Turn your crude drawings into AI masterpieces

Photo credits: piggledy, insanemilia

This is an advanced /remix tutorial. If you haven’t played with remix already, please start with part 1 to get the basic concepts down.

Warning, this gets a little NSFW unexpectedly

The AI enthusiast community is so ingenious sometimes that I want to scream for joy. Redditor Piggledy came up with an ingenious prompt that turns a simple pen doodle into a realistic photo, using the Realistic Vision 1.3 model (it comes preinstalled with Pirate Diffusion PRO), pictured above, using a random seed number.  But what if you didn’t know the right seed number and wanted to try this with your own doodle?  Let’s try.

ONE-TIME SETUP

First, we’re going to setup some reusable community styles, which work like little programs, sort of like an Instagram filter. Feel free to modify this to your own liking. It’s just text so anyone can do this, no programming required. After a person sets up a style, anyone in Pirate Diffusion can use it.

/style /new:unpaint $prompt (high detailed skin, 8k uhd, dslr, soft lighting, high quality, film grain, Fujifilm XT3, HD, Sharp)), [[deformed iris, deformed pupils, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime, text, close up, cropped, out of frame, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck]] [[canvas frame, cartoon, 3d, [[disfigured]], [[bad art]], [[deformed]],[[extra limbs]],[[close up]],[[b&w]], weird colors, blurry, [[[duplicate]]], [[morbid]], [[mutilated]], [out of frame], extra fingers, mutated hands, [[poorly drawn hands]], [[poorly drawn face]], [[[mutation]]], [[[deformed]]], [[ugly]], blurry, [[bad anatomy]], [[[bad proportions]]], [[extra limbs]], cloned face, [[[disfigured]]], out of frame, ugly, extra limbs, [bad anatomy], gross proportions, [malformed limbs], [[missing arms]], [[missing legs]], [[[extra arms]]], [[[extra legs]]], mutated hands, [fused fingers], [too many fingers], [[[long neck]]], Photoshop, video game, ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, mutation, mutated, extra limbs, extra legs, extra arms, disfigured, deformed, cross-eye, body out of frame, blurry, bad art, bad anatomy, 3d render]]

So I called my community style “unpaint” (I just made that up) and my future desired texts will go where the $prompt is shown above. Notice I added a bunch of extra words concerning realistic anatomy as ((positive tags)) and the words that describe crude drawings as a [[negative tags]], more commonly referred to as positive prompts.

While that is an insane amount of text, the point of this setup is that we never ever have to type that again. Styles are a one-time set it and forget it.

STEP 1:

Upload your fabulous crude MS Paint style drawing.  The figure below shows that my style was accepted, unfortunately credits me as its genius inventor (sorry Piggledy) and I’ve then uploaded their illustration.

STEP 2: UNLEASH THE REMIXES

To set expectations, it’s unlikely this will clean up in one take.  Take the time to observe what happens after each remix and iterate.

Right click or long touch the picture, click Reply (as if you were going to talk to the photo) and then type /remix with what you want your final image to be. Using some more positive and negative prompts, I describe the dream girl.

/remix (sexy) a realistic detailed modeling photograph of a ((((beautiful woman)))) with blonde hair wearing burgundy dress, standing in (yellow sand) desert, palm tree in the background, clear skies with scattered clouds, blue jeep in the foreground, (cracked soil), Nevada desert, hot weather, scorching sun, model photoshoot  <realvis13>

Well, then…

OK, totally didn’t nail it on my first try, but at least we have a human being and the basic scene set up. The white floor on the drawing is confusing the AI, and I’m losing the feet! Yikes!

STEP 3: MAKE SPOT EDITS

We can increase the guidance and make the floor a stronger part of the original prompt and go back to step one, or we can try /edit, which is a different command that’s useful for very specific spot fixes.

/edit what if the floor was desert sand

And yay, that worked well.  White floor, poof!  We have sand and shoes again. But that lady looks like a quick and dirty magazine collage cut-out, so let’s keep going.

Now that we have a semi-realistic photo to work with, I will repeat my original remix prompt verbatim.  We’re saying, forget the crude drawing, make the photo I wanted in the first place using this better reference material.

Also, I’m not feeling her conservative clothing, so let’s spice it up a bit with (sexy) and maybe we’ll get something that’s less of a shower curtain…

And we lost the feet again, but she got taller and who needs panties?

Well, let’s try that again.  Maybe the description of the desert can be reduced. Also her face looks weird and her legs are disappearing. Despair! Will this ever really work?

STEP 4: sWITCH TO A SIMPLER ANATOMY PROMPT IN YOUR REMIX NOW THAT YOU’RE ALMOST THERE

Let me stop calling the unpaint style and switch to /style:fuji, which is just a bunch of words that describe ideal anatomy, nothing fancy. We made up in the realistic AI characters tutorial.

Admittedly, I forgot to mention the tree, but that went so much better:

I love these, but the facial features are still a bit uncanny without that seed number, so let’s give her some identity, and ditch the sunglasses:

/remix a hot blue eyed, Swedish woman in the desert, smiling in focus, detailed face, next to a jeep, in the desert, (((realistic photo)))) <realvis13>

Success!  All it took was my sanity.

These are great pictures. Look how far we came! We lost the dress and the jeep somehow started turning into a futuristic police cruiser. Upgrades!

And we weren’t too tied to the initial drawing, changing it up was possible in a few different ways. I forgot to prompt for the tree, but I ain’t even mad.

STEP 5: GO CRAZY WITH THE SMALL DETAILS

From here on out, I can use /more to get different clothing and clouds, or go back to /remix get the facial expression I want (or remove add/more clothes) and /facelift to upscale it and smooth out imperfections.

Say your demanding art director was expecting someone younger, curvier, like a Texan beauty. At a flick of a wrist, it can happen. Hell, get 9 back at once.

Go try it yourself.

More community examples

Awesome art from insanemilia who “mostly finished off with <protonova>”

credit: Hampsteri

/remix a girl standing in a flower field , sun , birds in sky
by Anonymous_Baba

This is particularly useful for fixing unwanted glitches in images as well, or turning a simple paint drawing into a masterpiece.  Stack /remix with a /concept to hit the specific aesthetic you’re looking for.

Where to find the pen tool in Pirate Diffusion Telegram

Don’t see the pen tool?

It isn’t available in most versions of the desktop version of Telegram, in which case you can use your own tool like MS Paint or Photoshop and paste the image back in. Just don’t forget the compression step.

Image isn’t remixing?

The three most common reasons are (1) your bot isn’t administrator in the room that it’s in (2) the photo wasn’t uploaded in Telegram’s photo compression format, and (3) when those two things are fixed, Telegram can’t read your old messages prior to being an admin, so upload it again.  See detailed photos and steps on how to fix this in our troubleshooting guide

Go further 

You can also change the person’s poses by drawing the arms where you want.  Remember to use negative prompts for ideal anatomy as we we covered in the realism guide for best results.

The AI will interpret what you made will be unpredictable to some degree, so clearly drawing what you want and also clearly prompting with positive prompts goes a long way. Changing the model (concept) will also help if you are not getting the results you’re looking for.

Also, in PirateDiffusion, anytime you call a render, remix, or more command the /guidance and /strength are randomly assigned in the background.  You can increase or decrease those results manually for more control.

Render-Ahoy!