[ AI Art ] - stable diffusion 討論 分享 01

Install 完poseX, enable 埋controlnet,但係都冇個editor 出嚟

大家有冇咩解決方法

有冇例子

主要係kpop嗰啲明星lora 動漫lora我就唔係好清楚有冇
同埋最近喺discord聽個韓仔講 (未fc) 話有人喺佢哋韓國ai art論壇度拎人哋免費lora嚟賣 搞到好大件事咁 而且話驚俾啲韓國畫家告 所以啲嗰邊人依家唔太敢周圍免費放lora

呢個lora個強度好高
用左好似換左個checkpoint咁

整咗幅動態wallpaper


Link: https://steamcommunity.com/sharedfiles/filedetails/?id=2957933274
或者係Wallpaper Engine search Wanderer in Jungle
Model: https://civitai.com/models/24596/jellymix

全部人都用黎整J圖

咩屬性嘅lora，角色定係style

style黎

條link終於過到審查

超級方便!

sd_model_checkpoint,sd_vae,CLIP_stop_at_last_layers ,eta_noise_seed_delta

Noise seed有咩用？

當你gen圖嘅時候，佢會將呢個數字加到你嘅seed入面,如果有人用咗呢個[eta noise seed delta]，而你想再生成同一張圖片嘅時候，你就要設定同樣嘅數字.


平時基本上唔使理佢..

點整高res圖？
Canvas較16:9 再upscale 上2k/4k 再逐忽inpaint加details?
我自己整一係爆vram 一係load勁耐

我gen1280*840圖再用ultimate SD upscale放大4倍，重繪幅度揀0.4左右，之後再impainting，要揀填充同只修改遮蔽內容，就可以直接inpaint 5k圖，之後再resize做4k圖

steampunk and lots of pipe?

有無wildcard方法分享
有時諗唔到嘢好想用下chatgpt wildcard extension，但冇api用唔𠹌到

Civital有個分類係wildcards， 可以搵吓。

有冇人用colander遇過類似情況，想整大d解像度既時候，出錯誤訊息，試過setting剔左Upcast cross attention layer to float32，都冇用
modules.devices.NansException: A tensor with all NaNs was produced in Unet. This could be either because there’s not enough precision to represent the picture, or because your video card does not support half type. Try setting the “Upcast cross attention layer to float32” option in Settings > Stable Diffusion or using the –no-half commandline argument to fix this. Use –disable-nan-check commandline argument to disable this check.

有冇人用colab遇過類似情況 ，想整大d解像度既時候，出錯誤訊息，試過setting剔左Upcast cross attention layer to float32，都冇用
modules.devices.NansException: A tensor with all NaNs was produced in Unet. This could be either because there’s not enough precision to represent the picture, or because your video card does not support half type. Try setting the “Upcast cross attention layer to float32” option in Settings > Stable Diffusion or using the –no-half commandline argument to fix this. Use –disable-nan-check commandline argument to disable this check.

我記得有人問過點樣wedge prompt weight, 可以用 x/y/x plot 嘅 prompt S/R. (search and replace)





don't do (big laugh:2.0)

試吓重新喺civital down過個model? 聽講早幾日down 嘅file有問題。

試吓重新喺civital down過個model? 聽講早幾日down 嘅file有問題。

多謝回覆



搞掂左
呢3行最尾加返 —disable-nan-check 就得

唔知點解用唔到upscaler
次次禁都出error

MacBook m1 m2 都係垃圾
係AI 既世界，只有NVIDIA
已賣 M1 Max 換左4090laptop