We Let Machine Learning Design Logos
designwithai.com
designwithai.com
Why:
1. The logo icon is almost always from a service like The Noun Project. Probably fetched with 'category' of the company user enters. If it's not an icon then the first letter of the company name, why not?
2. The generated logo has the company name in different combinations of fonts, background and foreground colors. A simple algorithm with random value selection can do that. Where's AI in that?
Either AI is not being used at all or not to its full potential at least.
Why don't I see a logo for my product that has an icon like, say Airbnb's "A" or Apple's "Bitten Apple". All I get for my tech product are small mobiles and browser windows.
Now to your last point, we are working on something like that. We are training a deep learning architecture that creates logos from scratch, the stuff you are talking about. But we wanted to ship a simple yet decent product before we started working on more sophisticated stuff. In a couple of months, users will have option to select icons that are entirely generated by an AI engine.
I hope that clears some things. AI is being used in many places but there are also atleast 2-3 features that we're working on right now which will really exhibit the effectiveness of AI here.
Let me know if you wanna chat more about this. Happy to talk!
For updates on our product.
We are training a deep learning architecture that creates logos from scratch. But we wanted to ship a simple yet decent product before we started working on more sophisticated stuff. In a couple of months, users will have option to select icons that are entirely generated by an AI engine.
Doing this results in decreased revenue.
The term "AI" in our product comes from the fact that in long term, we actually want AI to create entire logos/icons/fonts and we're actively working on this right now. Most of our current models use Conv Nets, Word Embeddings, and Random Forests which still atleast come under the category of AI.
The reason I have been trying to explain everyone about the inner workings of our product is to demonstrate that we're indeed using different ML systems and they actually make sense where they're used.
For AI you’re supposed to give it a pencil and the computer teaches itself to draw.
It might work, eventually, for organizations with very simple needs - they need a logo just to seem like a real business. But notice how none of these logos communicate much.
I have to say, logos are not where I would expect AI to triumph in the next little while.
Logos work when there are subtle resonances between what humans know, how they perceive things, and the wealth of information humans accumulate over a lifetime of looking at things. Not to mention, how one uses all these factors to distinguish oneself from peers.
The history of computers has often proven pronouncements like mine wrong - but usually the computer is used to do something different rather than replace the subtlety a human can bring. I still find it hard to see where the space is between “I made us a logo in Word with Papyrus” and a proper design process.
What’s the disruptive underdog market here?
Maybe it’s the ancillary stuff. Like, you have a basic logo, now generate the files for all the other things; biz cards, stationery, online, .icos, etc etc etc
And we're not stopping here, no way! We are working on some features that will create a lot of stuff from scratch based on several deep learning architectures. We are excited to work on those features and we hope you'll like them once they are done.
Again, the goal is not to replace designers. The goal is to complement them or give founders a tool to get a decent logo in a few bucks. Happy to talk more.
Not sure if the result logos are intentionally blurry, but that definitely needs to be fixed. You will lose more business with the low quality results, than you will from people who make a logo and save the high quality version for free. People who are actually going to be willing to pay are going to be turned off by the bad result.
Thanks again for your feedback. If you happen to visit the website in a couple of days, you'll hopefully see much better logos in terms of clarity.
One feedback we got from another user was to reduce the number of logos we show. Asking people to select one from 50 logos seemed like a good idea but people struggle to select one from 50 from what I've observed. We'll still let everyone see 50 logos for a while until our ML models are so good that the top 5 logos we make are the ones you're gonna like.
I wish I had the time to get some icons in there and a better UI, but it's tough to compete with the awesome free logo generators out there, so I put it on hold.
I don’t really see where the ML plays into this. Cool concept though, I’d love to see it ask me if I like such and such and refine the process a bit more to land on something I actually like.
- The colors you see are generated by ML models, not us. If you don't like them, just hit the dislike button and our ML models will soon stop creating them.
- Icon selection based on the description is not an easy problem. You have to understand the semantics of what you're trying to say and match those with an icon's appearance. We're using word embeddings and auto encoders here.
- Initially, a genetic algorithm generates a bunch of stuff but that gets pruned by ML systems. As we get more feedback, that ranking/pruning is going to be more effective.
- Which fonts to combine together is not trivial. Convolutional neural networks come to the rescue here.
- Finally, one of the additional features we are working on is going to generate icons from scratch using GANs.
I hope this answers your question. Let me know if you wanna talk more about how we go about doing everything. Always happy to chat.
May I challenge you to explore these options?
- Not all logos needs colors, actually they should work in black and white as well.
- Your models don't seem to understand meaning or just to be fair, context. A name, tagline is probably too little information to infer a match. For my music tv app, I got 80% notes, keys and other music related icons. Not very imaginative. I usually write down / sketch my first 3 ideas and then throw them away since if it took me 15 minutes it would too for others.
- Why put always an icon there? An icon is a supplement, most successful logos just use a pretty (and expensive) typeface.
- How to do more imaginative typography? I type in The Brand Name and it the engine doesn't even try to change the case of the text.
I would expect to be imaginative, like:
THE..
BRAND
NAME.
Thanks again.
I'm not sure how/whether an AI-driven logo helps that process meaningfully. The execution of logo design is not the problem -- it's the concept which is the hard part. And that concept is a very human process.
Since the founders are in this thread: are you most passionate about design which led you to machine learning or are you passionate about machine learning and looked for a way to implement techniques? The engineering + ui/ux looks nifty, well done! As someone who has tried 99 designs, freelancers, and my own sketches for my projects' logos, my feedback is that this is a neat tool but I probably wouldn't find it useful were I ever to need to create a logo again.
Maybe a freelance web developer who has to create logos all the time for client projects would have a different perspective.
Edit: after looking through some more of the comments, if there was a feature that generated 100 different concepts of logos, let me share a link with other co-founders, and let us collaboratively tweak the results through rounds of feedback, THAT would be useful. As others commented, there are plenty of (free) 'design a logo' tools, but I don't know of any tools that 'help teams decide on a logo.' If the process helped us refine our concept, we would be so grateful for the value provided.
A couple of points. You are right when you say it's a long process but I believe you'll agree that it's not long for everybody. There are so many new companies that just want a decent and affordable logo as quickly as they can. Those are the people we are targeting now.
Regarding your comment about the logo concept - we are working on some deep learning architectures that generates the icon (concept) from scratch. That feature will take some time because of computational reasons but once that's in place, even startup founders who want to go a few rounds to get their logo concept would be able to use our website.
Let me know if you want to talk more about how we're gonna do it. Thanks again for your feedback, it's awesome!
Yet, most founding teams I've met end up spending a disproportionate amount of time (though not much money) discussing or changing their logo.
Also, startup founders usually only do this a few times. Web design consultants would do this more frequently.
- Our system figures out your needs by just looking at the description you provide. You don't have to select anything else to create very relevant logos to your brand. This is not the case with other websites where you have to provide plenty of information or at least do significant customizations. - Not one of these websites are generating icons from scratch. We will in a couple of weeks/months. This is going to be a very exciting feature that will set us apart. Let me know if you want to talk about this in more detail. - Our system learns as more people create logos. We have a feedback loop that continuous retrains the models to improve the logo creation process. In a few months, our logos will definitely be way better than free websites and other competitors.
Finally, at the end of the day, we still offer an alternative that we think has better features. If someone chooses to use a free website, by all means.
If I try to make a coffee shop logo, it won't design me a brand-new cup or bean graphic, as far as I can see, it'll just apply existing stock clipart?
edit: I ask because I genuinely can't tell if its generated on the fly or is existing human-designed vector art
This is by no means the final product that we want to have in a year or two. There was a reason we named the website "designwithai". We kept it generic because there are so many features that we're gonna add that are purely AI driven.
We can always talk more about the other features we're thinking about.
The assumption you are making, if I understand it, is that creativity needs to work like human creativity not just look like it. Similarly, design isn't design unless it works like human design, which involves sentience.
I'm just not sure that these assumptions are fair to make when we are discussing meaningful advances in artificial intelligence.
The ones on your home page look a lot less good to me. I'm not sure you've picked your best to show on the home page, I see the home page and think "eh, these logos are not that good, I probably don't want this."
Overall, I think the website design is very good. The barrier to entry is extremely low.
I dunno, you should probably get more feedback than two random people on the internet!
If you can share this - what kinds of organizations are you targeting? What do they care the most about?
I have a background in graphic design and computering, and I find it hard to envision the customer. They want a logo, want full control, don’t want to involve another human, have a price point below Fiverr? But maybe you’re serving people that I can’t see would need logo services.
I'd argue these little generators can produce better looking mathematically sound logos that are cheaper than dealing with a back and forth when you hire someone from upwork, 99, fiverr, your local college student.
Regarding mathematics and aesthetics: whenever you generate something with an algorithm, you'll inevitably give up some kind of variation. An algorithm only works within a subset of all possibilities. So I don't think logo generators can do everything that a designer can.
But for a lot of people it's the perfect option. You'll get a good logo for a friendly price.
Just from this post, over 500 logos have been liked/disliked by users which gives us a good amount of data to retrain our models for improvement and more fancy content.
Thanks for the feedback though. Really appreciate it.
https://www.designwithai.com/shareLogoSocial?logoHash=1422d7...
I'd use this for exploring and generating ideas. Which is what AI-design is good for, really.
We'll continue doing this until we are left with stuff that has not been used anywhere else in a logo.
https://www.designwithai.com/shareLogoSocial?logoHash=c82f33...
That was a fun diversion.
- What colors to choose for the user if he/she doesn't know the colors? - What icon to choose based on the text given? - What fonts to use together? - What font types to use (bold, italic etc) - Can we generate icons from scratch? - As we get user feedback, can we improve the logos?
And so many more areas where we can apply ML. Happy to talk at length about this if you want.
A bit part of logo design is ensuring that it can be trademarked, so this is pretty important.
- As soon as a logo is purchased with a unique icon, we delete that icon from our data. - We manually go through our icon repositories and try to find icons that have already been used in logos and delete them.
We'll continue doing this until we are left with stuff that has not been used anywhere else in a logo.
Keep up the effort. This looks promising.
Also product hunt is recommended a lot for submitting, there are tons of small startups checks product hunt daily.
More automation by AI - That's our main goal. If you create logos on Looka and Brandmark repetitively, you'll observe patterns. That happens because although Looka and Brandmark are applying AI, they are still getting a lot of information from the user before making logos i.e "select the logos you liked", "select the color schemes you like". We want to do away with all these things and only require the brand description. This cuts the time in less than half according to what we've measured. Below are some of the AI use cases we are working on. No other website is doing that right now.
AI Based Coloring - Our colors are not hand-picked by us. They are entirely generated by machine learning algorithms. We are also working on icon coloring using AI where the icon will have multiple colors entirely added by AI.
AI Based Icons - Almost every website including ours is using repositories of icons from websites like Flaticons, Nounproject etc. We are working on adding another layer where the icons are generated by the AI system itself.
AI Based Suggestions - We believe our in-browser customizer is relatively easier to use than others. We give the user an option to just select items that are suggested to him/her by our machine learning algorithms. We've got extremely positive feedback about this and are happy that people are liking it.
These are some of the areas where we're trying to be different than our competitors. We also offer more affordable prices than both Looka and Brandmark.
I don't particularly like charging for SVG and different sizes/colors, but it seems to be the standard for logo-making websites. Although I can turn a PNG into SVG in a few minutes with Inkscape/potrace, the target demographic probably has never heard of Inkscape and would pay $15 on fiverr or something for a quick Adobe job. So as usual the business is basically scamming ignorant fools.
Maybe an alternative is ads, I'd run an experiment for a few weeks. B2B ads pay absurdly high and hardware is cheap, so they should be able to pay hosting bills or whatever. And then you have a few years of runway to find a decent business model.
For higher tiers, charging for high usage seems reasonable, maybe there's an interest in a logo API.
I'd also look at cross-selling things like SEO, A/B testing, domain name registration, and general website design as part of a package, basically an affiliate program. https://www.pixelslogodesign.com/ looks reasonable although maybe too high-end since they have employees rather than freelancers. But charging $50 for ML logo design + $25 of fiverr labor seems reasonable.