How To Create Any Image From Text: The Wonders of AI in Computer-Generated Imagery
As a technology enthusiast taking a hiatus from writing on cryptocurrencies, I will indulge today in sharing an interesting discovery I made on the Internet (well, it is not exactly a ground-breaking piece of technology that just emerged, but still quite a novel one worthy of mention in my opinion!) But first, please enjoy a photo of a cute dog below.
Appreciate the texture, depth of field and quality of bokeh presented in this interesting shot — a piece of art presumably captured using a decent portrait lens on a DSLR. But what if I told you that this photo was generated by AI from scratch using purely word prompts, in under 2 minutes? Pretty mind-boggling, isn’t it?
To be sure that the AI did not just pluck the photo from the plethora of cute dog photo galleries available online, I did a reverse Google Image search, and was satisfied to find that the photo generated was indeed unique, as far as Google is concerned.
Introducing DALL.E
Dalle E is a “trained neural network” that can generate images merely from text descriptions. According to its site, Dalle E has a diverse set of capabilities, including creating “anthropomorphised versions of animals and objects, combining unrelated concepts in plausible ways, rendering text, and applying transformations to existing images”.
DALL·E is a 12-billion parameter version of GPT-3 trained to generate images from text descriptions, using a dataset of text–image pairs. We’ve found that it has a diverse set of capabilities, including creating anthropomorphized versions of animals and objects, combining unrelated concepts in plausible ways, rendering text, and applying transformations to existing images.
To trace its origins, the original DALL.E model was first developed by OpenAI in January 2021 to generate digital images from natural language descriptions. Its successor, DALL.E2, was announced in April 2022, with an improved ability to generate more realistic images at higher resolutions that can combine concepts, attributes and styles. However, still in its beta stage, DALL.E is not readily accessible to the public yet, as users have to join the waitlist before being able to try it out.
Plus, DALL.E is not free to use!
In this first phase of the beta, users can buy additional DALL·E credits in 115-credit increments (460 images) for $15 on top of their free monthly credits. One credit is applied each time a prompt is entered and a user hits “generate” or “variations.”
For those of us enthusiastic to relish this new piece of technology, Craiyon (formerly DALL-E Mini until a change of name was requested by OpenAI in June 2022) comes to the rescue! Released in 2022, Craiyon is an open-source community-created project that is based on the first version of DALL·E. Trained on unfiltered data from the Internet, Craiyon can generate whatever image you want the AI to produce by only using text captions, however random the scenario presented might be; the only limit is your imagination!
How To Generate Images From Text
Head over to the Craiyon site here or huggingface.co.
Enter a specific caption of the image that you intend to generate. From my limited experience, Craiyon worked best at generating landscapes/ inanimate objects.
Click run, and expect results in less than 2 minutes.
If the images generated are not up to expectations, press run again to generate a new batch of images.
Interesting Results
The quality of images generated range from visually pleasing and impressive to downright grotesque, especially if it involves complex imagery/ faces. However overall, the creative process makes for an interesting time-killer, what with the array of fascinating images generated.
Limitations
The AI presented in DALL.E is a fascinating piece of work that sees a huge potential, but currently still plagued with a few limitations.
Faces appear to be distorted, due to the limitations of the current image encoder.
Images generated have a low resolution.
Because the model was trained on unfiltered data from the Internet, it may generate images that reinforce certain stereotypes.
Final Thoughts
Indeed, the concept of ‘text-to-image’ is an impressive technology that would have been inconceivable just a few years back. Its applications know no bounds, especially for content creators.
What are your thoughts on this? Feel free to leave comments in the section below!
If you are interested to learn about generating passive income (which is what my blog is mainly about), feel free to follow me, and check out my previous articles and links below! Cheers!
🎁 Honeygain A passive income app to earn money off your unused internet bandwidth. Get $5 for free, no investment required.
🎁 Cake Defi A one-stop investment platform that bakes passive cashflow at APYs of up to 100%! Get a $50 bonus in DFI with a $50 deposit.
🎁 Nexo An advanced, regulated digital assets institution offering instant crypto loans, daily earning on assets with APYs of up to 36%, an exchange, with services in 40+ fiat currencies in more than 200 jurisdictions. Get a $25 bonus with a $100 deposit.
🎁 Hodlnaut A robust crypto lending and borrowing platform that generates passive cashflow from your idle cryptocurrencies with APYs of up to 9.4%. Get a $30 bonus in USDC with a $1000 deposit, or $50 with a $1500 deposit.
🎁 Finblox A fintech platform offering APYs of up to 90% on your cryptocurrencies! Get up to a $250 bonus in USDC with a minimum deposit of $100 (tiered).
🎁 BlockFi A cryptocurrency exchange and wallet. Get a $10 bonus with a $100 deposit.
🎁 Kucoin An expansive cryptocurrency exchange, with interesting offerings like staking, free trading bots and bitcoin cloud mining services.
🎁 Huobi A cryptocurrency exchange with diverse offerings, free airdrops and trading bots.
🎁 MEXC A cryptocurrency exchange with interesting listings and frequent airdrops from holding the MX token.
🎁 Crypto.com A cryptocurrency exchange based in Singapore. Get $25 in CRO on staking for a Ruby card.
🎁 Pionex A free multifunctional arbitrage trading bot that automates the process of buying low and selling high, 24/7.
For Malaysian investors
🎁 Luno Get a RM25 bonus in BTC with a RM100 purchase of BTC!
🎁 Stashaway Get free investing for 6 months!
🎁 Wahed code ‘KENLIE1’ RM10 signup bonus
🎁 Capbay P2P code ‘8879c6’ RM100 signup bonus
🎁 Versa Get a RM10 bonus with a RM100 deposit!
🎁 KDI Get a RM10 bonus with a RM250 deposit!
Connect with me Medium | Read.cash | Youtube | Twitter | Linktree
