[ from : to : when ] commands no longer work in the prompt since TAMS1 was removed
Feature Example: https://discord.com/channels/1108668896959025214/1108693680321208320/1298436766700797953
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features
Prompt for the lady is "bartender standing behind bar art by Anna Dittmann artistic Cosplay Photograph"
To simulate the effects of [from:to:when] , I will run the prompt as img2img on this abstract noise pattern
Bit crude, but you can see that by having some " noise" in the image initially , output is improved
Time for practical example
Prompt for these are: "[many photography abstractart dark incadescent pink tints dark abstractart with dark lean saturations foggy dark on night grape tones sleek 1girl:dissolve silhouette in this composition barcode Oblivious!Bucky looking at viewer5 steampunk humanoid creature deformed mangled bloody En el fondo hay un espejo. Chas Quilter Swept from the Sea \(1997\) BioSteel Camp :0.1]"
...or some variation. Examples:
Some additional examples using this technique
The model used is the SD 1.5 Indigo Furry model , for those curious
In summary ; pre-rendindering the first 10-20% with a colorful and dark pattern can offset many inherent flaws in SD related to color variation and light contrast
Note thar this feature does not work on TensorArt yet , as TAMS2 syntax for this command is still borked.
Please do me a favour , remind the devs to implement this feature on the discord:
https://discord.com/channels/1108668896959025214/1179345714443190302/1298008202675486810
Cheers,
Adcom
