SD终于把XL 1.0放出来了,有两个,我只测了base,先说结论,进步很大,很少数的提示词就能出很优秀的图。 重点:
- 分辨率:推荐1024*1024以上,太小了图质量很差
- 采样器:推荐DPM adaptive、DPM++ 2M Karras、DPM++ 2S a Karras。Euler a效果不太行了,太柔和了,细节不够
- 采样步数:正常20步就足够了
- 提示词:现在MJ的提示词大部分都能直接拿到SD XL 1.0模型上用了,甚至不怎么需要写反面提示词
简单对比一下:
动漫风格
提示词:anime artwork a girl looking at the sea, dramatic, anime style, key visual, vibrant, studio anime, highly detailed
负面提示词:photo, deformed, black and white, realism, disfigured, low contrastCopied!
很明显,sd xl 1.0比1.5模型好太多了,采样器对比也可以看出来DPM adaptive、DPM++ 2M Karras、DPM++ 2S a Karra这三个细节足一点,Euler a细节不够了,而且偏柔和
照片风格
提示词:cinematic photo of a construction worker looking down at city. 35mm photograph, film, bokeh, professional, 4k, highly detailed
负面提示词:drawing, painting, crayon, sketch, graphite, impressionist, noisy, blurry, soft, deformed, ugly
猫猫
提示词:a white cat with blue eyes running through a fire filled field of rocks and rocks,with a castle in the background,Chris LaBrooy,magic the gathering artwork,an airbrush painting,fantasy art
负面提示词:drawing,painting,crayon,sketch,graphite,impressionist,noisy,blurry,soft,deformed,ugly,
评论区