Post Ac8RYGkxWMIV8dzvaS by john@sauropods.win
(DIR) More posts by john@sauropods.win
(DIR) Post #Ac60TANoheMAqUazom by john@sauropods.win
2023-11-23T10:56:02Z
0 likes, 0 repeats
Can somebody please get an AI image generator to generate a fox with a human face or head for me?The other way around is really easy and does not count.#AI #StableDiffusion #DALLE #Midjourney
(DIR) Post #Ac60oecPU3TshTfSUa by miekeroth@socialserver.science
2023-11-23T10:59:53Z
0 likes, 0 repeats
@john seriously..
(DIR) Post #Ac60x7g7K9ON90RKeu by john@sauropods.win
2023-11-23T11:01:26Z
0 likes, 0 repeats
@miekeroth It’s trained on way too much furry art!
(DIR) Post #Ac60ygJ63pc6phPokC by miekeroth@socialserver.science
2023-11-23T11:01:45Z
0 likes, 0 repeats
@john yep!
(DIR) Post #Ac616jSaDohOvcgwJE by AggroBoy@mastodon.social
2023-11-23T11:03:04Z
0 likes, 0 repeats
@john I spent about half an hour trying last night and just couldn't get it to do it. In a great many attempts, I got *one* example out of DallE4 (that I didn't save) where it had sortof superimposed a textureless grey human face on top of a fox's head, but it didn't look like it was actually part of the head. The rest were either normal foxes, or humans wirh fox heads. Weirdly, it really liked generating humans with HUGE foxes heads.
(DIR) Post #Ac61XEdiv2qt9xXXHM by AggroBoy@mastodon.social
2023-11-23T11:04:39Z
0 likes, 0 repeats
@john I guess it's just a rare (unique?) enough composition that there are no examples to extrapolate from in the various training sets.
(DIR) Post #Ac61XFYnUqVQ0yF7Eu by john@sauropods.win
2023-11-23T11:07:57Z
0 likes, 0 repeats
@AggroBoy Yeah, but I would have thought it could just composite the concepts, like it does with subject and background, for example.
(DIR) Post #Ac61clwMXGZ9t5tlS4 by catselbow@fosstodon.org
2023-11-23T11:08:34Z
0 likes, 0 repeats
@john By "other way around" do you mean "get a fox with a human head to generate an AI image generator"?
(DIR) Post #Ac61jwYJTijwpN4LFQ by john@sauropods.win
2023-11-23T11:10:08Z
0 likes, 0 repeats
@catselbow That’s how we got them, right?Joking aside, this is serious, be serious and get prompting!
(DIR) Post #Ac650X6Hf4qUftYoxk by pbloem@sigmoid.social
2023-11-23T11:46:44Z
0 likes, 0 repeats
@john DALL-E 3 seems to manage, and it's suitably majestic.
(DIR) Post #Ac658GR5VH6HJ7yhOK by john@sauropods.win
2023-11-23T11:48:17Z
0 likes, 0 repeats
@pbloem Damn, you did it! How many times did you try?
(DIR) Post #Ac65CL5u7rU2uh0qzw by pbloem@sigmoid.social
2023-11-23T11:49:03Z
0 likes, 0 repeats
@john First try, I promise...
(DIR) Post #Ac65Oghbc5BtP9XGTI by john@sauropods.win
2023-11-23T11:51:16Z
0 likes, 0 repeats
@pbloem I burned all my credits on Bing and it wouldn't work. @AggroBoy tried for ages on DALL-E4 and nope.I wonder if putting it through the chat interface made it work?
(DIR) Post #Ac65cWVYJRMFqEUknw by pbloem@sigmoid.social
2023-11-23T11:53:36Z
0 likes, 0 repeats
@john @AggroBoy The success rate is 3/5. Note counting this one, which is tehcnically correct, I guess.
(DIR) Post #Ac664tcRyl2hd0FbGa by pbloem@sigmoid.social
2023-11-23T11:54:52Z
0 likes, 0 repeats
@john @AggroBoy ChatGPT does do some stuff behind the scenes, including writing its own prompt. You can see part of that in the filename when you download the image. The prompt it wrote for the first image I posted was "A surreal and imaginative depiction of a fox with a human head, blending the natural orange and white fur of the fox with the distinct features of a h..." (the filename cuts off there).
(DIR) Post #Ac664uaMO0xscoHReC by pbloem@sigmoid.social
2023-11-23T11:57:26Z
0 likes, 0 repeats
@john @AggroBoy Ah, I can just ask it what prompt it used. Here's the full thing.
(DIR) Post #Ac664vPPKDnXB8ADDM by john@sauropods.win
2023-11-23T11:58:52Z
0 likes, 0 repeats
@pbloem @AggroBoy AIs are better prompters than humans. There we go.
(DIR) Post #Ac66AN97crbYeV8yky by hrbrmstr@mastodon.social
2023-11-23T11:59:50Z
0 likes, 0 repeats
@john one try (via ChatGPT+)
(DIR) Post #Ac66U1jOxIL61jU9PE by john@sauropods.win
2023-11-23T12:03:24Z
0 likes, 0 repeats
@hrbrmstr It seems that ChatGPT knows how to get DALL-E to do what it wants, but people less so.
(DIR) Post #Ac67sAEZ5uCZIl8RMG by hrbrmstr@mastodon.social
2023-11-23T12:18:51Z
0 likes, 0 repeats
@john i asked “Please create a realistic image of a fox with a human head/face.” and the resultant prompt it created was this (full text in alt-txt)
(DIR) Post #Ac67tI7ZFuCI3KOqAa by TEG@mastodon.online
2023-11-23T12:19:07Z
0 likes, 0 repeats
@john Enjoy this lovely and not at all horrible creation!
(DIR) Post #Ac685uMqsj4e9BZ5Yu by john@sauropods.win
2023-11-23T12:21:28Z
0 likes, 0 repeats
@hrbrmstr Yeah, I pasted the ChatGPT generated prompt upthread into Bing and got this. You need to write a short story to get what you want.Stable Diffusion still utterly fails though.
(DIR) Post #Ac689z4rJS0due5n8q by john@sauropods.win
2023-11-23T12:22:13Z
0 likes, 0 repeats
@TEG What AI is this? It looks Stable Diffusion-ish. What was the prompt?
(DIR) Post #Ac68Ory21vykCkzs92 by TEG@mastodon.online
2023-11-23T12:24:54Z
0 likes, 0 repeats
@john I have very little clue, it's from https://www.craiyon.com/, with the prompt: A realistic orange vulpine lammasu, with the body of a fox. The head belongs to Albert Einstein. The fox is orange. The head is smiling. The body is visible. The tail is bushy. The head is a scientist with his tongue out. The face is pink. The head is fully human.
(DIR) Post #Ac68jdYMoekxa8tYxc by zillophane@mastodon.online
2023-11-23T12:28:34Z
0 likes, 0 repeats
@john nature beat us to it: the Tibetan fox
(DIR) Post #Ac69dC8bvu72lsL8bI by john@sauropods.win
2023-11-23T12:38:42Z
0 likes, 0 repeats
@TEG Dall-e mini, apparently, which is an attempt to match Dall-e with an open source model. It's seems to pay more attention to the prompt than Stable Diffusion, which gave me this:
(DIR) Post #Ac69spPSCiLeuiflsu by john@sauropods.win
2023-11-23T12:41:31Z
0 likes, 0 repeats
@zillophane And God prompted, “let there be a fox, with the face of a fox, and yet of a man, so that no man knoweth why it unsettles him so" and so it was.
(DIR) Post #Ac6B0XltlWMITuPiue by zillophane@mastodon.online
2023-11-23T12:54:00Z
0 likes, 0 repeats
@john midjourney "a person whose body ONLY has been transformed into the body of a fox, but still with the face of a human" partial success
(DIR) Post #Ac6BlKhy9VwHZnt332 by john@sauropods.win
2023-11-23T13:02:32Z
0 likes, 0 repeats
@zillophane Eh, that's the old fox head on a human body, which Dall-e also really likes to do.Interesting the aesthetic aspect to this. There's more than your prompt going on.
(DIR) Post #Ac6BwSYxI0iN7eyanI by zillophane@mastodon.online
2023-11-23T13:04:30Z
0 likes, 0 repeats
@john some of the failures have been absolutely beautiful pictures of foxes in fancy clothes
(DIR) Post #Ac6CRlwTHtpkCfuHQW by zillophane@mastodon.online
2023-11-23T13:09:55Z
0 likes, 0 repeats
@john another failure "a photograph of a man caught roaming around in a realistic fox costume in the style of nature photography"
(DIR) Post #Ac6Cjsto9zyCzcIVge by john@sauropods.win
2023-11-23T13:13:30Z
0 likes, 0 repeats
@zillophane Your fox costume idea was good. Didn't work of course, but you never know.
(DIR) Post #Ac6DA4a7LdAikpZiSm by zillophane@mastodon.online
2023-11-23T13:18:13Z
0 likes, 0 repeats
@john this obviously didn't work, but it's still beautiful. Midjourney must have been really selective about their training set
(DIR) Post #Ac6DPQ3PnOfth5b2f2 by john@sauropods.win
2023-11-23T13:20:59Z
0 likes, 0 repeats
@zillophane I think they're adding stuff to your prompts to get stylised results.
(DIR) Post #Ac6DYETAEjhAN3qJbE by zillophane@mastodon.online
2023-11-23T13:22:34Z
0 likes, 0 repeats
@john maybe
(DIR) Post #Ac6DiTp60TXosxzIRc by BlueTurtleAI@hachyderm.io
2023-11-23T13:21:36Z
0 likes, 0 repeats
@john @pbloem @AggroBoy At least ChatGPT, because it is trained for that purpose. But it can only talk to dall-e, I guess other image generation ais don’t like the generated prompts very much.
(DIR) Post #Ac6DiUxHnVgqOepLqi by john@sauropods.win
2023-11-23T13:24:24Z
0 likes, 0 repeats
@BlueTurtleAI @pbloem @AggroBoy I thing Midjourney has a prompt-rewriter in there.
(DIR) Post #Ac6E1WfLf4TdCjVgiO by BlueTurtleAI@hachyderm.io
2023-11-23T13:27:53Z
0 likes, 0 repeats
@john @pbloem @AggroBoy Yes, I think they add in the background a few things to make the images look good. But imo this is not on the level of what ChatGPT does. ChatGPT really understands, what you want and translates it then for dall-e.
(DIR) Post #Ac6FQt3Nf5nDpePM6S by Twarda@sauropods.win
2023-11-23T13:43:33Z
0 likes, 0 repeats
@john Out of curiosity. Why do you need human faced foxes xD
(DIR) Post #Ac6Fcgj9JgYF4sZXCC by john@sauropods.win
2023-11-23T13:45:43Z
0 likes, 0 repeats
@Twarda Well it would save me a lot of time painting things like this
(DIR) Post #Ac6GaCjlwnaDRNF8T2 by Twarda@sauropods.win
2023-11-23T13:56:19Z
0 likes, 0 repeats
@john Lol that's valid
(DIR) Post #Ac6IHeqvOI4N2qsAJk by AggroBoy@mastodon.social
2023-11-23T14:15:30Z
0 likes, 0 repeats
@john yeah; I was surprised that it couldn’t do it. But with some playing around it seems quite bad at novel modifications to a subject. Putting it a novel context seems to work well, as does simple stuff like changing the colour or what something is made of. But (for example) “an elephant with three trunks” just didn’t work for me. I’m guessing, but perhaps because the training set didn’t include examples close enough for it to even have example of “three trunks” to extrapolate from?
(DIR) Post #Ac6JxPRr9bORNzcTrs by trachelipus@masto.ai
2023-11-23T14:34:17Z
0 likes, 0 repeats
@john I'm insufficiently motivated to spend my own money buying DallE credits in service of this question, but I wonder how it would handle the prompt "A sphinx with the body of a fox"
(DIR) Post #Ac6KHAmnNT13hfq04u by john@sauropods.win
2023-11-23T14:37:51Z
0 likes, 0 repeats
@trachelipus You asked for: soft-core furry porn. Vending:
(DIR) Post #Ac6LOrDm6RSvILgXtg by superflippy@mastodon.xyz
2023-11-23T14:50:30Z
0 likes, 0 repeats
@john as someone who has made a lot of AI art, yes: it’s really difficult for AI to make anything that’s uncommon or subverts the norm. I tried recently to make a woman wearing a corset like a hat. After many tries, this was the best I got: "A corset wrapped around a woman’s entire head, in the style of a vintage ad illustration "https://creator.nightcafe.studio/creation/WPh6b5iI16ELl2nxILmC
(DIR) Post #Ac6MksB3KRNuaLTQLQ by superflippy@mastodon.xyz
2023-11-23T15:05:40Z
0 likes, 0 repeats
@john SDXL is *terrible* at this, no surprise, even using inline prompt weights. It hates to buck convention.https://creator.nightcafe.studio/creation/P2wdOYS8LMzmiSAV4DVl
(DIR) Post #Ac6MllFuY4x1PLAa3M by trachelipus@masto.ai
2023-11-23T15:05:43Z
0 likes, 0 repeats
@john Furry porn indeed, lol. I was hoping it would know a sphinx has the head of a human and the body of a lion,and thus understand it was supposed to keep the human head while replacing the lion parts with fox. Instead it went full on Egypt while keeping the fox face. A centaur with the body of a fox isn't quite what you want, since I assume you don't want the human torso. Hmm. Interesting challenge.
(DIR) Post #Ac77ItOftc8e6b0NCi by akira28@mastodon.social
2023-11-23T23:47:16Z
0 likes, 0 repeats
@john first try in dalle-3 via chatGPT. Prompt: a fox with a human head and face. A bit creepy for the proportions, but I would say it’s a good job
(DIR) Post #Ac77dufON2jM7lqcfA by john@sauropods.win
2023-11-23T23:50:55Z
0 likes, 0 repeats
@akira28 Yes, if you can load the rest of the thread (unreliable on Mastodon, I know), you'll see that ChatGPT gives a more elaborate prompt to Dall-E, based on your prompt. If you use Bings version you need a fairly elaborate prompt to get it to work.
(DIR) Post #Ac85XZdfdJ5kQ6qsm8 by BlackPhi@geekdom.social
2023-11-24T11:02:06Z
0 likes, 0 repeats
@john Something like this? The trickiest bit was getting it so it doesn't look photoshopped. The inpainting was denoised at 0.9. I don't know if the Americanisations made any difference, I suspect not. There is cleaning up to do around the tail and legs, but that is to be expected with #stablediffusion. The initial model was SD's 768-v-ema and the inpainting used the Realistic Vision 5.1 inpainting model, masking the eyes and muzzle.
(DIR) Post #Ac8GDlSv3FFSnta11E by john@sauropods.win
2023-11-24T13:01:56Z
0 likes, 0 repeats
@BlackPhi Yeah, infilling is possible, but I guess what I'm interested in is not so much how to get a certain thing done (I can paint/photomanipulate/etc. myself), but what is going on with AI imaging in general .I'm finding the coaxing aspect interesting. It's just not as simple as the model doesn't have reference images, because DALL-e makes some things hard, but with some cajoling will actually do a pretty good job.
(DIR) Post #Ac8Ql2dfDn17AD70Ua by kromeboy@mastodon.uno
2023-11-24T14:59:47Z
0 likes, 0 repeats
@john Something like this?Stable DiffusionModel: ArtUniverse method: inpainting then image to image
(DIR) Post #Ac8RYGkxWMIV8dzvaS by john@sauropods.win
2023-11-24T15:08:47Z
0 likes, 0 repeats
@kromeboy I don’t think I painting counts. It’s too much like just compositing in photoshop.
(DIR) Post #Ac8TL1V784sFy4Ojlg by kromeboy@mastodon.uno
2023-11-24T15:28:55Z
0 likes, 0 repeats
@john by prompt alone I think that is really hard because the models are trained on man dressed as animals but not animals dressed as man 😄
(DIR) Post #Ac8TSBgpe443azIMfw by john@sauropods.win
2023-11-24T15:30:11Z
0 likes, 0 repeats
@kromeboy Yes, they are conservative. Dall-e doesn't want to do it, but you can make it by giving it a really long flowery prompt.