With OpenAI’s current announcement of the brand new ChatGPT interface, I knew precisely what I wished to attempt first. Not like Midjourney, DALL-E couldn’t actually settle for picture prompts up to now. It was strictly textual content to picture.
Once I realized it is now doable, I had so many concepts run via my head. What can you employ picture -> textual content -> picture to do?
So, I opted to start out easy: Creating digital avatars of myself.
On this article, I am going to information you thru the easy steps to create your personal customized avatar utilizing ChatGPT’s language understanding and DALL-E’s creativity, in addition to present totally different examples to point out you the way good it really is.
Only a fast warning: You’re going to see a whole lot of my face on this article. So, undoubtedly keep tuned for that.
How To Make Customized Avatars with ChatGPT & DALL-E 3
It’s fairly straightforward to make customized avatars utilizing ChatGPT and, as promised, it will solely take two minutes.
First, it’s a must to allow GPT-4 by urgent the mannequin drawer on the highest left facet of the display.
Subsequent, press the paperclip emblem on the immediate bar to add your reference picture.
And, lastly, merely say “Create an avatar of me.” You’ll be able to add no matter you need, so long as the “avatar” key phrase is there.
Right here’s the ultimate product in comparison with my authentic picture:
I do know what you’re pondering. It doesn’t look fairly like me, proper? So, why does that occur?
ChatGPT doesn’t straight use your picture as a reference for DALL-E 3. As a substitute, there’s a intermediary between the 2: GPT-4V, which takes your authentic picture, turns it into textual content, after which generates a immediate primarily based on the textual content. In reality, you possibly can even see what GPT-4V thinks of my authentic picture after I examine the avatar:
What GPT-4V does is choose probably the most defining options of the unique picture and makes use of that to create the avatar. For me, it was my black hair, white shirt, and nice expression. And, positive sufficient, it resulted in an inaccurate depiction of me.
So, might I make this higher?
I additionally tried creating a greater avatar earlier than by offering extra context to GPT-4 comparable to my ethnicity, higher hair description, and different options that weren’t distinguished within the reference. Right here’s what it seems like:
One other factor I did was discover DALL-E’s creativity by asking it to create totally different variations of my avatar utilizing totally different kinds like a extra reasonable 2D illustration, a 3D render, doodle, and extra.
7 Different Examples Utilizing Well-known Individuals
To higher showcase how good DALL-E 3 is at creating avatars, listed here are seven avatars I made utilizing ChatGPT of well-known individuals from totally different ethnicities, every characterised by distinctive facial options.
Since GPT-4V extracts the defining options of an individual, this technique works so much higher for individuals with distinctive traits. Out of all these individuals, I’d say that — if I’m being actually beneficiant — solely 3 out of seven are recognizable with these avatars.
It’s like a sport of phone, the place I give ChatGPT a picture, GPT-4V passes it on, till it reaches DALL-E 3. It’s solely pure that there’s some components which might be misplaced within the combine.
So, for those who’re on the lookout for an correct avatar creator, I counsel trying on-line for image-to-image editors as an alternative. This will work, but it surely actually relies on the way you look.
A Fast Comparability In opposition to Midjourney
I wished to see if DALL-E 3 was really higher than Midjourney at creating avatars because it really permits image-to-image tweaking. So, I used my picture earlier and right here’s a comparability of them.
And yep… I’ll persist with DALL-E.
Not solely does Midjourney’s output look nothing like me, it’s additionally too stylized for my liking. Unusual.
Would I Advocate It?
No can be too harsh — so, I’d say, not but.
So long as ChatGPT can’t settle for photos and straight use them as enter for DALL-E 3, this wouldn’t work persistently. In case you’re on the lookout for avatars that you simply’ll really use, like I stated earlier, it’s higher to spend money on image-to-image editors on-line.
That stated, I do imagine that this supplies some good perception into how the brand new ChatGPT atmosphere works. In my quick expertise, I discovered that this new interface is much more streamlined and allowed me to complete my duties extra effectively.
As for efficiency, nicely, I’ll be trustworthy and say that I’ve had many retries earlier than I received what I wanted and never an error — however that’s to be anticipated as a result of quantity of individuals attempting it out.