|
Gave it a shot, it looks like it's the 'bare skin' part of the first prompt that's rejecting. the longer one in your second post worked fine for me.  Mordiceius posted:1990s fantasy anime screenshot, Highrise rooftop infinity pool, evening, neon lights of the city, woman relaxing in pool, she has black hair  e:f,b, so here's new stuff for a new page. Rankin-Bass Stop-Motion Horror Special Creatures + one Darth Vader:       "Death of a Goomba" by Francisco Goya 
Soulhunter fucked around with this message at 20:38 on Feb 1, 2024 |
#
?
Feb 1, 2024 20:28
|
|
|
|
| # ? May 29, 2024 15:55 |
|
|
Soulhunter posted:Gave it a shot, it looks like it's the 'bare skin' part of the first prompt that's rejecting. the longer one in your second post worked fine for me. That's the other option, the chatbot has completely different rules from the manual prompt but the images still show up in the same place.
|
|
#
?
Feb 1, 2024 20:29
|
|
|
Where do you guys go to "chat" with the Bing AI art tool? All I can do is give it prompts. Also, it makes me somewhat uncomfortable when AI starts using emoji :|
|
|
#
?
Feb 1, 2024 20:31
|
|
|
credburn posted:Where do you guys go to "chat" with the Bing AI art tool? All I can do is give it prompts. Go to bing.com press "chat" then scroll up if it doesn't move you up to the AI chat above the search that now loaded. Tell Copilot you want it to send a prompt it doesn't craft prompts like GPT+Dalle but you can convince it very simply by asking it to make a prompt for an AI image generator from your input and then send that prompt sample to it's image generator software. This has a 70% or so success rate of working in 1 shot, you only get 5 messages before you need a new conversation (you need to pay for GPT4 if you want unlimited). GPT4 does much better at reworks beyond the first but it's also easier to get bing to send exactly what you want. GPT4 likes to be helpful sometimes and change your prompt to be "better" without asking. pixaal fucked around with this message at 20:35 on Feb 1, 2024 |
|
#
?
Feb 1, 2024 20:32
|
|
|
Earwicker posted:
I read this as hyperbole from one person's perspective, but I imagine use of generative tools will become more common in the near future. moist banana bread fucked around with this message at 20:49 on Feb 1, 2024 |
|
#
?
Feb 1, 2024 20:43
|
|
|
moist banana bread posted:I read this as hyperbole from one person's perspective, but I imagine use of generative tools will become more common in the near future. Photoshop generative fill is the exact same thing. It's in use right now. And has been for years.
|
|
#
?
Feb 1, 2024 20:48
|
|
|
 what is "getting eggdogged"?
|
|
#
?
Feb 1, 2024 20:48
|
|
|
pixaal posted:stop trying to make softcore porn with bing use Stable Diffusion for that I have an AMD card.  pixaal posted:That's the other option, the chatbot has completely different rules from the manual prompt but the images still show up in the same place. Oh huh. I've never hosed with the chatbot. Just the website.
|
|
#
?
Feb 1, 2024 20:48
|
|
|
Earwicker posted:
bing error for using a banned word or name, or you described someone well enough that you set off a likeness check.  e: this can go here too 
|
|
#
?
Feb 1, 2024 20:50
|
|
|
pixaal posted:Photoshop generative fill is the exact same thing. It's in use right now. And has been for years. Yeah this is what I'm referring to in general, I had no idea until some time last year though.
|
|
#
?
Feb 1, 2024 20:55
|
|
|
For some reason, there's lots of keyphrases that aren't banned, or which work around the filter when prompting with bing chat, making the eggdog filter more of an annoyance and sometimes a complete crapshoot when you have an otherwise benign prompt that might be rejected for a random word. Bing chat lets you use phrases that probably should be banned still, like "Garfield executing an assassination", and has no problem with requests like "Charlie Brown and Lucy reenact scenes from the walking dead where Glenn's head is smashed with a spiked bat", "Spongebob learning to use a condom on a banana", and "Gritty curbstomping an NHL ref on the ice", so you can still pretty much do anything you want if you break out the thesaurus and get crafty. Some old examples of things that made it past the filters:       
Soulhunter fucked around with this message at 21:09 on Feb 1, 2024 |
|
#
?
Feb 1, 2024 21:03
|
|
|
you know we did banana phone, but what about corn phone?
|
|
#
?
Feb 1, 2024 21:15
|
|
|
Earwicker posted:
what the hell did it do to the barn lol
|
|
#
?
Feb 1, 2024 22:27
|
|
|
        AARD VARKMAN posted:what the hell did it do to the barn lol it seems quite capable of handling the barn well when asked to do a painting or etching, but when asked to create a photograph it can't handle the scale of the barn compared to the cow and renders it as a bunch of individual pieces for some reason, and it either floats in midair or its on a shelf or table. this is stable diffusion with control net Earwicker fucked around with this message at 22:37 on Feb 1, 2024 |
|
#
?
Feb 1, 2024 22:28
|
|
|
I just want you to know this post grossed me out, especially the dude with a hole in their head.   MidJourney is absolutely unable to recreate the magic of Demvr.   However --sref makes it very easy to make lovely fan art
|
|
#
?
Feb 2, 2024 01:40
|
|
|
Earwicker posted:
Holy poo poo, I didn't expect to see Cow Tools make an appearance.
|
|
#
?
Feb 2, 2024 02:02
|
|
|
awesome
|
|
#
?
Feb 2, 2024 02:14
|
|
|
Swagman posted:Horror prompts   for egg yolk eyes / empty eye sockets, shoutout to the Remnants read-along thread in The Book Barn for nightmare inspiration   
|
|
#
?
Feb 2, 2024 02:37
|
|
|
That's pretty tame IMO, it's a shame Microsoft has to be so prudish. Got SetArea doing SDv1-5 in quadrants, then passing that to SDXL to gloop it all together, so I can do stuff like this: Top Left: looking up at clear blue sky, the sun surrounded by clear blue sky above grassy hills Top Right: looking up at clear blue sky, with fluffy white marshmallow clouds Lower Left: grassy hills as far as the eye can see Lower Right: a (stone monolith:1.5) surrounded by grassy hills Global Positive: anime style, computer graphics, volumetric lighting, twilight hour. Global Negative: people, character, trees, faces, buildings SDXL Positive: A stone monolith in grassy hills, under clear blue sky filled with fluffy white marshmallow clouds, sun shining through  It's no Midjourney, but it is fun to mess with. I will share the workflow, but I'd like to parameterize some things and add a few short usage notes. (my favorite happy little accident to pop out while node wrangling)  Needs more than the 20 steps used for experimenting though, pretty grainy. moist banana bread fucked around with this message at 03:51 on Feb 2, 2024 |
|
#
?
Feb 2, 2024 03:37
|
|
|
   
|
|
#
?
Feb 2, 2024 04:34
|
|
|
likely a lack o leakage that makes it more macabre these abominations come 
|
|
#
?
Feb 2, 2024 04:43
|
|
|
      wanted to see if I could get bing to do a bag of cocaine, had to do ziplock full of baking soda, it wouldn't let me say "white powder"     lol baking soda works great for doing cocaine pictures my prompt was "party people sitting around coffee table, ziplock back full of baking soda on table, rolled up dollar bills, people have baking soda on their noses"   
LifeSunDeath fucked around with this message at 05:11 on Feb 2, 2024 |
|
#
?
Feb 2, 2024 04:48
|
|
|
moist banana bread posted:
This looks like a Worms map.
|
|
#
?
Feb 2, 2024 05:31
|
|
|
Haha, we could make it a worms map with mspaint I bet. Well at least I think everything's plugged in right. It's several other examples I found put together.        moist banana bread fucked around with this message at 09:39 on Feb 2, 2024 |
|
#
?
Feb 2, 2024 08:18
|
|
|
I got this bit of ludicrously violent Grand Guignol from a run of Bing images.  I guess Eggdog took the night off.
|
|
#
?
Feb 2, 2024 12:04
|
|
|
LAX security by Hieronymus Bosch 
|
|
#
?
Feb 2, 2024 19:19
|
|
|
here's some of my favorites from ~Christmas/January's trips to the image mines                        
|
|
#
?
Feb 2, 2024 21:28
|
|
|
Hey Swagman, you've posted a lot of great stuff in this thread. Tell you what, you make me a Monster Truck Bigfoot in a mechanics shop in parts and pieces without its chassis for my birthday tomorrow and I'll buy you plat. My profile has discord or email to get ahold of me for the certificate. I like irony, and that picture would be a "Naked Bigfoot Pic". For everyone else there have been a few interesting things of note and I'm just going to throw out a bunch of links to stuff. https://github.com/haotian-liu/LLaVA LLaVA: Large Language and Vision Assistant 1.6 released. This is a multi-modal chatbot that can "See" images you give it to describe what's in them and talk about them. This can't generate images on its own which is something I want to put together, an all-in-one Text/Image/Vision multi-modal generator, but you won't have to wait months for me to finish, this exists here and now. I look forward to being able to ask the AI Model directly what the deal with fingers actually is and getting an answer. There's a demo page you can try it without having to download or run anything on your own: https://llava.hliu.cc/ https://sliders.baulab.info/ Research paper about making concept sliders for SDXL, you use these LoRAs at various positive or negative strengths to get precise control of the effect you're looking for. Trained Concept Sliders link on that page has some premade to download and test out. Bigfoot (the beast) with a Perm sounds amazing. Actually, Bigfoot (the monster truck) with a Perm also sounds amazing. https://huggingface.co/Mitsua/mitsua-diffusion-one Exists as a public domain trained Stable Diffusion 1.5 model. The quality is not amazing but the fact that a free model without license or copyright issues that can be used locally without restrictions exists can not be overlooked. That models quality can always be increased at a later date with further fine-tuning. Earwicker posted:like you made some extremely silly claims about "anyone who's a 'real professional' is already using this stuff" and there are so many areas in which that isn't at all true, there are plenty of professional artists in many fields (yes including visual arts) who have no interest in using it. and there's nothing wrong with that. making AI images and words in this thread is fun and silly, making weird bullshit gatekeepery statements is not. I'm not touching the "Real Professional" part of this with a ten foot pole, I only want to link to this survey showing how wide spread this already is because it's only going to increase massively from here. https://arstechnica.com/gaming/2024/01/game-developer-survey-50-work-at-a-studio-already-using-generative-ai-tools/ near 50% of studios surveyed are already using these tools in some form. In the long run I believe it's going to be end up being really common that people just end up making their own sets of these tools custom to their own needs and wants. The sliders I linked above are just a start. https://www.fairlytrained.org/ https://www.theverge.com/2024/1/17/24041518/generative-ai-copyright-violation-fair-training-label-certification I've never believed for a single second that people actually cared about copyright issues and instead used that as some sort of rallying flag to point at how badly everything else has gotten under capitalism. Regardless, there's never been anything special about the creation of these Generative AI Models themselves outside of their massive compute requirement. It's only been a matter of time until eventually Foundational Models would be made without any issues surrounding them. Of course all these "certified" sites want your money though. I'll be keeping an eye out for more Free and Local Run "Fairly Trained" models and will point them out when I see them. If you run into someone that is irrationally mad about AI you can just point them to this site and tell them to knock themselves out. cumpantry posted:if you were a real professional you would suck the blood dry of your peers' past and present work https://www.fairlytrained.org/ Knock yourself out.  real professional sucking the blood dry of your peers' past and present work Steps: 48, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 1, Size: 512x512, Model hash: 99fd5c4b6f, Model: seekArtMEGA_mega20, Denoising strength: 0.32, RNG: CPU, Hires upscale: 2, Hires steps: 32, Hires upscaler: 4x-UltraSharp, Version: v1.5.2
|
|
#
?
Feb 3, 2024 01:14
|
|
|
I posed this question in the D&D AI thread thinking it would be more appropriate there, but it doesn't seem to be getting any traction so I'll ask here instead: Can anyone here recommend to me an “audio file to text” AI transcription github project that I can install locally to chew over a bunch of .wav files I’ve got kicking around, that doesn’t require me to upload those files to a third party service for processing? Think hundreds of hours of noisy lecture audio in raw 24 bit 48khz *.wav format that I can let an RTX card chug away at, that also isn’t the newest version of Microsoft Word. That’s what I’m looking for. Does such a thing exist?
|
|
#
?
Feb 3, 2024 02:33
|
|
|
Google's new AI art generator ImageFX is out and it can do dimetrodon   
|
|
#
?
Feb 3, 2024 15:00
|
|
|
XYZAB posted:I posed this question in the D&D AI thread thinking it would be more appropriate there, but it doesn't seem to be getting any traction so I'll ask here instead: I've had decent luck with this windows build of Whisper: https://github.com/Purfview/whisper-standalone-win/ It whipped through a bunch of hour+ long audio files on my 4080, and the results were pretty accurate. Be sure to get the cuBLAS/cuDNN libraries.
|
|
#
?
Feb 3, 2024 15:21
|
|
|
 at least chatgpt tells me what the issue is. I don't even know where to start with this one
|
|
#
?
Feb 3, 2024 15:27
|
|
|
It seemed like it failed for no reason the first time for me too
|
|
#
?
Feb 3, 2024 17:05
|
|
|
Apparently Street Lamps is a fail with Google's content policies. Was trying to make street lamps in a snowy night. Made this instead.
|
|
#
?
Feb 3, 2024 17:32
|
|
|
And now Street Lamps works fine.
|
|
#
?
Feb 3, 2024 17:36
|
|
|
AARD VARKMAN posted:Google's new AI art generator ImageFX is out and it can do dimetrodon I wonder when/if we can point AI at fossil bones and tell it to produce a scientifically accurate recreation of the live creature
|
|
#
?
Feb 3, 2024 17:38
|
|
|
AARD VARKMAN posted:Google's new AI art generator ImageFX so far this thing is pretty lovely tried my current favorite prompt "LAX security by Hieronymus Bosch" but that's against policy, apparently imitating the style of artists who've been out of copyright for over five centuries is now bad and wrong  so i tried the more generic and very simple "renaissance painting of LAX security" and got... this  think i'll be sticking with dall-e and sd lol
|
|
#
?
Feb 3, 2024 18:09
|
|
|
Earwicker posted:so far this thing is pretty lovely I'm not really surprised, for some reason Google seems extremely far behind anything to do with "AI" be it LLMs with their wet-fart Gemini or this new ImageFX thing. AARD's first dimetrodon looks ok, but that second one has the same sort of artifacts that midjourney had in it's extremely early versions, that kinda streaky, grain thing to it. However Google has been and still continues to be pretty decent with music to my not-music-making ears, recently updated after Suno was the darling: https://aitestkitchen.withgoogle.com/tools/music-fx
|
|
#
?
Feb 3, 2024 18:33
|
|
|
 cross your eyes - almost works as a "magic eye" stereo image Mescal fucked around with this message at 18:40 on Feb 3, 2024 |
|
#
?
Feb 3, 2024 18:38
|
|
|
|
| # ? May 29, 2024 15:55 |
|
|
bing prompt: corn people performing a séance around a big wooden table, ghost corn is floating above the table, the room is spooky  bing prompt: corn people sitting around a coffee table, there is a ziplock bag full of baking soda and some straws on the coffee table, and a mirror with baking soda on it, the corn people have baking soda on their faces, it is a party    
LifeSunDeath fucked around with this message at 18:46 on Feb 3, 2024 |
|
#
?
Feb 3, 2024 18:43
|
|