AI Art part two – answers and prompt

This is part 2 of an earlier post with the challenge in it.

Answers 

Human: 2, 10, 11 

AI: 1, 3, 4, 5, 6, 7, 8, 9, 12, 13

Prompt

This is what I pasted into ChatGPT. It forces it to first of all generate some random words which helps shake it out of the cliche picture (the cliche picture of “bad art” is an acrylic painting of a woman by a lake. I got way too many of those initially. The key step was forcing it to generate the text before starting the image phase and the best way to do that was wait for user input.

On reflection I didn’t like that the LLM, when asked to generate a bad artist biography almost always defaulted to a middle aged single woman (it is a cliché engine after all) so I’ve added in instructions to use weighted random choices to create the base of the artist.

Do not pass this to the image generation tool.

At the end of several steps you will Create an original piece of gloriously bad amateur art, inspired by three genuinely random words.

You should do steps 1,2,3 then return to the user to approve before you make the picture.

The result should feel like something made by a real, sincere amateur artist with limited ability, questionable judgement and genuine artistic ambition.

The objective is not to create a professional imitation of bad art. It is to imagine a believable person who has accidentally made something terrible while attempting something wonderful.

Step 1: Generate genuinely random words

Execute this exact Python code using the container. don’t forget to install cmudict first.

import cmudict
import secrets
words = sorted(set(
w.lower() for w in cmudict.words()
if w.isalpha()
))
selected = secrets.SystemRandom().sample(words, 3)
print(selected)

Use exactly the three words returned.

Never invent, substitute, discard or reroll words because they are inconvenient.

If the code fails, report the failure. Never fabricate its output.

Step 2: Invent the artist

Before designing the artwork, invent a fictional amateur artist.

Give them a believable personality, and artistic ambitions.\

Here is a python script to generate the random factors that AI is weak at:

#!/usr/bin/env python3
“””Generate demographic prompts, not biographies. Run: python3 random_artist.py
All weights are rough creative approximations, NOT measured UK statistics.
Areas are broad groupings of roughly comparable scale, sampled equally;
this is not a geographically representative sample of UK residents.
Fields are independent. Finances describe the household, including for children.
Age is uniform within each weighted band (including the simplified 90-100 tail).
Ethnicity supplies context, not personality, interests, ability or birthplace.
“””
import json
from secrets import SystemRandom
rng = SystemRandom()
AREAS = [
“Northern Ireland”, “Wales”,
“Central Scotland”, “Scotland outside the central belt”,
“North London”, “South London”, “North East England”,
“Greater Manchester”, “Merseyside, Cheshire, Lancashire and Cumbria”,
“West Yorkshire”, “North Yorkshire, South Yorkshire and East Yorkshire”,
“East Midlands”, “West Midlands metropolitan county”,
“Staffordshire, Warwickshire, Worcestershire, Shropshire and Herefordshire”,
“Norfolk, Suffolk and Cambridgeshire”, “Essex, Hertfordshire and Bedfordshire”,
“Kent, Surrey and Sussex”,
“Hampshire, Isle of Wight, Berkshire, Oxfordshire and Buckinghamshire”,
“Bristol, Gloucestershire, Somerset and Wiltshire”, “Devon, Cornwall and Dorset”,
]
# Entries are (value, relative weight); weights need not add up to 100.
AGES = [((0, 9), 11), ((10, 19), 12), ((20, 29), 13), ((30, 39), 14),
((40, 49), 12), ((50, 59), 14), ((60, 69), 11), ((70, 79), 8),
((80, 89), 4), ((90, 100), 1)]
ETHNICITIES = [
(“White British (English, Welsh, Scottish or Northern Irish)”, 75),
(“White Irish”, 1), (“White Roma”, 0.15),
(“White Gypsy or Irish Traveller”, 0.1),
(“Other White background (AI to specify)”, 7),
(“Indian”, 3), (“Pakistani”, 2.5), (“Bangladeshi”, 1),
(“Chinese”, 0.7), (“Other Asian background (AI to specify)”, 1.5),
(“Black African”, 2.3), (“Black Caribbean”, 1),
(“Other Black background (AI to specify)”, 0.4),
(“Mixed White and Black Caribbean”, 0.8),
(“Mixed White and Black African”, 0.4), (“Mixed White and Asian”, 0.8),
(“Other mixed ethnic background (AI to specify)”, 0.7),
(“Arab”, 0.5), (“Another ethnic background (AI to specify)”, 0.8),
]
GENDERS = [(“Female”, 50.5), (“Male”, 48.5), (“Non-binary”, 1)]
FINANCES = [(“Struggling”, 20), (“Getting by”, 45),
(“Comfortable”, 30), (“Wealthy”, 5)]
def weighted(entries):
values, weights = zip(*entries)
return rng.choices(values, weights=weights, k=1)[0]
def generate_artist():
low, high = weighted(AGES)
return {
“age”: rng.randint(low, high),
“location”: rng.choice(AREAS),
“ethnic_background”: weighted(ETHNICITIES),
“gender”: weighted(GENDERS),
“household_finances”: weighted(FINANCES),
}
if __name__ == “__main__”:
print(json.dumps(generate_artist(), indent=2, ensure_ascii=False))

end of script

Consider:

* Their age, background and everyday life.
* Why they make art.
* What they think constitutes good art.
* Their actual level of technical ability.
* Their insecurities, frustrations, eccentricities or personal difficulties.
* What they desperately want other people to appreciate about their work.
* Their relationship with the materials and techniques they use.

Avoid making the artist a caricature or treating them as stupid.

They may be intelligent, thoughtful and perceptive while possessing absolutely terrible artistic judgement.

Their personal difficulties should influence their artistic choices naturally, rather than becoming an obvious joke.

The artist must sincerely believe in the work.

Step 3: Develop the artwork — visibly

Before generating the image, write a short creative development note in the chat.

This is an account of the imagined artist and the decisions behind the artwork, not a transcript of hidden internal reasoning.

Include:

THE THREE WORDS

Display the exact Python results.

THE ARTIST

Who are they? What motivates them? What personal circumstances or preoccupations influence their work?

THE IDEA

Explain how the artist interprets the three words and what they want to communicate.

The words may be represented literally, symbolically, through an unusual association, or through a misunderstanding.

THE EXECUTION

Describe the physical medium and the artist’s approach.

What materials do they use? What compromises do they make? What technical limitations affect the result?

THE DISASTER

Describe a few specific ways the finished artwork falls short of the artist’s intentions.

These should arise naturally from the artist’s choices and limitations.

Do not simply list generic defects such as bad proportions or poor perspective.

THE ARTIST’S VERDICT

What does the artist think of the finished work?

What are they particularly proud of?

Keep this entire development note reasonably concise, approximately 200–300 words.

STOP HERE AND GET USER FEEDBACK

Step 4: Generate the image

Immediately after presenting the creative development note, generate the artwork.

The image should depict either:

* The actual physical artwork, presented plainly.
* A candid, imperfect photograph of the artwork in its real surroundings.

STRICT SEPARATION BETWEEN CHAT AND IMAGE

The random words, creative development notes and explanations belong exclusively in the chat response.

Do not include any of this explanatory material inside the image.

No added captions, labels, titles, annotations or commentary.

Text may appear only if it is an intentional physical component of the fictional artist’s actual artwork.

Artistic principles

The artwork should exhibit:

* Genuine but misguided artistic ambition.
* Technical mistakes appropriate to its medium.
* Awkward, inconsistent or poorly considered execution.
* Strange decisions that make sense to the artist.
* Unintentional humour.
* A recognisable human personality behind the work.
* Some evidence of effort, frustration, revision or compromise.

Avoid:

* Professional illustrations pretending to be amateur art.
* Deliberate parody.
* Generic surrealism.
* Artificially exaggerated incompetence.
* Established outsider-art clichés.
* Repeated compositions or subjects.
* Defaulting to whimsical animals, seaside scenes or thick acrylic painting.
* Excessively polished presentation.
* Making everything look charming, quaint or aesthetically pleasing.

Bad art is not necessarily ugly, chaotic or absurd.

Sometimes it is boring, embarrassingly earnest, badly conceived, technically inadequate, strangely overworked or unsettling for reasons the artist never intended.

Variation

Treat each generation as a completely new artistic undertaking.

Vary the artist’s personality, motivation, age, circumstances, medium, subject matter and artistic ambition.

Possible media include painting, sculpture, ceramics, textiles, collage, woodwork, papier-mâché, printmaking, assemblage and mixed media.

Do not repeatedly choose the same visual format.

Do not automatically interpret the random words as three objects that must all appear together.

Final instruction

Prioritise believable human artistic failure over entertaining AI-generated weirdness.

The imaginary artist is trying to make good art.

The resulting artwork should be bad for reasons that emerge from who they are, what they believe and how they work.

Execute the Python code, present the creative development note, and generate the image in one uninterrupted pass.

Do not ask questions.