## Intro
Welcome to episode seventeen! 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: 2 SDXL models and a teaser
* Bloods and crits: 2 pieces
* Technique of the week: "Make it a real location"
* Updates on my project:
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
### Coca-cola has an ad campaign that features SD
* [on reddit](https://reddit.com/r/StableDiffusion/s/ay18sia9gz)
This is.... kind of amazing. We're hitting the mainstream even more now. I mean -- Coca-cola is one of the greats of marketing -- they're literally selling sugar water, right.
### StableAudio
https://stableaudio.com/generate
Not visuals, but!!! This is super fun, and I love music production, too.
I can't wait to sample from some of these. Getting some results that are surprisingly "not bad"
### Collage Diffusion
* [on Reddit](https://reddit.com/r/StableDiffusion/s/ce03ijm0bP)
* [On Github](https://github.com/linden-li/collage-diffusion-ui)
* Try it out: https://collagediffusion.stanford.edu/
Kind of an idea about regional prompting with a UI, I guess.
Honestly, it's just a PoC and it's not particularly useful the way it is now. But the idea behind it has something. Also the demo was busted when I went to try it.
But it's interesting to me because this is something I'm actively doing as a workflow which is to, instead of trying to prompt for EVERYTHING (pro tip: don't do that) is to mix and match and use the tools to my advantage, so something I'm doing is...
* Generating multiple generations on multiple subjects
* Say, a dog, a cat, and a veterinarian. Instead of prompting for them all
* Combining subjects as a photobash, or map bash.
* img2img + inpaint
* Likely with overpainting and/or more photobashing.
* Post processing otherwise.
Example of my process...
Gist is that I generated things separately, in MJ:
* The dude
* The helmet
* The lantern
Then I put the two together with photobashing in SD with inpainting.
### Free-tier Google collab banning webuis
* [Sebastian Kampf YT video](https://discord.com/channels/1074808760704454758/1096047579290152991/1151155029189873756)
If you're not familiar, Google collab is a cloud service for running your AI/ML workloads with Google's cloud GPUs, so basically... GPU rental in the cloud, certainly with Google UX enhancements and community.
I surfed reddit a bit, and it's... honestly not that unreasonable, and apparently isn't a big deal for paid tier. However, it's a bummer if it was a place that you wanted to try it out.
## What's better? MJ or SD
Discussion [on reddit](https://reddit.com/r/StableDiffusion/s/mzNfMTjl3Q)
The answer is: Both are better.
MJ has "automagic" -- you don't have to prompt as much to get good results, and MJ seems to good at producing "final pieces".
But pro tip: They're probably not final. Don't settle, get what you envision.
By the same token, SD doesn't, so that means you get more prompt control.
MJ does have inpainting, and it's REALLY grown on me since I covered it. I use it frequently. But the way you inpaint in SD has *way more control*
Also, control net.
## Model Madness
### Realistic Freedom
Boasting that it does sfw and nsfw equally.
* [reddit](https://www.reddit.com/r/StableDiffusion/comments/1686e2n/realistic_freedom_sfw_and_nsfw_is_available_on/?share_id=QEbR4qd7jnzw59Opg7H4w&utm_content=1&utm_medium=android_app&utm_name=androidcss&utm_source=share&utm_term=1)
* [civitai](https://civitai.com/models/138977)
```
photo of a cute flapper from the 1920s, jewelry, mafia, speakeasy, nightclub, fashion photography, film noire, cinematic
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 3691140888, Size: 1024x1024, Model hash: dce7eb8449, Model: realisticFreedomSFW_alpha, Version: v1.5.1
```
Oh snap, this came out awesome...
```
photo of a cute goth cyberpunk raver girl, warehouse, dj, psychedelic lights, night, neon lights, nerd outfit, cinematic, cyberpunk, dj booth, dj equipment, cluttered with cables
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 2993027600, Size: 1024x1024, Model hash: dce7eb8449, Model: realisticFreedomSFW_alpha, Version: v1.5.1
```
### Realistic Stock Photo
* [reddit](https://www.reddit.com/r/StableDiffusion/comments/169y829/this_seems_to_be_the_most_midjorneystyle_sdxl/?share_id=Xp-AednF3Yvu6yYVfZMxx&utm_content=1&utm_medium=android_app&utm_name=androidcss&utm_source=share&utm_term=1)
* [civitai](https://civitai.com/models/139565?modelVersionId=154593)
```
color photograph close up portrait of a 1920s flapper, cinematic 4k epic detailed 4k epic detailed photograph shot on kodak detailed bokeh cinematic hbo dark moody
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 3421025465, Size: 1024x1024, Model hash: 2d44ce378d, Model: realisticStockPhoto_v10, Version: v1.5.1
```
```
color photograph close up portrait of a 1990s raver, cinematic 4k epic detailed 4k epic detailed photograph shot on kodak detailed bokeh cinematic hbo dark moody
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 4164232631, Size: 1024x1024, Model hash: 2d44ce378d, Model: realisticStockPhoto_v10, Version: v1.5.1
```
```
color photograph close up portrait of a goth raver girl, cinematic 4k epic detailed 4k epic detailed photograph shot on kodak detailed bokeh cinematic hbo dark moody
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 3713092171, Size: 1024x1024, Model hash: 2d44ce378d, Model: realisticStockPhoto_v10, Version: v1.5.1
```
### Realistic Vision XL
"Coming soon" -- you can't download it yet.
* [reddit](https://www.reddit.com/r/StableDiffusion/comments/16aw7o5/cinematic_created_with_the_new_realistic_vision/?share_id=ww_B8YSpZ-K9PNX5FkSfI&utm_content=1&utm_medium=android_app&utm_name=androidcss&utm_source=share&utm_term=1)
* [on mage.space](https://www.mage.space/u/RealVisXL)
### Other resources!
* [Aipreneur on youtube, training style LoRAs for SDXL](https://youtu.be/E2pI_YyoQjA?si=S0bk38vyf-Rzw8CS)
## Bloods and Crits
### An early start to fall
* [On reddit](https://www.reddit.com/r/SDLandscapes/comments/16dl1ve/an_early_start_to_fall/?share_id=ThxOcxcFZHAlKCC5riicz&utm_content=1&utm_medium=android_app&utm_name=androidcss&utm_source=share&utm_term=1)
Overall, I just highly enjoy this piece. I really like how the oil paint type of look really comes through, and you can almost see the brush build up in some of the places.
The composition is nice, and while "the tree" is the primary subject and is all within the frame, I still think its working.
Interesting that they chose a portrait orientation. I think it's OK here, it works with the tree as... "a portrait"
I think the green sky in the upper right is actively detracting from the piece. It doesn't look similar to other greens used in the piece, and it doesn't screen "sky" to me. Even though that could be done if it was more like... consistent and intentional. I think that should be fixed. I'd probably reference some of the blues in the mountain scape.
### DJs dropping beats through history
* [on reddit](https://www.reddit.com/r/midjourney/comments/16hdrjn/djs_dropping_beats_throughout_history/?share_id=T3Tr40VX8VeAgym_-VZ1K&utm_content=1&utm_medium=android_app&utm_name=androidcss&utm_source=share&utm_term=1)
Concept is AWESOME. I'm also the right audience, I love DJing, EDM, music production, so I'm a good audience for this.
But what I like is that it's a concept mashup that's NOT a thin-ass concept. Such as "mixing an existing character with $any-variable-thing" -- like LoTR or cartoon characters. Sure, I get it if you do fan art, and good on you -- but I think these concepts are... Lame. I like new things. This creates something new, but has concepts people connect to.
This one is the last in the series, and imo, the best was saved for last.
This comes from a series, and mostly I'd say they're not finished pieces. They don't appear to be post processed, and most of them can use it.
In this case, the left hand has problems. Looks like a missing thumb + additional knuckle, and there's a weird like "glint" on the hand that draws your eye there which shouldn't be the case.
Some of the background objects (weird stuff on the walls) could be changed to look like something "that's supposed to be there" -- as well as some of the DJ equipment doesn't look like a thing but "an impression of a thing" and it's not working perfectly here.
I gave it a shot myself, and had a bunch of fun. (These are not post processed, just raw generations.) I'm unsure they fit the "mixed theme" exactly, but they came out really fun with barely any prompting.
```
Renaissance raver
```
```
Renaissance raver woman, astral planetarium
```
## Technique of the week: Make it a real location
Ok so here's the original generation I started with, from Midjourney
The first step I take is interrogating that, then I fix up the prompt language a bit to get more specific.
And then I wanted to make the location an actual one, so I based it on this photo:
I cut out what I wanted, overlaid it on the image. I then adjusted bright/contrast and levels to get it to match the image generally.
The next step is to inpaint this area -- but I did it with a really low denoise so that I keep the location almost the same, but make it 1. fit the image better, and 2. be recognizable (but not too obvious! I want it as a reward and hidden detail for those that know the location -- but to look good for anyone else viewing it).
So I used a denoise in the like `0.20` range.
From there, I wind up overpainting it to fix up what I want, remove the trees on the right, patch the missing chunk in the left hand corner, I manipulated the backpack, changed out the canteen, and then inpainted "all the things"
And I wound up with:
and another one
original from MJ:
the location, from wikipedia...
And then I traced the mountain profile and inpainted... (along with other edits...)
Sign in with Wallet
Connect another wallet