@Crody I think you undo the vpred unless you specifically train with vpred. I just tried running with vpred settings and its all grey

@antiman9 Vpred is available for forge out the box, no reason to keep using A1111 

@metal079 came here to say the same thing. Got a grey image in reForge using the "Advanced Model Sampling" option for vpred, works fine for any other vpred model.

Not a big deal I guess, but certainly not vpred either.

The permission was already granted to me, so stop bothering me.

@Miguel45 

Is this your account? https://tensor.art/u/728067922689492343

@miao1973 Yes, he gave me permission. In fact, he even mentioned that he has no problem with anyone sharing his models without prior permission, as long as they’re not monetized and the proper links are included. That’s why I find it absurd that you’re complaining just because someone else uploaded it first.

Also, what monetization are you talking about? I removed it several days ago because, honestly, I only added it as an experiment, and I didn’t even make any money from it. So stop whining like a child and causing trouble just because someone else uploaded the model before you. Crody has no issue with people sharing his models without asking, as long as his conditions are respected.

@Miguel45 

https://civitai.com/models/503815?commentId=649212&dialog=commentThread

I'm talking about tensorART's Buffet plans.

I believe that any attempt to gain benefits from a model created by someone else is unacceptable, even if it is an experiment.

I understand that you removed it because you couldn't monetize it.

@miao1973 For your information, I never tried to monetize someone else’s models. The Buffet plan was something I created a while ago when I was experimenting with the idea of monetizing my two original models. However, it never generated any income. If you don’t believe me, I’d be happy to show you my profile, where you can see that I’ve earned a total of 0 since I had the Buffet plan active.

That said, I don’t understand why you’re bringing this up if I deleted the Buffet plan a long days ago. Could you please stop harassing me just because someone else uploaded a model before you? As I mentioned, today Crody gave me permission to upload this model, and a few days ago, he also gave me permission to upload his models. He even told me that he didn’t mind if other users uploaded his models, as long as his conditions were respected.

So, I don’t understand why you’re behaving this way, like a child, every time someone uploads a model before you. Your attitude seems childish and absurd, especially when Crody himself doesn’t care who uploads his models, as long as his terms are respected. Please mature and act like the adult I assume you are.

@Miguel45 

By "a long days ago," do you mean before I pointed this out on 1/5 on Tensor's discord?

I don't think a week ago would be considered a long days ago.

And even though it was a different model, you can't deny that you set it up to monetize.

Do you think it's okay because you have zero income?

The setting is supposed to be like YouTube's Superchat, where if no one uses it, there's no benefit.

Your setting must have been wrong, or no one used the feature.

Well, it seems like you were helped by it this time.

You're acting like a child.

@Crody didn't you use noob 1.1 in "illustrious v1.0"? is there any advantage in going back to noob 1.0? really just asking out of curiosity.

@Crody Oh wow, am i just really missing it this hard, or is neither noob nor v-pred really mentioned outside the small version description? I would maybe add "Noob 1.0 v-pred" to the version name, instead of Illustrious... because thats not it anymore. :'D

But i mean this is good, and why i asked, because i had no interest in using another IL model after using Noob models, and almost wouldnt have downloaded this because it says IL. Now ill ofc download it.

Maybe you need to check this again, as this loads as an EPS model instead of a VPRED model, and it does not do the darker and more contrast-rich gens that base noobAI VPRED does ...

@Xeno443 Yeah, ive gotten around to test it a bit now, and its definitely an eps model, itd be interesting to know what exactly @Crody meant. o:

I know, right??? This makes anthros turn out so good on illustrious!

What's the reason why we can't use on civitai? :/

@Crody You mean cel_shading? I tried , cel shading, anime_coloring, uncensored, newest on positive, , scan, old, oldest, (blurry, blurry_eyes, blur:1.14), white pupils, glowing eyes, watermark, artist name, bad anatomy, questionable anatomy, realistic, photorealistic details, photorealistic on negative. But it can't give 2d artstyle. Could you give some example pictures?

@laoxs I've had the most luck using terms like 'minimalism' and 'flat colors' along with putting 'shiny skin' in the negatives. You can lower the strength of the terms if you want to soften the effect, and removing your quality terms (Masterpiece, best quality, etc) can help as those tend to push things toward 2.5 styles.

That said, the easiest and most consistent way is to put artist names with the kind of style you're looking for in the prompt (mizumizuni, yukie \(kusaka shi\), kakure eria), or to find a LoRA of the style you prefer and use that with this checkpoint.

@Crody May I have your example prompt generating interactives? I failed to distinguish which character lying on back, about one third of the result is reversed position

I figured out since, apologies, my startup options were the issue. One of the following caused it: --opt-sdp-attention --opt-channelslast --autolaunch --disable-nan-check

I changed all of them and now it seems to be fine. Though --medvram keeps it from running out of VRAM a lot more.

@Crody Thank you, it means my guesses were correct, I'll try it soon.

@MAXMORD BREAK can be essential in keeping "bleed" to a minimum. It isn't perfect but it can help. Using it properly can really improve gens a lot.

The prompt gets sent in 75 token "chunks". If you type a prompt less than 75 tokens, your prompt takes up x amount and the rest is "padded". If you use 76, 1 chunk of 75 gets sent, and then the next chunk would have 1 token from your prompt- with the rest padded. it would still be 150 tokens that get sent in total whether your prompt was 76 tokens or 143. In some software, like webui and it's derivatives, typing BREAK in the prompt will pad the rest of the chunk it is in (in comfyui it is "conditioning concatenate"), and after BREAK will be the start of a new chunk. People use this to reduce concept bleeding. Concept bleeding is when "blue" is in one chunk and it describes the hair, but then the image also has blue clothing, or a blue background, etc. Separating the blue hair into a different chunk from the clothing would reduce this bleeding effect.

Be warned, however, the more "chunks", the slower the generation, and the less impactful any given chunk is on the whole image.

@Crody Hi, this question and answer has me intrigued. Is natural language/prompt language referring to the way the model interprets the prompt? If yes, where could I find more about how to combine/improve them, as you suggest?

@Crody Ah, I see. Thanks for the explanation

Where's this?

https://imgur.com/a/SWT52cr

@Crody Seriously??? Figured they'd be prepped for this sort of thing, but I guess not. lmao

@Crody Got it. Then I don't understand why my images with this model are not as sharp as others. It's kind of blurry. (I use Forge and Comfy)

@Crody Turns out it was my fault. If the forge has Euler and Euler a, then in Comfy I did not notice that I used Euler. 😑

Lora are still required for any characters, styles, or concepts the model is not trained on,

I have done some testing with Illust v3, and sometimes Pony LoRA will work kinda OK, but it will always be better to use Illustrious trained LoRA for better results.

You can absolutely use Pony (and SDXL) lora with Illustrious models, but:

*They are often very muted in effect. You'll probably need a higher weight on them to get the same effect as with the type of model they're trained on. For style lora, my usual rule of thumb is a pony lora used on Illustirous (or vis versa, this all applies in the other direction too!) is you need twice as strong a weight to get an equivalent result. Which holds true until it doesn't, some seem to work fine.

*Some types work better than others

*Some will not have quite the same result

*Some will be more corrupted - I notice more trouble with details getting garbled sometimes.

*Some will just flat-out not work, or at least not to any useful extent

So... Basically, "Maybe, maybe not, YMMV, try it and see what happens, management is not responsible for any abominations beyond human comprehension that get produced, may the gods have mercy on your soul."

@Crody Im sorry, i wasn't very clear. What i meant is i don't know a lot about programming, and software related stuff in general in the first place, so i have a follow up questions, what is IN06-08, and what multiplier and other settings should i use in the merger?

@Crody Just by looking at this merge plan example i'm getting suicidal thoughts. I don't understand three things, first is this line : "CM DHP + MKF +S SXL 0.2 0.3 _A" what is DHP, there was nothing named DHP, so where is it coming from? The second is "CM E + B "0,1,1,1,1,0.3,0.2,0,0,0,0,0,0,0,0,0,0,0,0,0,0.4,1,1,1,1,1" NovaFurryV3" what's with this alpha? Are these values? If yes, what does this number sequence represent. And the third thing is what's the example for alpha using IN06-08 is it like "CM ABC + XYZ IN06 _A"? I'm really sorry for asking you this much, i really know nothing about programming.

@Crody Ok, let's say i understand, but do i need to swap anything if i want to just merge NoobAI with NovaFurry? Wouldn't this work

"+NF3, https://civitai.com/models/503815?modelVersionId=1279960

+NAI, https://civitai.com/models/833294?modelVersionId=1190596

CM NF3 + NAI IN07 NovaFurryxNoob"
or am i fucking retarded, and that's not how alpha works?

@Crody Okay i made big advancements, that is i completed and now (almost) fully understand the script, completed merge plan, ran main.py and got txt file, what do i do with it now? (i use forge, not kaggle)

@Crody Hmm, i see. Welp im gonna stay on Forge for now, maybe will try Kaggle later. Anyway, thank you very much for help.