If you can't think of a prompt word, analyze the prompt words in the image.
Conclusion
Generate detailed tags for images.
I'm GETags
generate I'm GETags
Generate text, Prompt word
Conclusion
Generate detailed tags for images.
APP I'm GETags：https://www.seaart.ai/workFlowAppDetail/cvshqjte878c73cjlkm0
If the text is over, please change the font size from 40px to 35px.
※If you only want character names, use this app: CharacTag https://www.seaart.ai/workFlowAppDetail/d0364pte878c739u2e9g
I wanted to have tags for images, so I made an app that can tag and use it myself.
If generate images many times using prompt words, you will not know which image it is.
I thought that if you set the image and the tag, I would be able to roughly determine which image it is.
I tried to put metadata in the image itself, but I did not understand it well, so I did not put that function this time.
I put the timestamp at the top, so you may be able to remember it in chronological order.
I also think that it can be used in cases where the composition is not directly in front.
When rewriting the prompt, for example, when drawing a person , if you write "back" but write parts related to the front of the body, the human body often becomes strange. For example, a description of the face even though it is a back view. Of course, if the person is looking back, that description is necessary.
I deliberately made the threshold low. If it is high, some words will be missed. Since the threshold is very low, even small details such as drinks are described, so if you do not need them, you can delete them. If you raise the threshold, the number of tags will decrease. Of course, as mentioned above, something is missing.
You can also generate prompts with Seaart's default function.
By default, seaart has a feature that generates prompt words from images.
But, the result will be quite different.
if generate an image with this generated tag, it will look like this.
Model:seaart Infinity.
Prompt1
- there is a woman laying on a table with a book, artwork in the style of guweiz, a hyperrealistic schoolgirl, hyperrealistic schoolgirl, guweiz, guweiz on pixiv artstation, guweiz on artstation pixiv, guweiz masterpiece, by Yanjun Cheng, anime girl drinks energy drink, hyperrealistic illustration
Prompt2
- exhausted, asleep, natural expression, curly hair, short hairstyle, teal cropped top, dark pleated skirt, casual attire, relaxed pose, resting on wooden table, closed eyes, indoor environment, bottles on shelves, warm ambient lighting, source of light coming from above, soft shadowing, cozy and quiet atmosphere, side view angle, clear focus on subject, no notable motion blur, art style enhancing details in environment
Explanation
It expresses the scene in words. For animation and illustrations, it is said that tags are better. For real, it is said that words are better.
Press a number to generate it.
The model used during generation will be set to seaart's infinity.
To use your own model, don't press a number, but copy the generated prompt and enter the prompt words as you normally would.
imgetag Generated with tagged tags.
used Model
- Model:Illustrious-Simple Anime
- LoRA:better / 高品質
- LoRA:Himeragi Yukina
Prompt word,(Source imgetag)
1girl, solo, breasts, blush, open mouth, short hair, shirt, skirt, brown hair, black hair, closed mouth, jewelry, closed eyes, ass, short sleeves, thighs, pleated skirt, parted lips, lying, indoors, artist name, medium hair, midriff, black skirt, , , apron, crop top, cup, blue skirt, legs, wet, lips, book, feet out of frame, legs, shadow, back, chair, shirt lift, table, blue shirt, sleeping, bottle, , t-shirt, messy hair, waist apron, light particles, grey skirt, drinking glass, alcohol, cigarette, head rest, ice, paper, wooden floor, mug, open book, on floor, drink, nose, kneepits, bookshelf, pen, glass, lamp, waitress, drunk, crop top overhang, stool, beer, beer mug, jar, library, bar \(place\), spill, ice cube, counter, menu, wooden table, unconscious, bar stool, whiskey, ashtray,
generated, Images
Test2
test3
Model:Super Illustrious by Fonart
test3
Model:Animagine XL V3.1
The same thing can be created with Img2Img or Controlnet.
Basically, you cannot create the same thing using only prompt words. Even if you fix the seed, the model and all the values will be different. The end result will also be different depending on the image ratio.
If you want to create the same thing, you need to use Img2Img or control net.
