VN Ren'Py How Hard Can It Really Be? [v0.6.1] [Waterthose]

[Lerd0]

How the fuck is the spreading render made? It alludes me how people can get good AI renders. I do art, and I've tried manipulating AI to get references for some of my work (especially for hands or such), and I never get stuff with pretty much perfect anatomy

...get some loras ma boi......and use better models....

You must be registered to see the links

 

[The Loli Lover]

...get some loras ma boi......and use better models....

You must be registered to see the links

When I say not good anatomy, I don't mean like 6 fingers or such. It's more so just that I have 0 clue how it draws the hand-relative to vagina ratio so great and also very clearly and well. Most of the highest-rated images I've seen on Civit have much worse quality than this

 

[Lerd0]

Most of the highest-rated images I've seen on Civit have much worse quality than this

...dats cuz ppl there have really shit taste lol......

 

[Waterthose]

How the fuck is the spreading render made? It alludes me how people can get good AI renders. I do art, and I've tried manipulating AI to get references for some of my work (especially for hands or such), and I never get stuff with pretty much perfect anatomy

That's a secret! Just kidding, it's just a model that probably has a lot of images like this trained in it, this was done without a LoRa.

 

[FrozenPie1]

When I say not good anatomy, I don't mean like 6 fingers or such. It's more so just that I have 0 clue how it draws the hand-relative to vagina ratio so great and also very clearly and well. Most of the highest-rated images I've seen on Civit have much worse quality than this

I found that the "Illustrious" based models like "Hassaku XL (Illustrious)" or "WAI-illustrious-SDXL" or "Nova Anime XL" have really good anatomy and posing even without specialized LoRAs. There even things like the "ear licking of elven ears" work out of the box (AFAIK). The "Pony" models are far worse in that regard.

...dats cuz ppl there have really shit taste lol......

That's somehow very true

 

[Lynava]

Oh my god that intro was amazing
claps* claps* claps* claps*

 

[Waterthose]

Welp, I didn't find the you/you're issue but I've fixed the CG/renamed the other thing/changed the line on beach scene and adjusted some sprites.

I pray there aren't anymore glaring errors. (Whenever it gets updated to v0.6.1 here.)

 

[The Loli Lover]

That's a secret! Just kidding, it's just a model that probably has a lot of images like this trained in it, this was done without a LoRa.

I just don't understand AI in general as an artist. I get it with like 5000+ pictures, since it's a large sample size, but idk how the hell you train AI without being that elaborate, since I highly doubt all the AI games on here are training their own LORA's on 1000+ images?

 

[FrozenPie1]

Welp, I didn't find the you/you're issue [...]

Then I was just hallucinating maybe... Sorry for the inconvenience

I just don't understand AI in general as an artist. I get it with like 5000+ pictures, since it's a large sample size, but idk how the hell you train AI without being that elaborate, since I highly doubt all the AI games on here are training their own LORA's on 1000+ images?

This stuff is just one thing: Statistics. It gets pictures which are tagged and the training tries to find similiarites or the same things between pictures which have the same tags (like long hair) paired with an image recognition step to split different parts of the pictures apart. The better the tagging, marking, splitting apart and quality of the images on which the model is trained, the better the resulting model. LoRAs mostly specialize for a specific tag like "Boobjob" for example and work with different models and give them a "better grip" on what should be generated how.

 

[Waterthose]

I just don't understand AI in general as an artist. I get it with like 5000+ pictures, since it's a large sample size, but idk how the hell you train AI without being that elaborate, since I highly doubt all the AI games on here are training their own LORA's on 1000+ images?

Some consider AI a "Black box" where you insert prompts and it magically generates things close to what you want.
Models are far more complicated to explain and is what causes your generation to show in a certain "style", these use a LOT of training data to become a model.
LoRa's are usually not trained on that many images as that can impact the output not to mention the tagging for 1,000 images... you could train a LoRa with sub 10~30 images. Creating your own LoRa is easier than making/merging your own models, so it is very possible for the AI games on F95 to have LoRa's for their own characters and such.

 

[FrozenPie1]

Alternatively to LoRAs you could also use a ControlNet (like a scribble node which also works with normale pictures and not just scribbles) to use an example image on which you base the output. This is extremly helpful when generating complex scenes but can also hinder other parts if not used carefully

 

[Waterthose]

Then I was just hallucinating maybe... Sorry for the inconvenience

Eh I was only looking at you're which was sub 150 entries but your is over 700 so...... next time if you see it write it down.

 

[FrozenPie1]

Eh I was only looking at you're which was sub 150 entries but your is over 700 so...... next time if you see it write it down.

Heh, I will for sure but looking through the text while screenshotting and marking stuff is much more difficult then reading the script and correct in-place or write the correction behind it. You have too know the point in text too after all
Could also be that it was a "you" which should have been a "your" or something like that

 

[Waterthose]

Heh, I will for sure but looking through the text while screenshotting and marking stuff is much more difficult then reading the script and correct in-place or write the correction behind it. You have too know the point in text too after all
Could also be that it was a "you" which should have been a "your" or something like that

What Milton does is give me the entire sentence and highlight what the error is.

 

[Avaron1974]

I've only just got round to playing this and it's genuinely one of the funniest games i've played.

I love it.

 

[FrozenPie1]

What Milton does is give me the entire sentence and highlight what the error is.

So also just screenshotting with the marked mistake? I think you could make it much easier for him if you give him the script as a text-file

 

[Waterthose]

I've only just got round to playing this and it's genuinely one of the funniest games i've played.

I love it.

Omg yes! You were one of the readers I really wanted to see try out my AVN.

So also just screenshotting with the marked mistake? I think you could make it much easier for him if you give him the script as a text-file

Nah he writes the entire sentence or copy n pastes it, he uses URM so he's looking at the script the entire time.

 

[FrozenPie1]

Nah he writes the entire sentence or copy n pastes it, he uses URM so he's looking at the script the entire time.

Ah well, didn't know there was this possibility with URM

 

[Waterthose]

Ah well, didn't know there was this possibility with URM

Don't ask me! I'm just guessing.

 

[MiltonPowers]

So also just screenshotting with the marked mistake? I think you could make it much easier for him if you give him the script as a text-file

I use the script. But it's better to use the whole sentence and context, that way the dev can decide if my suggested changes will be used. Because it is spoken, not written word, sometimes the correct grammar is not the right thing. Rather than me changing it and getting away from the dev's intention, he/she can decide.