HOWTO: Craft beautiful outputs from prompt combination

[bmaltais]

Sorry to use the Issues to document this. This is not a bug or an issue but rather a demonstration on another possible way to use the variation feature recently introduced in v1.13. Please delete if not appropriate

---

I have been experimenting with the new variation feature of the v1.13 release and it is AWESOME. I want to explain a different way to use the feature vs what is part of the provided readme file. I think you will find it interesting.

I created a few images using the following prompt and selected the best two:

"portrait female commander shepard (amber heard), cyberpunk futuristic neon, hyper detailed, digital art, trending in artstation, cinematic lighting, studio quality, smooth render, unreal engine 5 rendered, octane render, Illustrated, Perfect face, fine details, realistic shaded, fine-face, pretty face" -H704 -n4

I did the same with this prompt:

"ultra realistic style illustration of a cute red haired (amber heard), sci - fi, fantasy, intricate, elegant, highly detailed, digital painting, artstation, concept art, smooth, sharp focus, illustration, 8 k frostbite 3 engine, ultra detailed" -H704 -n4

I then combined the two prompt and the 4 images using the following weighted prompt with variations:

"portrait female commander shepard (amber heard), cyberpunk futuristic neon, hyper detailed, digital art, trending in artstation, cinematic lighting, studio quality, smooth render, unreal engine 5 rendered, octane render, Illustrated, Perfect face, fine details, realistic shaded, fine-face, pretty face":1 "ultra realistic style illustration of a cute red haired (amber heard), sci - fi, fantasy, intricate, elegant, highly detailed, digital painting, artstation, concept art, smooth, sharp focus, illustration, 8 k frostbite 3 engine, ultra detailed":1 -V1152589909:0.25,3282369810:0.25,2192299756:0.25,3674209958:0.25 -S1 -v0.2 -n4 -t -s50 -W512 -H704

I divided 1 by the number of seed images (4) and doing this ensured that only "noise" from those 4 seed images would be used as a base for the combined prompt output. I allowed for 20% random noise to be added (-v0.2) to create some variation of the output.

Here are the two best resulting output. You see how the overall portrait remained the same but the style of the output is a blend between the two original prompt output. Really nice feature:

NOTE: I have nothing for or against Amber Heard, I just used her as I could easily see if the resulting images were trending toward a known value.

P.S.: You can use more that 2 prompts for the mix. I did 2 just to keep things simple...

[bmaltais]

Here is another example using another easy to recognize person. The resulting styles are really surprising:

Images from 1st prompt:

Images from 2nd prompt:

Resulting combination using the same technique as above:

[i3oc9i]

Very interesting thank you soo much for this sharing

[lstein]

This is very nice! Can I add this to the WIKI as the inaugural posting for a new "Tips and Tricks" section?

[lstein]

Sorry to use the Issues to document this. This is not a bug or an issue but rather a demonstration on another possible way to use the variation feature recently introduced in v1.13. Please delete if not appropriate

It's probably more appropriate to post to the Discussion tab, but people don't check there as often and I'm fine with the post being in Issues.

[bmaltais]

This is very nice! Can I add this to the WIKI as the inaugural posting for a new "Tips and Tricks" section?

For sure. I am happy to help others move the SD prompting forward by using all the nice features added to the fork.

[LipcaCZ]

Just add by Greg Rutkowski and BAM!, instant masterpiece from almost any prompt :D