Daz to SD (img2img) - Spat's ZGFX-CHILLS

[Legolas18]

This one is a requested conversion of @spats 's render 🙂

AI Version:

Daz Version:

I used ControlNet Canny (to retain text and lines) 🙂

I used NO Photoshop.

[Spats]

Wow, cool!

While it's not Cassandra anymore, it's difficult to say that I mind, at all.
Great reproduction of tone and lighting, skin tone, ...and the realism. (The, erm, pumpkins are incredible.)

Great trick for keeping text and lines. Didn't know it was even possible.

Some of the detail lost on a couple anomalies in the background, the trophy animal is something even weirder and scarier now (and is there a new creepy doll's head behind the center of couch?); but man, everything else a different and interesting level.

Great work!

[Legolas18]

@spats It's not perfect, but I hope to have done a good enough job at it. I am glad you liked the "fruits" of the AI's labor. 😂

[Paul (Fuzzy3000)]

Nice work.

Smoother face and plumper bosom...

Face-wise even the DS version is a bit lacking, so the improved features more than make up for the not-quite likeness.
Will have to give a few attempts when time allows... 🙂

[Paul (Fuzzy3000)]

@spats
OK - couldn't resist it, had to try... 🙂

Just when you think the face is improving the txt goes crappy [and vice-versa] - Oh what fun.

Still... this is what I ended up with;

[right click for new tab, for full size]

[Legolas18]

@Fuzzy3000 Here are a few tips 🙂

1) Use Adetailer (mediapipe_face_full)
2) Use ControlNet Cammy (balanced, also tick the "Pixel Perfect" box)
3) The model used also has a great effect on the image. I am using RunDiffusionXL @0.5 denoise strength.

Hope it helps 🙂

[Paul (Fuzzy3000)]

@legolas18
Thanks
Every tip is worth it's weight in gold to start with.

So many options that are not even hinted at what they do - even with lots of YT vids there is no straight forward 'AtoZ of how to proceed, or the best order in which to try things.

Frustration thy name is SDXL

[Spats]

@fuzzy3000 Nice one. While the face is a bit more Cassandra-like it looks like it did go a touch wonky in the eyes. But, again, the colors, tone, textures - all great again. It seems you're right on the same SD img2img path, too.
Less worried about text preservation as compared to realism and original concept preservation, as stated originally - but also great that it's turning out as well as it is in these instances, an interesting proposition. At some point it seems it will be interesting to see what SD can not reproduce (or improve upon) in most respects.

"So many options that are not even hinted at what they do - even with lots of YT vids there is no straight forward..."

So, are you saying it's like using, oh I don't know, some multi-faceted graphics tool with esoteric features and plugins, but which doesn't have a manual?

[Paul (Fuzzy3000)]

@spats

What could you be referring too....?
🙂

[Author]

Posts