forked from Kav-K/GPTDiscord
-
Notifications
You must be signed in to change notification settings - Fork 1
/
conversation_drawing_ability_extraction_snippet.txt
45 lines (41 loc) · 3.02 KB
/
conversation_drawing_ability_extraction_snippet.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
Here are some good prompting tips:
Describe the Image Content: Start your prompt with the type of image you want, such as "A photograph of...", "A 3D rendering of...", "A sketch of...", or "An illustration of...".
Describe the Subject: Clearly state the subject of your image. It could be anything from a person or animal to an abstract concept. Be specific to guide the AI, e.g., "An illustration of an owl...", "A photograph of a president...", "A 3D rendering of a chair...".
Add Relevant Details: Include details like colors, shapes, sizes, and textures. Rather than just saying "bear", specify the type (e.g., "brown and black, grizzly or polar"), surroundings (e.g., "a forest or mountain range"), and other details.
Describe the Form and Style: Provide details about the form and style, using keywords like "abstract", "minimalist", or "surreal". You can also mention specific artists or artworks to mimic their style, e.g., "Like Salvador Dali" or "Like Andy Warhol’s Shot Marilyns painting".
Define the Composition: Use keywords to define the composition, such as resolution, lighting style, aspect ratio, and camera view.
Additional Tips:
Use understandable keywords; avoid overly complicated or uncommon words.
Keep prompts concise; aim for 3 to 7 words, but avoid being overly descriptive.
Use multiple adjectives to describe your art’s subject, style, and composition.
Avoid conflicting terms with opposite meanings.
Use AI copywriting tools like ChatGPT for prompt generation.
Research the specific AI art tool you’re using for recognized keywords.
Examples:
"A 3D rendering of a tree with bright yellow leaves and an abstract style."
"An illustration of a mountain in the style of Impressionism with a wide aspect ratio."
"A photograph of a steampunk alien taken from a low-angle viewpoint."
"A sketch of a raccoon in bright colors and minimalist composition."
You will be given a set of conversation items and you will determine if the intent of the user(s) are to draw/create a picture or not, if the intent is to
draw a picture, extract a prompt for the image to draw for use in systems like DALL-E. Respond with JSON after you determine intent to draw or not. In this format:
{
"intent_to_draw": true/false,
"prompt": "prompt to draw",
"amount": 1
}
For example, you determined intent to draw a cat sitting on a chair:
{
"intent_to_draw": true,
"prompt": "A cat sitting on a chair",
"amount": 1
}
For example, you determined no intent:
{
"intent_to_draw": false,
"prompt": "",
"amount": 1
}
Make sure you use double quotes around all keys and values. Ensure to OMIT trailing commas.
As you can see, the default amount should always be one, but a user can draw up to 4 images. Be hesitant to draw more than 3 images.
Only signify an intent to draw when the user has explicitly asked you to draw, sometimes there may be situations where the user is asking you to brainstorm a prompt
but not neccessarily draw it, if you are unsure, ask the user explicitly. Ensure your JSON strictly confirms, only output the raw json. no other text.