AI generator - i am lost

[coffeeaddicted]

I am just wondering, which of these prompts is actually good?

From what i see, there are many options. It seems you have to pay if you want to own the image otherwise it will be shared.
Images i had seen still have problems with fingers, though overall some look actually pretty good.
So, if i just want to play around and create something and my knowledge is rather limited (so far), which of these offers is good or good enough?

Would you have some recommendations?

 

[Doorknob22]

Yes. Download

You must be registered to see the links

and run everything locally. Also: why do you care if the images are shared?

 

[coffeeaddicted]

Yes. Download

You must be registered to see the links

and run everything locally. Also: why do you care if the images are shared?

Dunno. Never thought about it but i am not sure if i want to share everything. lol.
Thanks for the link

 

[coffeeaddicted]

Well, i create monster now.
I have to read more on it because it seems i can't set a higher resolution without getting multiple people in the image.
In any case, my monsters.

lol

 

[Dark Lord Nova]

Yeah, I'm running into the same problem. Every time I generate something, the faces come out warped—like totally off—and sometimes the body too, even when I'm really specific in the prompts (even the negative ones). I'll add stuff like "soaking wet," makeup details, and try to get real emotions—happy, sad, excited, terrified, angry—you name it. They still end up looking like some weird glitch. Lol. I've tried ComfyUI, but honestly, it's kinda more complex than Stable Diffusion or even Fooocus. I just found this new one called SwarmUI, thought it might be different, but nope… still getting the same results. Lol. I've gone through a bunch of different models too, from Civitai and even some from Hugging Face, but still no luck.

 

[cesid]

Try Forge. You will be limited but if you just do images is ok, it is very bad for videos.
The easiest checkpoint is epicphotogasm_ultimateFidelity. You don't need even the negative prompt and works with ancient computers since it is SD 1.5
The same guy of epicphotogasm did something good for SDXL that is more "intelligent" and require less loras. Then there is Flux that require more work to start but is even smarter than SDXL.
If you learn machine language and manage to speak with the computer 10101010101010 you should keep trying with Comfyui and try to master it.

 

[coffeeaddicted]

Yeah, I'm running into the same problem. Every time I generate something, the faces come out warped—like totally off—and sometimes the body too, even when I'm really specific in the prompts (even the negative ones). I'll add stuff like "soaking wet," makeup details, and try to get real emotions—happy, sad, excited, terrified, angry—you name it. They still end up looking like some weird glitch. Lol. I've tried ComfyUI, but honestly, it's kinda more complex than Stable Diffusion or even Fooocus. I just found this new one called SwarmUI, thought it might be different, but nope… still getting the same results. Lol. I've gone through a bunch of different models too, from Civitai and even some from Hugging Face, but still no luck.

Well, i found out that if you don't use their resolution, it will add the empty space with figures. So, if i just use the 512x512, it works fine. Nevermind the fuckup on every part.

I have to figure out how to set the higher resolution though. Plus, fill more in the negative prompt. I looked at some and they added like "ugly", "deformed face" or such.

I just started yesterday, so i am not too wise. I take any lead if there is any.

And then i look at something like this. While not perfect, it looks somewhat real. But in the end, AI gives you still horrible images no matter how good it is done. Unless there is some AI that masters fingers or nails.

View attachment 678ee6fd027ef991b5e7e872.webp

 

[coffeeaddicted]

Try Forge. You will be limited but if you just do images is ok, it is very bad for videos.
The easiest checkpoint is epicphotogasm_ultimateFidelity. You don't need even the negative prompt and works with ancient computers since it is SD 1.5
The same guy of epicphotogasm did something good for SDXL that is more "intelligent" and require less loras. Then there is Flux that require more work to start but is even smarter than SDXL.
If you learn machine language and manage to speak with the computer 10101010101010 you should keep trying with Comfyui and try to master it.

Again another AI.
Forge? Sounds like Mordor.

 

[tanstaafl]

Again another AI.
Forge? Sounds like Mordor.

Comfyui is fine, so is stable diffusion, forged is good too. The UI is not your problem, the problem is just lack of knowledge of prompts, aspect ratios, and models (checkpoints) to use. Go watch some youtube videos for beginners. Pick whatever UI you like.

 

[coffeeaddicted]

Comfyui is fine, so is stable diffusion, forged is good too. The UI is not your problem, the problem is just lack of knowledge of prompts, aspect ratios, and models (checkpoints) to use. Go watch some youtube videos for beginners. Pick whatever UI you like.

Thats correct. I am a bloody beginner in the genre. I will get there. Hopefully

 

[tanstaafl]

Thats correct. I am a bloody beginner in the genre. I will get there. Hopefully

For what it's worth I prefer comfyui and sdxl models.

 

[Dark Lord Nova]

Forge is basically Stable Diffusion. I've been around the block with this stuff—used ComfyUI, Forge, Fooocus, Wan 2.1, and lately I've been playing with SwarmUI. Some of you probably know the JuggernautXL model—I've used that a bunch, along with Flux and SDXL models for testing out different checkpoints and LoRAs.

This is the negative prompt I usually throw in. Sometimes it nails it, sometimes it totally misses—depends. I also mess with resolutions like 512x512 or 1024x1024, tweak the steps (anywhere from 20 to 50), and adjust CFG scale between 5 and 7, sometimes higher if I'm chasing something specific.

Negative Prompt: 2d, 3d, cartoon, anime, drawing, painting, illustration, dull, glitch, cropped, out of frame, (((double heads))), bad hands, bad fingers, deformed hands, deformed fingers, deformed arms, deformed legs, deformed feet, deformed toes, deformed face, bad face, cross-eyed, censored, deformed, bad anatomy, disfigured, mutation, mutated, extra limb, ugly, disgusting, poorly drawn hands, missing limb, floating limbs, disconnected limbs, disembodied limb, oversized limbs, bad hands, malformed hands, (((mutated hands and fingers))), six fingers, 6 fingers, 4 fingers, four fingers, 5 fingers, five fingers, 3 legs, three legs, four legs, 4 legs, five legs, 5 legs, six legs, 6 legs, watermark, watermarked, oversaturated, distorted hands, obese, doubled face, doubled hands, prosthesis, unreal, fake, blur, blurry, out of focus, out of frame, angle shots, (closeup:2), (bust shot:2), (half shot:2), (badhadc4:1.2), (worst qulity:2), (low quality:2), (normal quality:2), lowres, ((monochrome)), ((grayscale)), B & W, black and white, whole body out of frame, full body out of frame, complete body out of frame, full body out of frame, close ups, sepia, clones, clone, multiple girls, multiple women, double heads, through walls

 

[tanstaafl]

Forge is basically Stable Diffusion. I've been around the block with this stuff—used ComfyUI, Forge, Fooocus, Wan 2.1, and lately I've been playing with SwarmUI. Some of you probably know the JuggernautXL model—I've used that a bunch, along with Flux and SDXL models for testing out different checkpoints and LoRAs.

This is the negative prompt I usually throw in. Sometimes it nails it, sometimes it totally misses—depends. I also mess with resolutions like 512x512 or 1024x1024, tweak the steps (anywhere from 20 to 50), and adjust CFG scale between 5 and 7, sometimes higher if I'm chasing something specific.

Negative Prompt: 2d, 3d, cartoon, anime, drawing, painting, illustration, dull, glitch, cropped, out of frame, (((double heads))), bad hands, bad fingers, deformed hands, deformed fingers, deformed arms, deformed legs, deformed feet, deformed toes, deformed face, bad face, cross-eyed, censored, deformed, bad anatomy, disfigured, mutation, mutated, extra limb, ugly, disgusting, poorly drawn hands, missing limb, floating limbs, disconnected limbs, disembodied limb, oversized limbs, bad hands, malformed hands, (((mutated hands and fingers))), six fingers, 6 fingers, 4 fingers, four fingers, 5 fingers, five fingers, 3 legs, three legs, four legs, 4 legs, five legs, 5 legs, six legs, 6 legs, watermark, watermarked, oversaturated, distorted hands, obese, doubled face, doubled hands, prosthesis, unreal, fake, blur, blurry, out of focus, out of frame, angle shots, (closeup:2), (bust shot:2), (half shot:2), (badhadc4:1.2), (worst qulity:2), (low quality:2), (normal quality:2), lowres, ((monochrome)), ((grayscale)), B & W, black and white, whole body out of frame, full body out of frame, complete body out of frame, full body out of frame, close ups, sepia, clones, clone, multiple girls, multiple women, double heads, through walls

Yeah, JuggernautXL is good, I also used DreamShaperXL and RealCartoonXL a lot. As for negatives I usually went with

You must be registered to see the links

.

 

[Dark Lord Nova]

Yeah, JuggernautXL is good, I also used DreamShaperXL and RealCartoonXL a lot. As for negatives I usually went with

You must be registered to see the links

.

Really appreciate you sending me that link to the embeddings — they're seriously making a difference. Getting some solid results already.

 

[tanstaafl]

Really appreciate you sending me that link to the embeddings — they're seriously making a difference. Getting some solid results already.

View attachment 4818839 View attachment 4818840 View attachment 4818866

Okay, sorry for this dump. But this is my text file dump that I use to keep notes for ultimate embeddings. You'll want to download the positive version of the ultimate embeddings to use along side the negative for this to do anything. It helps with adding...let's call flare and style to your images. Obviously the <name>_Neg is for negative and those without are for positive prompts.

Ultimate Embeddings:

ZipRealism = UHD, 8k, F2.8, RAW Photo, ultra detailed, sharp focus

ZipRealism_Neg = weird eyes, mutated hands, ugly, drawing, illustration, low quality, watermark, tiling, mutation, poorly drawn hands, poorly drawn face, bad anatomy, missing fingers, crooked teeth, extra fingers, disfigured, error, abstract, 3d render, cgi, fused fingers (sometimes when I type "deformed pupil, deformed iris" like I normally do in SD15, it often makes things worse...so they are not in the list)

Zip2D = masterpiece, best quality, ultra-detailed, sharp image, high resolution

Zip2D_Neg = worst quality, low quality, lowres, poorly drawn hands, poorly drawn face, bad hands, bad anatomy, 3d, blurry, watermark, signature, ugly, jpeg artifacts

AC_Neg1 = text watermark signature,low-quality,moiré pattern, downsampling, aliasing, blurry blur, glossy, jpeg artifacts, compression artifacts,poorly drawn,low-resolution bad ,error,pattern,beginner, grainy, glitch (More on poor quality image errors)

AC_Neg2 = distorted, distortion, twisted, excessive, exaggerated pose, exaggerated limbs, symmetrical,duplicate, pixelated, fake, hyper, overexposed, high-contrast, bad-contrast (More on style, composition and colors)

I do not recommend use all these negative prompt at the same time.

SK_3D = professional 3d model, octane render, highly detailed, volumetric, dramatic lighting

SK_Analogfilm = analog film photo {prompt}, faded film, desaturated, 35mm photo, grainy, vignette, vintage, Kodachrome, Lomography, stained, highly detailed, found footage

SK_Anime = anime artwork, anime style, key visual, vibrant, studio anime, highly detailed

SK_Cinematic = cinematic film still, shallow depth of field, vignette, highly detailed, high budget Hollywood film, bokeh, cinemascope, moody, epic, gorgeous, film grain

SK_Comic = comic, graphic illustration, comic art, graphic novel art, vibrant, highly detailed","photograph, deformed, glitch, noisy, realistic, stock photo

SK_DigitalArt = concept art, digital artwork, illustrative, painterly, matte painting, highly detailed

SK_Fantasy = ethereal fantasy concept art , magnificent, celestial, ethereal, painterly, epic, majestic, magical, fantasy art, cover art, dreamy

SK_CyberPunk = neonpunk style, cyberpunk, vaporwave, neon, vibes, vibrant, stunningly beautiful, crisp, detailed, sleek, ultramodern, magenta highlights, dark purple shadows, high contrast, cinematic, ultra detailed, intricate, professional

SK_Ghibli = Studio ghibli style, Art by Hayao Miyazaki, Anime Style, Manga Style, Hand drawn, cinematic, Sharp focus, humorous illustration, big depth of field, Masterpiece, concept art, trending on artstation, Vivid colors, Simplified style, trending on ArtStation, trending on CGSociety, Intricate, Vibrant colors, Soft Shading, Simplistic Features, Sharp Angles, Playful

SK_Vectorart = Vector art, Vivid colors, Clean lines, Sharp edges, Minimalist, Precise geometry, Simplistic, Smooth curves, Bold outlines, Crisp shapes, Flat colors, Illustration art piece, High contrast shadows, Technical illustration, Graphic design, Vector graphics, High contrast, Precision artwork, Linear compositions, Scalable artwork, Digital art (This one has a very strong effect, please use higher weight on your main prompt)

 

[coffeeaddicted]

I just discovered that i am not only a bloody beginner but also lame.

No wonder it didn't work as i was expecting. I had no models. I found a helpful video to make sure i have everything before even starting.

You must be registered to see the links

Btw. the images i was downloading for my enjoyment suffer all from finger curse. Are they creations that don't suffer from that? What i am trying to find out is a) if that can be avoided and b) if AI is maybe not there yet.
Quasi a newbie question.

 

[tanstaafl]

I just discovered that i am not only a bloody beginner but also lame.

No wonder it didn't work as i was expecting. I had no models. I found a helpful video to make sure i have everything before even starting.

You must be registered to see the links

Btw. the images i was downloading for my enjoyment suffer all from finger curse. Are they creations that don't suffer from that? What i am trying to find out is a) if that can be avoided and b) if AI is maybe not there yet.
Quasi a newbie question.

For distorted hands there are a couple options. There's a Loras for hands and you may want to checkout hand detailer

You must be registered to see the links

 

[coffeeaddicted]

Kind of works now.
Judge for yourself.

Edit 1: added some information.



I think changing the checkpoint is what you need to do. Still a bloody amateur but at least i understand a little more.

Lovely.

 

[cesid]

Btw. the images i was downloading for my enjoyment suffer all from finger curse. Are they creations that don't suffer from that? What i am trying to find out is a) if that can be avoided and b) if AI is maybe not there yet.
Quasi a newbie question.

If you stick with 1.5 you should need a lora but with SDXL and more advanced you should never have the problem unless the checkpoint itself is terrible. Mostly don't even require "bad hands" or "extra fingers" as negative prompt.

 

[bigbootyluver69]

Honestly looking through your post history I'm impressed that you've stuck with ComfyUI as long as you did, most people try it out and dip out because honestly, it's fucking hard. I'll give you some pointers since it's a LOT to take in but once you understand the basics you'll be able to generate good images.

Kind of works now.
Judge for yourself.

Edit 1: added some information.

View attachment 4820491

Looking at your workflow you've got the basics down, I notice you're using a lighting model, in short a lighting model generates images faster with less steps but there's a cost for the faster generation (i don't remember what the troll takes from you). Which is fine, but you'll want to lower your steps to 8-14 ish (look up what the model recommends in the description), since doing more only adds time to your generations and time is gold in the world of AI.

Also, you'll want to change your Sampler_Name to dpmpp_2m since it generates real images better than euler, keep the Scheduler on Karras since dpmpp samplers work good with that scheduler. There are guides that explain what samplers to use with what scheduler but since you're learning just keep it simple for now.

Another thing with your image, you see how it's kinda burned/overexposed? That usually means the model had a hard time creating the image which could be a million things but the most common are 1) too high CFG, bring it down to 4-5 instead of 8, this allows the model to have more freedom to interpret your prompt but might not follow what you write. 2) The checkpoint/model isn't really trained on what you're prompting it. Specifically the keyword I noticed was "beautiful older woman", I use "mature woman" and it tends to work better but at lot of the models on Civitai are trained on younger women instead so maybe try a different model to see if you get better results? Prompting in it's self is a whole other talent and will drastically change your images.

Besides those things you should be able to generate decent images with that workflow you have, your only limitation will be your GPU and how much VRAM you have and of course your time.

I just discovered that i am not only a bloody beginner but also lame.

No wonder it didn't work as i was expecting. I had no models. I found a helpful video to make sure i have everything before even starting.

You must be registered to see the links

Btw. the images i was downloading for my enjoyment suffer all from finger curse. Are they creations that don't suffer from that? What i am trying to find out is a) if that can be avoided and b) if AI is maybe not there yet.
Quasi a newbie question.

To answer your question about fingers, it depends....sometimes you get lucky and the model produces a good image with good hands/fingers and not looking like some creature from the deep. Here are two images I created awhile back with only a text prompt.


I want to say I generated at least 10 images and out of those only like 3-4 had usable hands, but yes you can avoid the finger curse but what most people do is they fix the hands later down the line with inpainting or detailer nodes. Which is a more advanced concept and personally if I were you I'd spend more time learning how prompts work and how to at least get images you're happy about before going into fixing small parts of an image.


Edit: I was curious if I could recreate what you're trying to generate and the results weren't too bad, I couldn't figure out how to correctly fix the hands since the workflow I use didn't account for both hands, I usually only have one hand to fix lol. But as you can see I started with a pretty crap image and slowly iterated on it to what I wanted it to be. Left - > Right, 1) Initial text prompt image (really fugly) 2) Img 2 Img workflow to get some details before fixing hands and face 3) Face fixer and hands fixer...kinda, I still haven't learn how to fix hands especially realistic hands.