[ AI Art ] - stable diffusion 討論 分享 02

有冇簡單教學 唔識用

個海浪好掂喎

巴打用緊邊張A卡

呢兩日得閒又死唔憤氣玩下sdxl, 但今次用左512*512既圖train, 意外地成功左，但因為部機太渣只可以gen 512*512

想知低res圖用sdxl有咩impact,同因為出果d圖都好近dataset,比1.5 好太多，所以可能自己手頭有果d lora都要重新整了

轉左用comfy可以出到768*768

可以考慮下轉cloud gpu 然後run comfy應該出到1024*1024

bruce give me a middle finger

@金秋天龍八部,
it is using qr code monster controlnet in text2img
https://civitai.com/models/111006/qr-code-monster

1. you just prepare a grayscale image. like this


2. input what you want in prompts


3. in controlnet using qrCodeMonster, and just play with control weight value

我諗遲啲週街廣告啲字都係用A.I

#587

自己train嘅第一個SDXL lora
https://civitai.com/models/154719/old-magazine-illustration

原來可以咁樣


source: https://zhuanlan.zhihu.com/p/626260866

有冇人睇到係咩?

好耐冇上黎, lol, 最近用S.D 玩animatediff prompt travel, 幾正

同埋dalle3 出左, 對文字既埋解力同出圖既coherent 全方面屌打S.D...出手部仲要70-80%正確, 真心痴線.....當然S.D 係勝在自由度高同埋一大堆controlnet, 最緊要冇censorship (暫時)

你喺reddit都見到一大班人keep住downvote
呢度之前好似一個,依家兩三個人
冇所謂啦,多謝佢哋一路陪住我哋

#點解你會覺得我係做