This is a test of multiple concepts, even at 5k steps trained, this lora can make mistakes, so ill make a command list of what works and what not.
every command was tested on qwen edit nunchaku, so takes me less than one minute, see workflow here: https://civitai.com/models/2143038/luisap-advanced-qwenqwen-edit-nunchaku-workflow-with-insane-upscaler-rtx-4060-ti-8vram
Apply greenscreen
Apply pixelization on background
Apply pixelization on subject
Apply bokeh blur on background
Apply bokeh blur on subject
Apply zoom blur on background
Apply motion blur on background [Don't respect the token vertical and horizontal sometimes]
Apply motion blur on subject [Don't respect the token vertical and horizontal sometimes]
Apply greyscale on background [ need 2 strenght to work, sometimes struggles ]
Apply greyscale on subject [ Same, need 2 strenght on lora to works ]
Apply constrast on background/subject [ don't work ]
Apply matte [ this command gotta dumb, only isolate the caracter in a black background]
Apply inverse matte [ sometimes get black white, and sometimes get white black]
Apply drunk blur on background/subject [ struggles too ]