Art Review Generator
The Art Review Generator is a tool specializing in the generation and analysis of language used in the discourse of fine art and culture. Rooted in natural lang...
Last checked:
What is Art Review Generator?
The Art Review Generator is a natural language processing tool and text generator that creates art review text based on user-provided word prompts. It generates medium-length sentences that approximate its training data, which consists of 57 years of actual art reviews from Artforum magazine, totaling over 21,000 reviews. The tool uses a modified GPT-2 language model trained specifically on art criticism language rather than generic social media text.
Key features include generating unique art reviews that describe intention, emotion, technique, and impact of artworks, using deep matrices of probability to determine word combinations. Users submit prompts in English and receive generated results via email since the generation process takes 2-15 minutes. The tool is designed for generating and analyzing language used to describe fine art and other cultural expression, producing text that mimics the distinctive academic and esoteric jargon typical of art reviews.
This tool is primarily for artists, art students, art critics, and anyone interested in art discourse who wants to explore how language describes cultural expression. It was created by Joshua Curry (art name: Lucidbeaming) in San Jose, CA, who used it to generate introductory text for his book of procedurally generated photographs. The generator is currently paused but remains available as a web-based application for exploring machine learning in cultural analysis.
Art Review Generator pricing
Pricing model: Free
Free tier - the tool is free to use with no paid plans mentioned. Users submit prompts via email and receive generated results at no cost. The creator mentions avoiding expensive hosting and compute time costs by using the delayed email delivery process.
Art Review Generator pros
- Uses unique training data from 57 years of Artforum reviews
- Over 21,000 art reviews in training dataset
- Generates novel text word by word, not copied reviews
- Captures distinctive academic art criticism language
- Includes esoteric jargon challenging for traditional models
- Creates unique reviews for each prompt submission
- Explores bias and prejudice in historical art criticism
- Glitches and loops can be poetic and interesting
- Diverse results without over-optimization
- Free to use with email submission
- Web-based browser-accessible application
- Analyzes language of fine art and cultural criticism
- Simulates thesis and supporting statements structure
- Combines perspectives from multiple decades
- Created by professional artist Joshua Curry
Art Review Generator cons
- Review generator is currently paused
- Generation takes 2-15 minutes per request
- Requires email to receive results
- Not real artificial intelligence despite claims
- Interpreter doesn't understand prompt meaning
- Produces nonsense and non-sequiturs sometimes
- Contains biases from historical training data
- No real-time results available
Frequently asked questions about Art Review Generator
What is this tool?
This is a tool to generate and analyze language used to describe fine art and other cultural expression. It takes a set of words as a prompt and generates medium-length sentences that approximate the training data, which is 57 years of actual art reviews from Artforum magazine.
How did you make it?
Essentially, I took an existing language model (GPT-2) and trained it with 57 years of art reviews from ArtForum Magazine, over 21,000 reviews. I used a training setup in a Google Collab notebook to do that. Then I put together a web application that uses that model to generate results based on public submissions.
Where did the idea come from?
I published a small book of procedurally generated photographs and it seemed appropriate to also use a computer generated introduction. I tried various online tools but most were too limited or generic. The majority of NLP training uses text from social media with dubious vocabulary. I used Writing With Transformer by Hugging Face for my book but required extensive editing, so I was inspired to make my own targeted tool.
It generates nonsense sometimes. Don't you want to fix that?
No. By leaving it in this state without optimizing, results are diverse and interesting. It wasn't my goal to make a commercial product or something idealized for popular consumption. The glitching, repetition, and non-sequiturs are fascinating to me.
Doesn't it just copy the reviews you trained it with?
No, there is no duplication. All the text is generated word by word. It may seem copied because the results are so plausible, but they are novel statements.
Why do I have to give my email and wait for results?
Because I'm cheap! This sort of thing is processing intensive and delivering realtime results would require buying expensive hosting and compute time. By using a process that takes much longer time, I can do this affordably. The most expensive word in human language is 'now'.
I want to make my own, will you help me?
No, sorry. But I did write an explainer of how I built this and there are tons of online resources for this now. The turning point for me was finding the Max Woolf simplified Python repository, gpt-2-simple.
Who are you?
My name is Joshua Curry and I make art under the name Lucidbeaming in San Jose, CA. Contact me at [email protected].
Is this artificial intelligence?
It's not really artificial intelligence, though tools like this get used as examples of AI in mainstream news articles. The interpreter has no idea what the prompt or training data mean. What it has is a billion examples of what words get used next to each other in typical sentences. It generates new text based on deep matrices of probability.
Why did you choose art reviews for training data?
I chose modern art reviews to train with because the language used is distinctive and unique to human expression. The descriptions of art include intent, emotion, technique, and impact. Art reviews can also veer off into heavy academic and esoteric jargon that is a challenge for traditionally trained language models.