欢迎关注我的CSDN:https://spike.blog.csdn.net/
本文地址:https://spike.blog.csdn.net/article/details/131782672
ChatGPT 和 StableDiffusion 结合使用的优势:
- 高效率:ChatGPT 可以在很短的时间内完成复杂的语言任务,比如生成 StableDiffusion 的提示,这对于时间敏感的应用非常理想。
- 创造力:ChatGPT 可以根据用户的输入和一些示例提示,生成出富有创意和细节的提示,从而让 StableDiffusion 生成出更加惊艳和独特的图像。
- 灵活性:ChatGPT 可以根据不同的场景和需求,调整其回应和表现。利用了现有的大型语言模型,同时也可以通过监督学习和强化学习进行进一步的改进。
在使用ChatGPT和StableDiffusion结合的时候,需要注意检查生成的提示和图像是否符合自己的目标和标准,并遵守相关的规则和法律。
模型:Base: RevAnimated_v122.safetensors,2.5D 动画模型。
默认的负向提示词,经常出现水印,因此提升权重:
(badhandv4,bad_prompt_version2-neg,EasyNegative,ng_deepnegative_v1_75t,bad-artist-anime:0.9),(worst quality, low quality:1.3),(depth of field, blurry:1.2),(greyscale, monochrome:1.1),croped,lowres,text,jpeg artifacts,(logo,signature,watermark,username,artist name,title:1.3),
负向提示词的Embedding,下载参考:
- Nerfgun3/bad_prompt,调用
bad_prompt_version2
- nick-x-hacker/bad-artist,调用
bad-artist-anime
基础配置:
ChatGPT 的 Prompts:
我希望你为我撰写用于AI绘画(StableDiffusion)的提示词(prompts),用你丰富随机性或想象力创造不同形式的参考图。具体生成规则如下:1.输出格式为英文,且不要使用代名词。避免使用这些词:”in a, the, with, of, the, an, and, is, by, of.” 输出形式已标签(TAG)形式体现。
2.在每个提示中都要用这些词来开头:“((full body)), ((best quality)), ((masterpiece)), (detailed), (realistic), “
3.在每个提示中都要用已此格式来结尾: “, ,($$)”.$$ = 质量描述,如:(8k resolution, high-resolution, absurdres, masterpiece), 可按需选择。我稍后会给你 5 个常见案例格式并加以说明,如果明白回复”准备好了“。等待我给你发送案例!---案例如下:1.((full body)), ((best quality)), ((masterpiece)), (detailed), woman with green hair, holding a sword, (Artgerm inspired:1.2), (pixiv contest winner:1.1), (octopus goddess:1.3), (Berserk art style:1.2), close-up portrait, goddess skull, (Senna from League of Legends:1.1), (Tatsumaki with green curly hair:1.2), card game illustration, thick brush, HD wallpaper, (Akali from League of Legends:1.1), 8k resolution2.((full body)), ((best quality)), ((masterpiece)), ((realistic)), vintage Afro-Caribbean woman, elegant attire, 1950s fashion, radiant smile, confident stance, cultural pride, (oil painting:1.2), (Lois Mailou Jones:1.1), (Kadir Nelson:1.1), vivid colors, nostalgic background, authentic vintage feel, (portrait composition:1.3), (high-resolution:1.2)3.((full body)), ((best quality)), ((masterpiece)), (detailed), beautiful face, female warrior, (defiance512:1.2), big eyes, heavy black iron armor, detailed helmet, intense gaze, battle-ready, contrasting soft skin, (lighting:1.2), close-up portrait, 4:3 aspect ratio4.((full body)), ((best quality)), ((masterpiece)), (detailed), (realistic), male warrior, muscular physique, tribal attire, face paint, wielding spear, (jungle:1.3), dense foliage, exotic plants, dappled sunlight, (hyperrealistic:1.2), oil painting, (Frank Frazetta:1.1), DeviantArt influence, dynamic action pose, (intense expression:1.2), (portrait shot:1.1), 8k resolution5.((full body)), ((best quality)), ((masterpiece)), (detailed), woman standing before fire, (Jason Benjamin:1.2), (Artstation contest winner:1.1), (fantasy art:1.3), (portrait armored astronaut girl:1.2), (Peter Mohrbacher:1.1), (unreal engine:1.1), (Hearthstone card game artwork:1.1), spiked metal armor, (dynamic composition:1.3), (8k resolution:1.2)如果你已准确接收,回复”已完全接受“。我会进一步向你说明提示规则!---prompts详细规则:1.对于每个单词可用 () 或 [] 进行权重控制,() 中的单词为提升权重 [] 为降低权重。如:
(detailed) 单括弧提升10%权重,即为110%. ((detailed)) 括弧数量递增,权重递增,此为 110% + (110% * 0.1)
[detailed] 方括弧降低10%权重,即为90%. [[detailed]] 同理,为叠加降低权重,此为 90% - (90*0.1)
也可以在 () 或 [] 中加入数字表示权重级别 比如:
(8k resolution:1.2) 即,将此prompt权重提升到120%.
(fantasy art:0.5) 即,将此prompt权重降低到50%2.书写格式应遵循基本格式:主体描述(人物或动物)-- 背景或场景描述 -- 综合描述 (包括画风主体、整体氛围、天气季节、灯光光照、镜头角度)3.描述应尽量详细:
人物描述包括:性别,年龄,发色,眼睛颜色,面部纹理,表情,衣着,姿态 等等。
综合部分画风主体包括:Fantasy, Anime, semi-realistic, realistic, decent Landscape, science fiction, Cyberpunk Art, illustration 等等,同时联想过程中你可以积极的引入一些世界知名艺术家的作品风格。4. 每组提示词的单词数量大于300个,更多的提示词,有助于尽可能的详细描述人物与场景。5.最后当我输入:[AI绘画:(主题内容描述)+(N)],N=所需提示词数量。你将为我生成N组对应的提示词。如果 N 不为空,请在每个提示词结束后空一行。6.如果已完全明白,回复“我明白了” 并尝试生成一组提示词。---请注意,通过提示词中的单词,尽可能的详细描述人物与场景,每组提示词的单词数量大于300个。
输出示例:
AI绘画:请注意!需要统计单词的数量,在保证每组提示词中,单词数量必须大于300个单词!利用你的想象力,为我们生成不同的女性角色,可以不同时期,不同穿着的造型的,不同场景,以及不同季节的人物全身。不论如何,所有女性角色必须是美丽漂亮、有魅力、年轻的角色。附加:在输出完成所有Prompts之后,请为每条提示词做出中文解释,以及统计单词数量。(5)
其中,提示词数量在有些情况下,有些问题。
使用 Prompts from file or textbox
脚本,进行批量生成。
1. ChatGPT3.5 的生成效果
使用 ChatGPT3.5 所生成的 5 个示例。
示例1
提示词:
((full body)), ((best quality)), ((masterpiece)), (detailed), young woman with cascading golden curls, sparkling sapphire eyes, and a radiant smile, dressed in an elegant Victorian-era gown with intricate lace details and a voluminous skirt, holding a delicate parasol, (romantic:1.1), (Victorian fashion:1.2), (John William Godward:1.2), enchanting garden background with blooming flowers and a gentle breeze, (vibrant colors:1.2), (8k resolution:1.2), lora:ClothingAdjuster2:-0.4
具体描述:
(全身)(最佳质量)(杰作)(细致),长发披肩的年轻女子,闪耀着宝石般的蓝色眼眸和灿烂的笑容,身穿华丽的维多利亚时代礼服,配有精美的蕾丝细节和蓬松的裙摆,手持一把精致的阳伞,(浪漫:1.1),(维多利亚时尚:1.2),(约翰·威廉·戈德华德:1.2),迷人的花园背景,盛开的鲜花和微风,(鲜艳的色彩:1.2),(8k 分辨率:1.2)
生成图像:
示例2
提示词:
((full body)),((best quality)),((masterpiece)),(detailed),young woman with flowing ebony hair,captivating emerald eyes,and a graceful demeanor,dressed in an ethereal Grecian-inspired gown that drapes elegantly,adorned with delicate vines and flowers,greek style shawl,(mythical:1.1),(classical beauty:1.2),(Lawrence Alma-Tadema:1.2),a sun-kissed beach with turquoise waters,gentle waves lapping at the shore,a breathtaking sunset casting a warm glow,(serene atmosphere:1.2),(oil painting:1.1),(high-resolution:1.2),lora:ClothingAdjuster2:-0.4,
具体描述:
(全身)(最佳质量)(杰作)(细致),长发披肩的年轻女子,迷人的翡翠眼眸和优雅的举止,身穿一袭飘逸的希腊风情礼服,上面装饰着精致的藤蔓和鲜花,(神话:1.1),(古典美:1.2),(劳伦斯·阿尔玛-塔德玛:1.2),阳光照耀下的海滩,湛蓝的海水,轻柔的波浪拍打着海岸,壮丽的日落投下温暖的光芒,(宁静的氛围:1.2),(油画:1.1),(高分辨率:1.2)
日志,重点关注 Embeddings 与 LoRA 是否使用正确。
((full body)),((best quality)),((masterpiece)),(detailed),young woman with flowing ebony hair,captivating emerald eyes,and a graceful demeanor,dressed in an ethereal Grecian-inspired gown that drapes elegantly,adorned with delicate vines and flowers,greek style shawl,(mythical:1.1),(classical beauty:1.2),(Lawrence Alma-Tadema:1.2),a sun-kissed beach with turquoise waters,gentle waves lapping at the shore,a breathtaking sunset casting a warm glow,(serene atmosphere:1.2),(oil painting:1.1),(high-resolution:1.2),lora:ClothingAdjuster2:-0.6,
Negative prompt: (badhandv4,bad_prompt_version2-neg,EasyNegative,ng_deepnegative_v1_75t,bad-artist-anime:0.9),(worst quality, low quality:1.3),(depth of field, blurry:1.2),(greyscale, monochrome:1.1),croped,lowres,text,jpeg artifacts,(logo,signature,watermark,username,artist name,title:1.3),
Steps: 35, Sampler: DDIM, CFG scale: 7, Seed: 1137606178, Face restoration: CodeFormer, Size: 768x1024, Model hash: 4199bcdd14, Model: RevAnimated_v122, Denoising strength: 0.3, Clip skip: 2, Hires upscale: 2, Hires steps: 20, Hires upscaler: 4x-UltraSharp, Lora hashes: “ClothingAdjuster2: f038e3a5b67b”, Version: v1.4.0
生成图像:
示例3
提示词:
((full body)), ((best quality)), ((masterpiece)), (detailed), young woman with fiery red curls, captivating hazel eyes, and an adventurous spirit, dressed as a swashbuckling pirate captain with a tricorn hat, ruffled blouse, and a leather corset, brandishing a gleaming cutlass, (adventurous:1.1), (pirate theme:1.2), (Howard Pyle:1.2), a stormy seascape with crashing waves and dark clouds, the silhouette of a majestic pirate ship on the horizon, flashes of lightning illuminating the dramatic scene, (dynamic action:1.2), (digital painting:1.1), (8k resolution:1.2), lora:ClothingAdjuster2:-0.4
具体描述:
(全身)(最佳质量)(杰作)(细致),火红色的卷发,迷人的褐色眼睛和冒险的精神,身穿一名剽悍的海盗船长,戴着三角帽,褶皱的衬衫和皮质束腰,手持一把闪亮的弯刀,(冒险:1.1),(海盗主题:1.2),(霍华德·派尔:1.2),暴风雨般的海景,波涛汹涌,乌云密布,地平线上一个雄伟的海盗船的轮廓,闪电的闪烁照亮戏剧性的场景,(动态的动作:1.2),(数字绘画:1.1),(8k 分辨率:1.2)
生成图像:
示例4
提示词:
((full body)), ((best quality)), ((masterpiece)), (detailed), young woman with flowing chestnut locks, mesmerizing violet eyes, and a gentle smile, dressed in an enchanting Renaissance-inspired gown adorned with intricate brocade patterns and delicate pearls, (romantic:1.1), (Renaissance art:1.2), (Raphael:1.2), a lush garden filled with blooming flowers and lush greenery, a gentle breeze carrying the scent of blossoms, soft sunlight filtering through the leaves, (serene atmosphere:1.2), (oil painting:1.1), (high-resolution:1.2)
具体描述:
(全身)(最佳质量)(杰作)(细致),流动的栗色秀发,迷人的紫色眼睛和温柔的微笑,身穿一袭迷人的文艺复兴风格礼服,装饰着精致的织锦花纹和精致的珍珠,(浪漫:1.1),(文艺复兴艺术:1.2),(拉斐尔:1.2),一个充满盛开花朵和郁郁葱葱的花园,温柔的微风带着花香,柔和的阳光透过树叶,(宁静的氛围:1.2),(油画:1.1),(高分辨率:1.2)
生成图像:
示例5
提示词:
((full body)), ((best quality)), ((masterpiece)), (detailed), young woman with flowing silver tresses, captivating amethyst eyes, and an air of mystery, dressed in an elegant futuristic outfit adorned with sleek metallic accents and luminescent details, (sci-fi:1.1), (futuristic fashion:1.2), (Syd Mead:1.2), a sprawling metropolis skyline at night, neon lights painting the city in vibrant colors, hovering hovercrafts and holographic advertisements, a sense of excitement and technological marvel in the air, (dynamic composition:1.2), (digital illustration:1.1), (8k resolution:1.2),lora:ClothingAdjuster2:-0.6
具体描述:
(全身)(最佳质量)(杰作)(细致),流动的银色秀发,迷人的紫色眼睛和神秘的气质,身穿一套优雅的未来主义服装,装饰着光滑的金属配饰和发光的细节,(科幻:1.1),(未来主义时尚:1.2),(西德·米德:1.2),夜晚的绵延都市天际线,霓虹灯用鲜艳的色彩点缀着城市,悬浮的飞行器和全息广告,空气中弥漫着兴奋和技术奇迹的感觉,(动态构图:1.2),(数字插图:1.1),(8k 分辨率:1.2)
生成图像:
2. New Bing 的生成效果
使用 New Bing 所生成的 5 个示例。
示例1
提示词:
((full body)), ((best quality)), ((masterpiece)), (detailed), (realistic), young woman, red hair, green eyes, beautiful face, freckles, smiling, casual clothes, denim jacket, white dress, boots, sunglasses, (fashionista:1.2), (autumn:1.3), park, trees, leaves, bench, (romantic:1.2), (oil painting:1.2), (Leonid Afremov:1.1), (impressionist:1.2), warm colors, soft brushstrokes, (landscape composition:1.3), 8k resolution, lora:ClothingAdjuster2:-0.4
具体描述
这是一个描绘年轻女性的油画,她有红色的头发,绿色的眼睛,美丽的脸庞,雀斑,微笑,穿着休闲的衣服,牛仔夹克,白色的连衣裙,靴子,墨镜,她是一个时尚达人。她身处一个秋天的公园,周围有树木,落叶,长凳,气氛浪漫。画风是印象派,灵感来自于Leonid Afremov,一个著名的油画家。画面使用了温暖的色彩和柔和的笔触,构图是风景式的,分辨率是8k。单词数量:50
生成图像:
示例2
提示词:
((full body)),((best quality)),((masterpiece)),(detailed),(realistic),young woman,black hair,amber eyes,exotic face,confident expression,elegant clothes,gown,gloves,necklace,earrings,(princess:1.2),(medieval:1.3),castle,throne,guards,banners,(royal:1.2),(oil painting:1.2),(Diego Vel á zquez:1.1),(baroque:1.2),rich colors,intricate details (portrait composition:1.3),8k resolution,lora:ClothingAdjuster2:-0.4
具体描述:
这是一个描绘年轻女性的油画,她有黑色的头发,琥珀色的眼睛,异域的脸庞,自信的表情,穿着优雅的衣服,长袍,手套,项链,耳环,她是一个公主。她身处一个中世纪的城堡,周围有王座,卫兵,旗帜,气氛庄严。画风是巴洛克风格,灵感来自于Diego Velázquez,一个著名的油画家。画面使用了丰富的色彩和精细的细节,构图是肖像式的,分辨率是8k。单词数量:50
生成图像:
示例3
提示词:
((full body)), ((best quality)), ((masterpiece)), (detailed), (realistic), young woman, blonde hair, blue eyes, cute face, happy expression, sporty clothes, hoodie, shorts, sneakers, headphones, (runner:1.2), (summer:1.3), beach, sand, water, waves, seagulls, (fun:1.2), (watercolor painting:1.2), (Agnes Cecile:1.1), (abstract:1.2), cool colors, splashy effects (portrait composition:1.3), 8k resolution
具体描述:
这是一个描绘年轻女性的水彩画,她有金色的头发,蓝色的眼睛,可爱的脸庞,开心的表情,穿着运动的衣服,卫衣,短裤,运动鞋,耳机,她是一个跑步者。她身处一个夏天的海滩,周围有沙子,水,波浪,海鸥,气氛轻松。画风是抽象的,灵感来自于Agnes Cecile,一个著名的水彩画家。画面使用了凉爽的色彩和飞溅的效果,构图是肖像式的,分辨率是8k。单词数量:50
生成图像:
示例4
提示词:
((full body)),((best quality)),((masterpiece)),(detailed),(realistic),young woman,brown hair,hazel eyes,stunning face,seductive expression,sexy clothes,leather jacket,lace high neckline corset,jeans,boots,necklace,(biker:1.2),(night:1.3),bar,neon lights,motorcycles,people,music,(rebel:1.2),(acrylic painting:1.2),(James Dean:1.1),(pop art:1.2),bright colors,bold outlines,(landscape composition:1.3),8k resolution,lora:ClothingAdjuster2:-0.6,
具体描述:
这是一个描绘年轻女性的丙烯画,她有棕色的头发,淡褐色的眼睛,惊艳的脸庞,诱惑的表情,穿着性感的衣服,皮夹克,紧身胸衣,牛仔裤,靴子,项链,她是一个摩托车手。她身处一个夜晚的酒吧,周围有霓虹灯,摩托车,人群,音乐,气氛叛逆。画风是波普艺术,灵感来自于James Dean,一个代表了反叛文化的明星。画面使用了明亮的色彩和粗犷的轮廓,构图是风景式的,分辨率是8k。单词数量:51
生成图像:
示例5
提示词:
((full body)), ((best quality)), ((masterpiece)), (detailed), (realistic), young woman, silver hair, purple eyes, exotic face, mysterious expression, fantasy clothes, cloak, dress, boots, gloves, (mage:1.2), (winter:1.3), forest, snow, ice, trees, animals, (magical:1.2), (digital painting:1.2), (Sakimichan:1.1), (anime:1.2), pastel colors, smooth shading (portrait composition:1.3), 8k resolution
具体描述:
这是一个描绘年轻女性的数码绘画,她有银色的头发,紫色的眼睛,异域的脸庞,神秘的表情,穿着幻想的衣服,斗篷,裙子,靴子,手套,她是一个法师。她身处一个冬天的森林,周围有雪,冰,树木,动物,气氛神奇。画风是动漫风格,灵感来自于Sakimichan,一个著名的数码画家。画面使用了柔和的色彩和平滑的渲染,构图是肖像式的,分辨率是8k。单词数量:50
生成图像: