Hacker News new | comments | show | ask | jobs | submit login
New paint colors invented by neural network (lewisandquark.tumblr.com)
165 points by CPLX 189 days ago | hide | past | web | favorite | 54 comments

I want the algorithm for generating server names: Bunflow, Snowbonk, Stanky Bean, Dorkwood, Bank Butt.. these are great!


Not a neutral net, but adaptable to what you want to do.

I'm a bit skeptical of this whole report.

This is hilarious. But I don't understand how the network learned the words and color associations for "turd", "poop", "stanky" and "butt", since presumably these were not words Sherwin-Williams used for their own colors and thus would not be present in the training set all. What am I missing?

Probably bias on the author's side where he would select those words over the nonsense outputs or the unfunny ones.

The names are done on a letter basis rather than a word basis judging from some of the outputs.

Even still, the algorithm has no knowledge of these words. The likelihood of producing them is infinitesimal. Why would it produce real words more often than gibberish, or words in every other language on earth? I don't think this argument holds.

If the algorithm is trained on English words, it will learn English phonotactics (i.e., syllable structure, what sounds are allowed at the beginning of words, etc). That makes it much more likely to generate valid words, as the set of strings that follow English phonotactic rules << the set of gibberish strings.

Agreed, but I can't imagine with a sample space of just 7700 paint names that it would generate turdly any more frequently than bvlogd or a gazillion other marginally pronounceable six letter strings, without some additional bias not mentioned in the article.

If I were the author I would train another NN that compares the color with a huge set of photo+definition and decide from there. For instance "Coral Gray" looks quite accurate when compared with the average recreational diver photo of coral reefs (but not the real coral specimens, which tend to be very colorful when correctly photographed).

And there I thought someone came up with a way to mfg "impossible" colors[1] like reddish green.


HSB or HSV would be a better training set for it to discover human color associations than RGB is.

One of the perceptual spaces would probably be best, something like CIELAB. Make the gradient of perceptual change wrt change in the space the same for the entire space and you might get more generalizability.

A good NN should have no trouble modelling a simple perceptual transform like that on its own.

This is terrific. AI is going to take so much stress out of naming our next child.

That's what I thought, but my wife is not allowing me to name our son Wintermute.

Some countries would deny you that choice even if you were both OK with it.

e.g. http://mentalfloss.com/article/25034/8-countries-fascinating...

I am from Denmark; I thought this was much more common in the world. Definitely way more than being part of the eight countries to have these strict laws apparently.

Always thought it was strange you could call your child McBurger if you wanted or apparently Number 16 Bus Shelter.

I know one who got Jazz rejected even though people still plus twenty years later call her Jazz.

There are some limitations in Italy too, though a little less restrictive than those in Denmark or Germany, JFYI, a rough English translation:

Art. 34 (Name Attribution Limits)

    1. It is forbidden to impose on the child the same name as the living father, of a living brother or sister, a surname as a name, names ridiculous or shameful.

    2. Alien names that are imposed on children who have the Italian citizenship must be expressed in letters of the Italian alphabet , with the extension to the letters: J, K, X, Y, W and where possible, using the diacritic marks proper to the alphabet of the name source language.

    3. Childrens of unknown parents may not have names or surnames that may represent a reference to their natural origin, or surnames of historical importance or belonging to families particularly well known in the place where the birth act is formed.
I believe that some limitations do exist in all or most countries, the mentioned article just enumerates 8 of them.

I suspect it is more common, that was just the first google-result which was somewhat interesting.

I'm in Finland and there are official name-lists here too, as I found out to my surprise when it came to naming our child.

Please join us in welcoming little Bank Butt to the world!

...assuming you can get your hands on a large set of pictures of babies along with their names, to train your NN on.

It would be nicer to have the parents pictures instead. This way you can name the baby before birth.

Who wouldn't want to paint their walls Turdly brown?

It was popular in the late 70s in Northern Ireland https://en.wikipedia.org/wiki/Dirty_protest

I'd go for Homestar Brown with Dondarf highlights, myself.

I personally prefer Stanky Bean

...and Bank Butt might just be an entirely accurate name for that color.

Or Gray Pubic?

I find the title of this entry highly misleading. I clicked through expecting to see new colors (even though I knew that was going to be impossible). What I saw was weird color names.

These click bait AI articles with almost no content are becoming more common on YC. The fact that the author made no mention of seeding the results is a lie through omission. A lot of the hype machine around so-called "neural networks" has to do with the language used, that is so far out of touch with the reality of the work: "training", "learning", "creativity", "inventing." These are glorified Markov chains.

I found it to be quite the opposite. The title is simple and matter of fact, not clickbait, but the actual article is genuinely hilarious.

> New paint colors invented by neural network

These aren't new colors, they are new color names. And they weren't invented by a neural network, the names were generated with seed words like "poop" or "stanky", and a language processing algorithm matched the seed words based on certain criteria and statistical models of a 7500-word database. There is definitely an element of "invention" in the sense of "making up (an idea, name, story, etc.), especially so as to deceive." The deception here is that a simple language processing algorithm is capable of "invention."

"What color is this room?"


"Ah .. " backs away slowly

Have you ever read the color names on the chips at a home improvement store? Some of them are quite weird.

My favorite is 219, 209, 179 ingeniously called Dope

That one was pretty accurate as it approximates the color of dried aircraft dope on aircraft fabric.

It is tough to single one out with such excellent names; "Stanky Bean", "Woleebaph Ronder Wily", "Clardic Fog"...

I will never see #c5a2ab again and not think, "Stanky Bean." In fact, it looks like stankybean.com isn't registered. I'm imagining just having it be a solid wall of #c5a2ab.


It's possible to generate new colors using Multi-layer Recurrent Neural Network (RNN, LSTM, and GRU) based on Sherwin-Williams paint colors along with their RGB values (RGB = red, green, and blue color values)

I'm partial to Dorkwood.

If you're going to use this approach, you need to have it also train on images from a large dataset like something from shutterstock or google imagesearch.

A fun project!

The recipes entry further down the page is pretty impressive too.


1. I want to paint my room Caring Tan all over.

2. At last, stoners can have their own blue.

3. Turdly is apt.

"Homestar Brown" and "Pubic Gray" got me.

> 7,700 Sherwin-Williams paint colors

The xkcd color survey is a much bigger data set: https://blog.xkcd.com/2010/05/03/color-survey-results/

I use it a lot to color graphs in Python / Seaborn [0]. I love to combine Pale Red with Denim Blue and Medium Green.

[0] http://seaborn.pydata.org/tutorial/color_palettes.html#using...

Aren't most of those names Homestar Runner characters?

Stoomy brown, Gray Pubic, Dope all made me lol

There is also Power Gray which sounds really cool

Okay, I'll ask the question:

I stumbled on https://gist.github.com/nylki/1efbaa36635956d35bcc recently. This is rather random and chaotic; some parts are interesting, other parts not so much.

How could I build a neural network that surfaced information in the style that I've prepared below?

The following is taken from the above URL, rearranged somewhat for a certain effect.


 Categories: Chinese, Appetizers, Poultry, German, Casseroles, Diabetic, Main dish, Seafood
      Yield: 25 Salad

      2 c  Salad oil
      2 c  Chicken broth
           Steamed, Biscuits
     16    Boull of hot soy sauce
      1 ds Pepper sauce
      1 tb Creme de carrots
      1 tb Fat flour
      1 tb Cream, divided
      6 oz Published cheese; cut
      1 ea Stick semoves and pepper sauce
      2 tb Minced fresh celery
      2 tb Sour cream or flour
      8 oz Semisweet chocolate
    1/3 c  Flour; for frying
      1 tb Hot water; or or salt
      1 ts Gelatin shortening
      2 tb Butter, ground
      1 c  Crushed bananas
      1 ts Poppy strip powder
      2 tb Sesame seeds
      1 tb Parsley, dried
      1 ts Vanilla extract
    2/3 c  Lite red pepper sauce
      1 ts Ground cinnamon
    1/8 ts Pepper
           Brown sugar
Preheat oven to 350. Top with parsley.

In a lightly floured bowl, beat the egg/unsalted water 2 minutes. Remove base over chicken in bowl. Add meat in foil. Sprinkle each zucchini; add cheese, salt, flour, salt and butter.

Fold each balls of pan, reliated with glaze. Cook until crispy on a plate. Allow feed in baking sheets. Combine flour, baking powder, and baking soda and corn syrup.

Sprinkle with green onions and salt. Preheat oven to 375F. To cool completely onto prepared cheesecloth. Cover and simmer for 30 minutes, until all ingredients are done.

Let them from the heat and spread with a double boiler over high for 10 minutes. Combine sugar and olive oil and oregano. Remove chops.

For a couple of dish, pineapple pieces of leek and freeze. Prepare chops to serve.

Keep is some mayonnaise to cool spread on a baking sheet, place in a bowl, beat egg yolks, paprika, carrot, celery, thyme, onions, salt, pepper, oregano, and gradually until softened. Combine lemon juice and mozzarella cheese and mayonnaise in a bowl. Add baking chocolate, blending well. Make a little and cooked or all to the cherry and ice cream, and slice the crumbs, and fresh sauce.

In a large bowl, beat eggs and cook for 3 hours. Wash pork, freeze up, but not dough back to diet into a layer. YOWL THE COOKIE: ADD 1/2 cup of cheese. Stir in the carrots, remaining ingredients together, stirring constantly.

Place the oil in a large skillet over medium heat until chicken is boiled, to the center of the egg mixture. Pour over and roll it for an about 3-4 pounds and can be stored is changes have for sized onion from page in with pan with canned extratty fish on a glass or rounding pan.

Cook it with the batter. Set aside to cool. Remove the peanut oil in a small saucepan and pour into the margarine until they are soft. Stir in a mixer (dough). Add the chestnuts, beaten egg whites, oil, and salt and brown sugar and sugar; stir onto the boqtly brown it.

In large bowl, combine liquid, and add bay leaf.

Microwave each cake; boil 4 hours, until melted too mush of skillet in foil; add remaining 1/4 cup of cheese.

Stir together flour, baking powder and leaves to a platter.

Sprinkle with basil. Place fish filling into a large bowl; lemon juice only fill stems of gravy. Serve warm or serving colleed. Bake in a skillet into a heavy saucepan.

Cover with lemon weed. Stir and saute for about 30 minutes. Remove from heat; fluffy. Drain on both sides of the refrigerator.

Grind sauce about 1 1/2 inches in cold water combine with chopped pecans.

Cook skillet in a bowl until stiff bowl and stir often on a colored and bake for 8 to 10 minutes or until cheese is evaporated. Peel peppercorns and parmesan cheese, or all flour, 3" pieces, adding the weights: heat over all sweetened corn.

Break off doors along with tomatoes. Blend the beans and seasonings. Cook, uncovered, over medium heat, stirring occasionally, for 20 minutes, until the figs is absorbed.

Store in a high flavors of the meat mixture, and cover with a size of the egg mixture. Stir the beef into the center cups.

Bake at high heat and stir over soup and replace 3 tablespoons of the bowl. Combine cheese and buttermilk; cook over low heat, cook a microwave, 20-30 minutes.

Sprinkle the beans, then stirring frequently. When the meat is dissolved. Store to medium-side; then dice the fine side.

Sprinkle with milk. Gradually remove from heat and simmer for about 20 minutes. Mix vegetables, and add the cornmeal and the celery, and mix well. Let cool.

Beat Water and boil until light brown.

Cover and cook over low heat 30 minutes. Remove seconds. Place on a lightly floured board in a bowl.

Serve the sugar canned cheese with salt and pepper and tomatoes, and remove margarine, vanilla and bacon and sprinkle with milk and cook to the bag.

Let cool for at least 4 minutes.

NOTE: Cover with delicious dice them bread for about 30 minutes.

Source: Stuff Cooking by Chocolate Candy by Sunsafe by The Collection of Cookbook by Pet Farnes Weeks, in Your Rodled, Elmeasian. -- (You have) over the same and roll the edible of the pan of soup.

From Electric this unpeelerset Beans Kinch Bar "Suet Brightor, Nanc" and Information To you by Depth Chefs copyrighted by ISBN 0-90828-2 Microwave Optional Collection of Shellies


(YES, it is still random, but I tried to extract the most amusing parts. The full URL doesn't really do that.)

Bank Butt.

Thats hilarious. I loled.

Sucks you're getting downvoted. This was the hardest I've laughed in months.

The machines have surpassed us. No human could come up with paint names this good.

Of all the colors to follow "Homestar" I didn't expect Brown.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | DMCA | Apply to YC | Contact