yamit

joined 9 months ago
[–] yamit@lemmy.world 1 points 38 minutes ago

Ok after some tests, "aesthetic" is not needed, just use numbers, and it works only for anime, even illustrations or 3d-ish do not seems to work. Here the tests: number scores

Also the model was updated somewhere after june 30th, and I'm not saying it because "i feel the images are better now", but because ALL seeds before that date are -very- different, not the same usual difference that seeds already have, but a complete change between all variations from before X all variations now.

[–] yamit@lemmy.world 1 points 13 hours ago* (last edited 12 hours ago)

Editing my comment because looks like placebo, or there are more tags? i need to test some things and after that I'll make a new comment here.

7
submitted 1 week ago* (last edited 1 week ago) by yamit@lemmy.world to c/perchance@lemmy.world
 

Open my generator in private mode: https://perchance.org/ai-art-generator-advanced, there's no content, the page is almost blank, but it is labeled as NSFW, I deleted everything over 8 hours ago before creating this post. Edit: I had to put the gen back to normal because there are people who uses it. But I took a screenshot, look, there's nothing nsfw:

Another problem that i will just copy paste from another comment: "the 'Content Warning' should detect when the 'nsfw words' are from the generator content warning, if i put a warning message i get another and then the gen is set automatically to private and the meta don't appears on google... But if i remove the message from my page, that have potential nsfw content created by others or that can be found in searchs, then the message is gone and now i have a generator with nsfw content without any message..."

The problem could be solved by using an imported page, so I can put lists and regex to prevent people from creating nsfw without being punished for it... But things are different with the Ai art generator advanced, is there any way to solve this?

[–] yamit@lemmy.world 2 points 2 weeks ago* (last edited 2 weeks ago) (1 children)

In the example in my last comments is possible to edit. If you want the generation to use the edits made in the prompt box, then check the line 44 in the left panel, there is this code "prompt = [prompt.input || input.artStyle.prompt]"

So if there is something in there (prompt.input) it will use the edits. If not then the input.artStyle.prompt will be used. The "||" means "or", if there is no data (empty string) then it will skip to the next item, so that says (use the thing from prompt.input, OR, use the thing from input artStyle.prompt)

note that "prompt.input" is the variable declared in the right panel, it is not the same as "input.fullprompt", but you can use "input.fullprompt" or whatever is the name of your inputs without problem

[–] yamit@lemmy.world 4 points 1 month ago

There's a hidden button below the "Leave". Click in there.

[–] yamit@lemmy.world 3 points 1 month ago* (last edited 1 month ago) (4 children)

Is possible (like in the Koto comment) and you can make it to visualize and edit only if something is checked/cliked, I think I saw some generator with this option in the generator list.

I just made it to test, i think i will put it in my gen.. https://perchance.org/example-aaaa1

Also try learning js with AI, ask the AI how to do some functions in Javascript and try learning some js basics, then try to apply that in something simple in a new page on perchance. Also the basic perchance syntax is very helpful (selectOne, selectAll, getChildNames, etc)

[–] yamit@lemmy.world 1 points 1 month ago

The problem is the condition, that list is not an array, I'm pretty sure it is an object. An object is something like "BannedPromptPhrases = {banWord1: "banWord1", banWord2: "banWord2"}; They are values stored inside the var. So the commands below (join() split() map() filter()) converts it to an array, the condition should be AFTER. But i think you don't need that condition, and if you are lazy just change it to something that's always true like 2===2.

[–] yamit@lemmy.world 3 points 1 month ago (5 children)

I dont know if this is really the problem because i was testing using my own pics, but i forgot that i can still see my pics when they are banned if i am with admin mode disabled. But make sure your list is an array:

let furryJail = jail.bannedPromptPhrases.selectAll.join(','); furryJail .split(",") .map(tag => tag.trim()) .filter(tag => tag !== "") ; dataObj.bannedPromptPhrases.push(...furryJail);

like

also you need to add it in 2 functions, changeAll() and changeGallerySubChannel() so it runs at the start.

[–] yamit@lemmy.world 5 points 1 month ago (1 children)

The old model was so bad in doing many things, sure if you want a character with simple shirt and jeans or red pleated skirt (the only type of skirt it understands) and eyes and hair color, ok, then you get consistency. But try to add more details, hair accessories, colors, bandages, bangs. It simply can not, like i said in this post https://lemmy.world/post/29911427

To get consistency with anime character you need to describe everything, the lighting and shadows, the background and position, and do not use quality negatives. I have many characters that i had to simplify to use in SD, but now i get consistency in it's characteristics, colors and outfit, and they still look the same as before. Tip: use tags to describe characteristics, and describe positions, background, style and others elements with natural language.

[–] yamit@lemmy.world 2 points 1 month ago* (last edited 1 month ago)

I think some label that is displayed in the img data (label:::Label) would do the trick (but it can only be made in the text 2 image plugin i think), . For now you can use the negatives to label and filter with bannedNegatives, they affect less or nothing.

2
submitted 2 months ago* (last edited 2 months ago) by yamit@lemmy.world to c/perchance@lemmy.world
 

The new model is good. Compared to the older, now we have more control over the image. What we can do now that was impossible before:

  • Create male characters (extra: with feminine features)
  • Color control without mixing everything (hair, clothes, objects, skin, scenery, eyes, heterochromia with each color defined, etc)
  • Character characteristics control without mixing them up (now is possible to make each character distinguible and interact with the other)
  • Control over small details (pendant with specific format/shape, details and decorations in objects)
  • Control over location and quantity (so if you use "bandage" in the arm, the character will have it in the arm instead of random parts in the body and floor, and if you want a single flower, tomb or ribbon, the image will no be full of it)
  • Lighting control (lighting and shadows also makes more sense)
  • Text and symbols prints, is even 'possible' to define the shape and size, and even use emoticons.
  • Better cinematic effects (including glow, fire, haze, fog, lightning)
  • Better camera lens and effects
  • The resolution is still bad but somewhat the pics have less noise and blurred elements

And also all the others things from the older model except (things the model can't do):

  • Defined art style (the background/scenery elements and objectsnhave great impact in the composition, and they are so random)
  • Specific tags for anime, the model don't know what is serafuku for example

The new flux model also have so much bias for abstract, painting and expressionism. Overall, if the model will be really improved then I think it is better for perchance because:

  • Unique (there are many free sites with SD1.5 and SDXL, but not for flux, they are very limited and slow)
  • For perchance generators and games that likes rpg and fantasy things, the style, quality and variety are much better.

Personally i prefer Illustrious, but there are many good options in the web.

[–] yamit@lemmy.world 2 points 2 months ago (2 children)

To solve it, instead of:

item = value

change to:

item
 value

for all the items in options.

 

~~urgent issue~~

One of the problems with creating an img generator is the lack of "power" with galleries, you can't control what is posted in public, banning images doesn't solve anything since it only hides from that problematic people from the internet.

But downvoting is effective, it really does make the images disappear (unless if someone with knowledge edits the page just to see), so I was thinking, why not give high voting score power to the gen owners? A simple downvote and poof, that problematic image disappears from trending.

This still maintains the user's freedom in creating any content they want, while giving peace of mind to the creators (why did I make this gen? to makes me feel bad? maybe I'll delete it, screw the users...)

sorry not sorry the ping @perchance@lemmy.world

view more: next ›