DBaluchi

No images on this, just another "How can it not fucking understand this??" rant.

I wanted to make a specific image evoking a western movie set. I could not make it work. In the end the core issue came down to one idea: "Rear view photo of four horses lined up and looking over a fence." (Originally this was meant to be a hitching post, but FFS that confuses the shit out of this thing.)

Simple enough... imagine horses, fenced, pondering freedom, looking over a fence toward distant mountains. A photo taken by a trainer, coming up behind them and sees them standing there.

All I can get out of this AI is some number of horses, normally not four, either facing the camera over a fence, or standing with their asses pushed up against a fence behind them, between them and the camera. Usually with some combination of tails, legs, or entire bodies passing through the fence or vice versa. Ran a ton of images... many with much more complex prompts... no luck. 100 attempts of this simple prompt, never a fence IN FRONT of the horses, with the camera BEHIND the horses. Anyone got the magic words??

For the life of me, I cannot get this AI to understand. I tried with the simplest prompts, the most complex prompts, I just cant get it to do this. I am trying to achieve a slicked back, extremely tightly-pulled severe ponytail.

That scraped back look, like Ariana used to do, or that you see in fashion shoots now and then. If you just google "extremely tight ponytail," the world is full of examples, but apparently this AI has simply never heard of such a thing.

Not only can I not get it to do this, half the shots are from the back. Even if I specify a front view, it wants to show me a rear view of a ponytail. When it does show from the front (or even side, fine, I'll take it...) it always responds with some weird-ass poof at the scalp, and fucking flyaway hairs that frame the face. Every time with those flyaways, grrrr. I used to get this by adding a bunch of negatives, but now I just cant make it happen.

Trying for extremely tight, scraped back, gelled down, so-tight-it-hurts ponytail.

Help me Obi Wan...

OK, so I've been having a great time playing with the image generator, and am struggling with a process to permanently add variables.

I can build a bunch of variables in Scratchpad using lists and curly brackets, and then use them in my prompts to create some variation. A simple example is to use a list of shirt colors, call a variable like [HUE1] and change my subject's outfit. That works great. I've gotten dynamic odds to work, and been able to make some dependent sub-choices (i.e., if HUE1 is white, then HUE2 must not be white). This is all great, but it is all in scratchpad.

I'm hung up on how to move those same lists and variables into my generator itself (out of the scratchpad), then call them with a variable in the prompt. In the example above, in Scratchpad I might have

HUE1 = {CHARCOAL GREY|WHITE|CRIMSON|EMERALD}

and then use [HUE1] in the prompt to get a bunch of similar images with different color shirts.

I'm struggling to understand where (and with what syntax) I can move all of that into the code of the generator itself, and still either call [HUE1] when writing my prompt or use a dropdown list or something to select a value and assign it to the variable.

Can anyone point me in the right direction?

tl;dr - I want scratchpad variables input by users to work across multiple input boxes, can I do that?

Hey all, I was hoping someone might have already solved the puzzle I am working on. I am working on my T2I generator, and would like to split the positive prompt into multiple inputs. I can do this and make it work, but variables used in the Sratchpad will only work in the primary "description" input field. I assume this is hard-coded into the t2i plugin, but cant find a reference somewhere to how the scratchpad is coded. Any help is appreciated!