DreamShaper - V∞!

afaik A1111 generates the seed/noise on GPU, whereas ComfyUI generates the seed/noise on CPU.

still no reply?

yeah i wanna know this too.. all i find about diffusers is some old model fine-tuning back when all AI images still looked like Dali paintings, why is it being used lately by some creators?

That's odd honestly, I have more mature male characters appearing and have to use an age slider to get male characters to at least appear like they are in their 20's. What prompt are you using that you keep getting children in your images? I know "wide eyes" often gives me a younger look to any character but I can say I haven't had children appearing in my prompts' results.

Also what's your negative expression/prompt. Almost everything if you dont wish to have children should at least have a textual that removes children completely for your work. Like I can put a full prompt in no matter what model I use but without a negative the AI seems to put children in, even my own model I created of photography that I've done which are mature male models it will give people much younger. SO it's always safe to put a negative prompt to tell the AI you want nothing underage.

@A_Broken_Mind If you read the context in my comment, I explain that I was testing checkpoints for their conceptual representations of single words or expressions.
No negative prompt, only positives like "masterpiece" or "astonishing" etc. And those only generate with the model's biases. DSXL was the only one with so many children, that's why I'm asking my question.

I did not say I only got children in directed generations.

Hope that clarifies my comment.

@IndustrialVectors well it's pretty broad, when using single word responses. Like masterpiece, is a child not a masterpiece of two people, an AI may see children as the masterpiece of the adults.

@A_Broken_Mind SD models get results based on the bias acquired in their training dataset, images and tags. I often test models against single words on multiple generations to get a sense of the tokens to use.
DS 8 was strongly biased for children. I don't mind as I said, but I asked about the training data.
(I say that I don't mind because as a photographer who started with SD for outpaint/inpaint, I do have to work with children as well.)
Let's not argue anymore about that topic, pretty please.

Once I had a look over the subset from laion-art-aethestethic data, and it includes a lot of wedding photos with children in suits for some reason

I want to know too.

a lot of images on civitai are wrongly documented. straight up random stuff sometimes. maybe it's the case too, or other resource are used. either way, lot of gen data is wrong.

there is tons of parameters,settings,extensions that civitai doesn't show, they can dramatically change your outputs

@DRDtaker yeah i get that. there should be a tag or something that say what is used to generate.. either fooocus, comfyui, a1111, forge, civitai or else.. but sometimes, it's literally not the issue, i see generation data from a 1.5 model, with LoRAs from SDXL, and absolute random words in the prompt that have NOTHING to do with the image.

your loss

explain?

Oh no, anyway...

There's a whole lot of lost context in terms of making a comment like this.

what is this SAI?

@dilectiogames "StabilityAI." but don't listen to these children. they get free stuff and some models don't make the prolapsed waifu bukakke porn they expected, so they throw hissy fits and blame the authors for attempting to monetize across demographics (y'know, to spread beyond the ''teenage masturbating boy'' demo), and he shockingly didn't successfully navigate the minefield that amounted to a braying months-long tantrum thrown by a buncha suburban twats with their softening dicks in their sweaty little hands

it dependes of what u just downloaded, for example Checkpoints should be in the models/stable-diffusion folder, and Lora should be in the models/Lora folder and do that for every model that u need to use

@SsouxieE ok but sometime I see i model with type "Lora" or "Pony" but when they download their name end with safetensors? so should i put them in Lora Folder or checkpoint folder?
(I am using comfyUI, and it seems that even the models with type Lora when downloaded with extension Safetensor needs to be put in checkpoint folder otherwise they don't show up in comfyUI)
Any idea what's happening? I am new to this :)
Also where to add Pony models in comfyUI?

@KinkySensei I will answer as far as I know, so, the .safetensors is just an extension, it's very common to see it, and as far as I know it should be in their correspondent folder, Lora in Lora folder, Checkpoints in the stable-diffusion folder (That's how it appear for me) and u know how it continues, and about the ComfyUI idk, I just use Automatic1111, so im not sure how ComfyUI works, aaaand I'm new in this too lol so im just trying to help just with the things that I know, I hope this help. Have a good day

Didn't know there was a specific one? I usually use 8XNMKD but have used others...

I use Ultramix_Balanced. I tend to like its results if I need to use high denoise. Fantastic upscaler for 2D and 2.5D, in my opinion.

if you have the civitai browser extension you can automatically add the config files and stuff. Get A1111 running first, then put the model into the right folder, then run the civitai extension and it will update with the info

I don't believe you would. You should use the safetensors version with Comfy. The diffusers version is useful for Python programmers, especially folks who are training their own custom models. Great question!

yep

@fernes uh huh

@null Nope

Thanks for all your reviews they are very useful👍👍

@charlesp Thank you for your reaction! So all my work is not in vain. This is very valuable to me. Thanks!

Maybe not complete garbage but nothing useful here.

You can, but it's not good at it.

lost technology

ComfyUI/models/checkpoints

I would recommend creating directories within that folder that match the type of model, so SD15/SDXL/SD3 etc.

@gelikib398342 Thanks!

and it's cheap !!!!

DreamShaper XL - v2.1 Turbo DPM++ SDE | Stable Diffusion XL Checkpoint | Civitai

Look at the Dreamshaper xl page

Its "Helping" because that is the norm the engine has.

@PrimeMover ok, is there a way to stop it? There must be if I've seen many use it without getting humans

bro did you find a fix for this? Everything I make looks asian

@neon_sin459 there are a few loras out here thay you can use. But i suggest you pick a good checkpoint. Look at the sample images of others and see if you can spot the asian face there. If yes than dont use that.

cnm