Guide to Image Generation. Part 1
Part 2 here
Part 3 Here
My Links: twitter, discord, IG
Subscribe for FBB images @ https://patreon.com/fitCorderAI
We'll begin with simple subjects, using some of the most popular models that have become favorites within our community. These models offer a solid foundation for understanding the mechanics of AI artistry and provide a springboard for more complex explorations. Once we've mastered these basics, I'll be sharing some of the unique methods I employ with my personal models.
This article assumes you know how to install an embedding. (Guide coming). It also links to my guides for VAE & HiRes. It also assumes you know how to install a checkpoint. If you need further help please reach out using the links above.
I will be using the same seed / generation settings for this whole series.
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
The vae is mse-840k unless specified. Find it here.
We will start with Lykon's models as they are fantastic and very flexible. I will go over my own models later down the article.
The value of negative embeddings.
Here I will show you a basic prompt with & without a negative embedding on anime.
Anime Model: Anylora
Positive Prompt: woman, dress
Negative Prompt:
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
Anime Model: Anylora
Positive Prompt: woman, dress
Negative Prompt: BadDream
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
Here I will show you a basic prompt with & without a negative embedding on realism.
Realism Model: AbsoluteReality
Positive Prompt: woman, dress
Negative Prompt:
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
Realism Model: AbsoluteReality
Positive Prompt: woman, dress
Negative Prompt: BadDream
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
I hope this convinces you of the value of negative embeddings such as BadDream. You can find it here. I will continue using Baddream from this point.
The value of positive embeddings.
Here I will show you the same basic prompt and adding a positive embedding on anime. This one is for portraits specifically.
Anime Model: Anylora
Positive Prompt: woman, dress, fcPortrait
Negative Prompt: BadDream
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
Here I will show you the same basic prompt and adding a positive embedding on realism. This one is for portraits specifically.
Realism Model: AbsoluteReality
Positive Prompt: woman, dress, fcPortrait
Negative Prompt: BadDream
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
I hope this convinces you of the value of positive embeddings such as fcPortrait. You can find it here.
You may also achieve great effects through the right prompts. Which iin my case is available on every image I upload through the metadata.
The value of Hi-res. fix
This gets into the territory where some of you not have the vram available for this. If that is the case I may point you in the direction of Ultimate SD Upscale. You may find the guide on how to use Hi-res. fix here. Note: I used 512x512 for these instead of my usual 512x768.
Anime Model: Anylora
Positive Prompt: woman, dress, fcPortrait
Negative Prompt: BadDream
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
Realism Model: AbsoluteReality
Positive Prompt: woman, dress, fcPortrait
Negative Prompt: BadDream
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
Hi-res. Fix allows you to go past 512x512 with great results. I will be using Hi-Res Fix from here on out.
The value of VAE
Most models do not come with a "baked vae". If that is the case for you it is very easy to setup and allows you more control over your image. Read more Here.
None of my Models come with a baked VAE so I'm switching to them for this point forwards.
Here I will show you the same basic prompt before and after adding a VAE on realism.
Realism Model: fCBlendMix
Positive Prompt: woman, dress, fcPortrait
Negative Prompt: BadDream
VAE:NONE
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
Realism Model: fCBlendMix
Positive Prompt: woman, dress, fcPortrait
Negative Prompt: BadDream
VAE:vae-ft-mse-840000
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
vae-ft-mse-840000 here
Here I will show you the same basic prompt before and after adding a VAE on anime.
Anime Model: fCAnimeMix
Positive Prompt: woman, dress, fcPortrait
Negative Prompt: BadDream
VAE:NONE
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
Anime Model: fCAnimeMix
Positive Prompt: woman, dress, fcPortrait
Negative Prompt: BadDream
VAE: Blessed.vae
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
Blessed.vae here
The value of Checkpoints
So far we have used very standard checkpoints that aren't too artistic or stylized. They generally prompt what you ask and nothing extra. Hence my use of fcPortrait to set the image as a portrait with expected lighting / composition. Now we move towards art. Here we look at my Radiance series each wish tends towards different results with Apex being my favorite.
Artistic Realism Model: fCRadianceApex
Positive Prompt: woman, dress, fcPortrait
Negative Prompt: BadDream
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
Artistic Realism Model: fCRadianceAesthetic
Positive Prompt: woman, dress, fcPortrait
Negative Prompt: BadDream
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
Artistic Realism Model: fCRadiancePinaccle
Positive Prompt: woman, dress, fcPortrait
Negative Prompt: BadDream
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
Artistic Realism Model: fCRadiancePinaccle
Positive Prompt: woman, dress, fcPortrait
Negative Prompt: BadDream
Steps: 20, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: 3826206155, Size: 512x512
As we conclude this exploration of AI artistry, I want to assure you that this is just the beginning. There are still countless techniques, models, and innovations to discover and master. As we move forward, I promise to continue sharing my insights, offering more advanced strategies, and introducing you to the groundbreaking models that are shaping the future of AI-driven art. Stay tuned for more deep dives, more unique tricks of the trade, and more creative inspiration. Together, we'll continue to push the boundaries of what's possible in the mesmerizing world of AI art.
Please suggest which topics you want to cover next in the comments. I plan to do some more deep dives. Hope you guys stay with me.
My Links: twitter, discord, IG
Support my efforts at @ ko-fi.com/fitcorderai
Subscribe for FBB images @ https://patreon.com/fitCorderAI