# This is not an AI Art Podcast (Ep. 8)
## Intro
Welcome to episode eight! This is your host, Doug Smith. This is Not An AI art podcast is a podcast about, well, AI ART – technology, community, and techniques. With a focus on stable diffusion, but all art tools are up for grabs, from the pencil on up, and including pay-to-play tools, like Midjourney. Less philosophy – more tire kicking. But if the philosophy gets in the way, we'll cover it.
But plenty of art theory!
Today we've got:
* Model madness model reviews:
* Bloods and crits:
* Technique of the week:
* My project update: so you can learn from my process
Available on:
* [Spotify](https://open.spotify.com/show/4RxBUvcx71dnOr1e1oYmvV)
* [iHeartRadio](https://www.iheart.com/podcast/269-this-is-not-an-ai-art-podc-112887791/)
* [Google Podcasts](https://podcasts.google.com/feed/aHR0cHM6Ly9hbmNob3IuZm0vcy9kZWY2YmQwOC9wb2RjYXN0L3Jzcw)
Show notes are always included and include all the visuals, prompts and technique examples, the format is intended to be so that you don't have to be looking at your screen -- but the show notes have all the imagery and prompts and details on the processes we look at.
## News
Interesting thing to keep an eye on, found [discussion on Reddit](https://www.reddit.com/r/StableDiffusion/comments/13yk093/new_textual_inversiondreambooth_killer_code_and/?utm_source=share&utm_medium=android_app&utm_name=androidcss&utm_term=1&utm_content=share_button)
ViCo, there's an [arxiv paper](https://arxiv.org/abs/2306.00971) and it's also on github [github.com/haoosz/vico](https://github.com/haoosz/vico)
Claims only 400 steps to train.
### Theme of the week: Go to your room, and make some fucking art.
Everything takes practice.
I see a lot of posts on the stable diffusion subreddit about how people are "so excited they can make high quality art now" -- which is great, I'm all for having better digital tools to make people more effective at producing output they desire.
However, none of this comes for free. Even when you can get a lucky generation or two out of your models.
So keep practicing -- everything.
* Your prompting
* Your research (more on this later)
* Your process
* the tech workflow
* the creative workflow
* Your drafting and manual ability
* Go to your room, and make some fucking art
## Model Madness
### Absolute Reality
Saw it on the [sd subreddit post](https://www.reddit.com/r/StableDiffusion/comments/13xnlq7/absolutereality_showcase/?utm_source=share&utm_medium=android_app&utm_name=androidcss&utm_term=1&utm_content=share_button)
On https://civitai.com/models/81458/absolutereality
This one grew on me after trying it a few times! I really like what I was getting out of it, the example prompts weren't helping me.
At first -- didn't really get a flapper out of it? Most other models know what one is.
```
(masterpiece), (extremely intricate:1.3), (realistic), portrait of a 1920s flapper, the most beautiful in the world, night club, film noire, neon lights, professional photograph of a stunning woman detailed, sharp focus, dramatic, award winning, cinematic lighting, octane render, unreal engine, volumetrics dtx
Negative prompt: BadDream, (UnrealisticDream:1.2)
Steps: 35, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 3020411638, Face restoration: CodeFormer, Size: 576x768, Model hash: 9483a64615, Model: absolutereality_v10, VAE: vae-ft-mse-840000-ema-pruned, Clip skip: 2
```
Let's try another subject...
```
(masterpiece), (extremely intricate:1.3), (realistic), portrait of a 1990s (raver:1.2) fashion model, professional photograph of a stunning woman detailed, sharp focus, dramatic, award winning, cinematic lighting, octane render, unreal engine, volumetrics dtx
Negative prompt: BadDream, (UnrealisticDream:1.2)
Steps: 35, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 2902100042, Face restoration: CodeFormer, Size: 576x768, Model hash: 9483a64615, Model: absolutereality_v10, VAE: vae-ft-mse-840000-ema-pruned, Clip skip: 2
```
Umm, I got the same thing almost!?
```
portrait of a 1990s (raver:1.2)
Negative prompt: BadDream, (UnrealisticDream:1.2)
Steps: 35, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 2264179666, Face restoration: CodeFormer, Size: 576x768, Model hash: 9483a64615, Model: absolutereality_v10, VAE: vae-ft-mse-840000-ema-pruned, Clip skip: 2
```
### FFusion Model (2.1)
Saw it on [sd subreddit](https://www.reddit.com/r/StableDiffusion/comments/13xx7yr/ffusion_ai_limited_public_beta_release_sd_21_512/?utm_source=share&utm_medium=android_app&utm_name=androidcss&utm_term=1&utm_content=share_button)
Worth watching! I'm getting some proportional problems on people, but, I'm pretty sure this is a common complaint about 2.1 models -- anatomy problems (due to at least speculated dataset problems)
This one is in beta.
```
a 1920s flapper in a smoky nightclub, great digital art with details, beautiful surreal portrait, photomanipulation, amazing digital art, concept art, smoke, detailed smoke, mind-bending digital art, gorgeous digital art by Rory Lewis
Negative prompt: low resolution,bad face quality,missing detail,over sharpening,(sharp image),(contrast:1.6),(oversaturated:2),(saturated:1.6),big contrast,contrast white burn,white spots,over exposed,over saturated,burned image,ugly,hideous,deformed,mutant,butchered,gore,sloppy,artifacts,mutilated,poorly drawn,poorly detailed,smudged,sketch,pencil,glossy skin,doll,plastic,lowres,poorly drawn,sloppy,over exposed,burnt image,broken,fuzzy,aliasing,cheap,old-school,poor quality,pixelated,sleepy,closed-eyes,poorly drawn,crippled,crooked,weird,odd,distorted,erased,cut,mutilated,sloppy,hideous,ugly,pixelated,text,logo,watermark,deformed eyes,close up,((disfigured)),((bad art)),((deformed)),((extra limbs)),(((duplicate))),((morbid)),out of frame,extra fingers,mutated hands,poorly drawn eyes,((poorly drawn hands)),((poorly drawn face)),(((mutation))),((ugly)),blurry,((bad anatomy)),(((bad proportions))
Steps: 35, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 912865668, Face restoration: CodeFormer, Size: 512x680, Model hash: 382df72d4a, Model: ffusionAI_512BetaBFGBuild0401, VAE: vae-ft-mse-840000-ema-pruned, Clip skip: 2
```
And a more "traditional" flapper
```
photograph of a 1920s flapper, raw photo, analog style, film grain, fujifilm xt3
Negative prompt: low resolution,bad face quality,missing detail,over sharpening,(sharp image),(contrast:1.6),(oversaturated:2),(saturated:1.6),big contrast,contrast white burn,white spots,over exposed,over saturated,burned image,ugly,hideous,deformed,mutant,butchered,gore,sloppy,artifacts,mutilated,poorly drawn,poorly detailed,smudged,sketch,pencil,glossy skin,doll,plastic,lowres,poorly drawn,sloppy,over exposed,burnt image,broken,fuzzy,aliasing,cheap,old-school,poor quality,pixelated,sleepy,closed-eyes,poorly drawn,crippled,crooked,weird,odd,distorted,erased,cut,mutilated,sloppy,hideous,ugly,pixelated,text,logo,watermark,deformed eyes,close up,((disfigured)),((bad art)),((deformed)),((extra limbs)),(((duplicate))),((morbid)),out of frame,extra fingers,mutated hands,poorly drawn eyes,((poorly drawn hands)),((poorly drawn face)),(((mutation))),((ugly)),blurry,((bad anatomy)),(((bad proportions))
Steps: 35, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 3871437286, Face restoration: CodeFormer, Size: 512x680, Model hash: 382df72d4a, Model: ffusionAI_512BetaBFGBuild0401, Clip skip: 2
```
## Bloods and crits
### Hunters between realms [on reddit](https://www.reddit.com/r/StableDiffusion/comments/13z6fd1/hunters_between_realms/?utm_source=share&utm_medium=android_app&utm_name=androidcss&utm_term=1&utm_content=share_button)
**NOTE:** I accidentally didn't realize that the series wasn't a series but a SET OF STEPS OF WORK IN PROGRESS. So I critiqued one that wasn't finished. Sorry! It progressively got better and there was lots of iteration. Excellent work.
Great process! Takes something "sort of boring" a photo of policeman waiting around outside at night, and it comes out.
The original input does have interest, and this is good quality input, and we get good quality output. "People at night waiting under a light" and then expanding on the narrative is REALLY working out.
I think it helps the composition greatly, and we start with a fairly interesting composition.
There's detail problems that need to be worked out, especially the weird powerlines, Some of which could be kind of adjusted to work out the depth too, like some could be lighter, some could be brighter. There's also a weird artifact line in the very middle of the frame that's totally detracting from the piece. The signage looks good in this one, but I don't read kanji, so, I'd probably take into consider what I want to do with the sign based on my goals and audience.
## The crystal princess [on Reddit](https://www.reddit.com/r/StableDiffusion/comments/13wfxvc/the_crystal_princess/?utm_source=share&utm_medium=android_app&utm_name=androidcss&utm_term=1&utm_content=share_button)
The rendering is great, and the author was kind enough to share the prompt. There's so much potential here that's lost, I think though.
The composition isn't terrible, but there's a lot to work with here, between the clothing, flowers and landscape, I think it could be pushed.
Also the repetition of form is almost there. Like the flowers and the diamonds, it's really cool, but it's a bummer to not have any of that in the background. You could also take something from the background and put it in the foreground. Like leaves on trees, and then add leaves to your flowers. It'd really bring the whole piece together.
Great prompt and output, though.
### Buffins the most beautiful cat [on Reddit](https://www.reddit.com/r/StableDiffusion/comments/13v4ntl/buffins_1958s_cat_with_the_most_appealing/?utm_source=share&utm_medium=android_app&utm_name=androidcss&utm_term=1&utm_content=share_button)
Really great output overall. Excellent kind of "restoration job". Excellent example of a stepping stone to use on your way to a final piece. Good workflow == good outcome.
It's also super entertaining, which... I can't deny.
I wouldn't really call this "a final piece" per se, meme value -- yes, complete. As a finish piece, not really. There almost has to be something more. A like change in composition or something else, otherwise, it's kind of a like... craft type of outcome. Excellent craftsmanship, but there's another layer to go to a final piece of art.
From this control net:
## Technique of the week: "Do your homework"
This weeks' technique is to "do your homework"
Just dig a layer deeper into whatever you're working on, and go look it up.
My mother-in-law was an english teacher, and she would say that to become a member of the look-it-up club, you must, quote-un-quote, "don't just guess, look it up"
So do that! Don't just fish around and guess about what to prompt. Sure, use prompt templates, and it's important to learn from other people's prompts, but... You can really take a it to the next level.
Here's an example, I was looking up some info about portrait photography, and I went to look at news about it, and I found [this great article about a portrait photographer, Rory Lewis](https://amateurphotographer.com/technique/portrait_photography/rory-lewis-transform-your-portrait-photography/) (Amateur Photographer Magazine), who's done portraiture of celebrities, politicians, etc.
But I learned a few things. I learned about some of Rory's inspiration behind his work, and I applied it to something I was working on.
One thing he brought up in particular is [Tenebrism](https://en.wikipedia.org/wiki/Tenebrism)
So I ran that through midjourney and then added it to my piece. Speaking of which! I run *every new thing* I learn about through MJ. I have google TV and I have art running on it when it's on and no TV is playing, and if the TV gets left on and run past it, I'll try running that artist name through MJ for kicks.
I also borrowed a bunch of cool info from the article about what film was used, what kind of lighting was used. All helpful prompt terminology.
Here's a piece before I added a concept I learned from that article:
### Before
### After
Look at the dramatic lighting we get! That portrait photographer was on-to something.
### The final piece from that
Granted -- the thing is, we iterate from here.
Let's talk about why I selected it -- I liked the dark moody composition, I liked that it was already composed in black and white. And I *loved* the repetition of form in the windows. The strong contrast creates negative space and it adds to the composition.
Let's also talk about what I changed on it.
A bunch of hand problems, there's kinda "no right-most hand" so I added that back in, I also fixed the pinky on the guy on the right. The hand on the guy on the left isn't "bad" per se, but it's not doing anything to add to the piece.
I got the idea to add the pipe when trying to figure out what to do with the hand. Like maybe a hand holding a pipe. But I liked the idea of the pipe because it plays into the smoky train cabin, a kind of part-to-whole relationship.
The face on the right is a little "waxy" to me. And I'd almost like to fix it, but I don't think it's "that big of a deal" at least not for what I'm using it for now.
## Project update
So -- my success with LoRA's at first was guess what? Pure luck.
The more I do, the worse I seem to be getting.
Guess what? I need practice. I also need to create more scripts to automatically run x/y plots and run more variations.
I'm getting this weird thing where I up the "network alpha" and then my LoRA's become super inflexible? I'm clearly doing something wrong.
I tried DAdaption, but didn't have great results from it.
Few articles to pass along:
* https://rentry.org/59xed3
* Extremely detailed, author seems knowledgable
* https://www.kix.in/2023/04/07/sd-lora-finetuning/
* Author seems newer to LoRA's but the write up is amazing and worth a reference read.
Sign in with Wallet
Connect another wallet