Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 340033791, Size: 768x512

对照表

阶段结论

官方给出的提示词：chalkboard，monochrome，traditional media，chalk outline,

效果明显，如图所示。自行取用。

单词权重对照

基础信息

大模型：revAnimated_v11

lora：blackboard

正：

masterpiece, best quality,

soft light,

chalkboard, traditional media, (chalk outline:NUM), (monochrome:0.4),

upper body, looking at viewer,

(1girl,solo),short hair,jacket,beret, neckerchief,cute,(smile:0.6),floating scarf, floating book,

(park, garden),maple tree, flying leaf,

负：

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, lowres, text, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, text, signature, watermark, simple background, toony, dated, low res, line art, flat colors

Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 2767837104, Size: 768x512

Chalk outline 权重对照

Monochrome 权重对照

traditional media 权重对照

english text与chinese text的作用

阶段结论

Monochrome是单色，在使用它之后画面颜色会趋向单一，

想要生成单色，权重可以高于0.6，只是想让画面颜色不要太过鲜艳则可以选择低于0.6的权重。

要注意的是如果需要生成单色画面，就不要写带有颜色倾向的提示词，比如黑发，灰色外套等，会影响画面颜色。不过如果想在单色种凸显某个颜色，比如想要单底色情况下的红色眼睛，则可以使用红眼这样的提示词。

Chalk outline的意思是粉笔轮廓，可以一定程度的强化粉笔的风格质感，在权重超过0.9的时候开始较为明显。建议日常如果没有特别需求，控制在0.9~1.2之间即可。

traditional media的意思是传统媒体，具体指代的风格我也不太确定，可能是纸媒上手绘稿的意思。测试中不同权重情况下除了有些色彩的变化，没感觉有风格的区别，我测试了很多次，也没太搞明白是什么原理。不过看对比测试的时候，感觉使用它之后一定程度优化了颜色和构图方式，既然是作者推荐的，按照默认权重用上去准没错。

另外，还能用English text与chinese text去控制黑板上的文字生成，不过这些字只是有外貌倾向，而不是真的能生成有意义的句子。如果不想要字，可以将他们加入反向提示词或者使用simple background来生成相对干净的黑板画面。

大模型测试

基础信息

大模型：revAnimated_v11

lora：blackboard

正：

masterpiece, best quality,

chalkboard, traditional media, (chalk outline:1.2), (monochrome:0.2),

soft light,

upper body, looking at viewer,

1girl,short hair,jacket,beret, neckerchief,cute,(smile:0.6),

(park, garden, flower),holding balloon, colored balloon,

负：

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, lowres, text, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, text, signature, watermark, simple background, toony, dated, low res, line art, flat colors

Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 428481863, Size: 768x512

对照表

阶段结论

大模型测试中除了几个类真实照片的大模型表现不佳，其他表现都还不错，除了作者推荐的bl5之外，revAnimated的表现效果也意外的好，所以这次测试使用了revAnimated。

兼容性测试

人物lora测试（女性）

基础信息

大模型：revAnimated_v11

lora：blackboard

lora：misaka_mikoto_v18

正：

masterpiece, best quality, soft light, chalkboard, traditional media, (chalk outline:0.7), (monochrome:0.3), upper body, looking at viewer, 1girl,solo, misaka_mikoto,brown eyes,white_shirt,small_breast,school_uniform, classroom, 负:

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, lowres, text, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, text, signature, watermark, simple background, toony, dated, low res, line art, flat colors Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 4134330637, Size: 768x512

对照图

人物lora测试（男性）

基础信息

大模型：revAnimated_v11

lora：blackboard

lora：jojo_joseph_joestar-10

正：

masterpiece, best quality, soft light, chalkboard, traditional media, (chalk outline:0.7), (monochrome:0.4), upper body, (1man,solo) joseph joestar, (muscular:1.1) male, brown hair, tank top, jeans, fingerless gloves, multicolored scarf, striped scarf, midriff 负：

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, lowres, text, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, text, signature, watermark, simple background, toony, dated, low res, line art, flat colors

Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 1107534963, Size: 768x512

对照图

服装lora测试

大模型：revAnimated_v11

lora：blackboard

lora：hanfu_v29

基础信息

正：

masterpiece, best quality, soft light, chalkboard, traditional media, (chalk outline:0.7), (monochrome:0.4), upper body, (1girl,solo), hanfu,song, smile, 负：

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, lowres, text, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, text, signature, watermark, simple background, toony, dated, low res, line art, flat colors Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 8784925, Size: 768x512

对照图

画风lora测试（盲盒）

基础信息

大模型：revAnimated_v11

lora：blackboard

lora：blindbox_V1Mix

正：

masterpiece, best quality, soft light, chalkboard, traditional media, (chalk outline:0.7), (monochrome:0.4), full body, (1girl,solo), chibi, 负：

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, lowres, text, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, text, signature, watermark, simple background, toony, dated, low res, line art, flat colors

Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 1039761364, Size: 768x512

对照图

阶段结论

兼容性测试中使用了御坂美琴和jojo的人物lora，兼容性一目了然，没什么好说的。

建议黑板绘lora权重0.8~1.1，人物lora按喜好随意。

服装类lora的兼容性测试使用的汉服lora，黑板绘lora权重与汉服lora过高都会让画面元素变得更多，个人更喜欢汉服lora在0.7以下，黑板绘lora在0.6~0.8的画面效果，更像是黑板上应该画出来的样子。

画风类lora的兼容性测试使用的是盲盒lora，出现了画风互相覆盖的问题，均处于低权重情况下勉强能共存，权重较高就会互相污染。不太推荐使用。如果一定要用，建议黑板绘lora权重设置在0.5~0.7，其他画风lora控制在0.3左右。

图生图测试

原图

基础信息

大模型：revAnimated_v11

lora：blackboard

正：

masterpiece, best quality, chalkboard, traditional media, (chalk outline:0.7), (monochrome:0.4), close-up, looking at viewer, (1girl,solo), lace-trimmed veil, brown hair, blue eyes, lace-trimmed collar, black lace, aqua headwear,

负：

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, lowres, text, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, text, signature, watermark, simple background, toony, dated, low res, line art, flat colors

Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 1209696489, Size: 768x512

对照图

阶段结论

图生图测试的效果不太理想，不论是背景的黑板，还是人物的粉笔效果，多次调整关键词都没能捣鼓出来，最终成品的效果更像是底模型的功劳。我自己整活的时候在表情包后面p了个黑板，再次生成的效果还行，可能还是需要对画面进行一定程度的处理之后才能使用。