Generate Pictures / Images from text

Comments: 80

• 20 Jan, '20
  Joaquin Phoenix

  Uh, have you seen the pedo/perverted stories people have made? Pictures of those would be disastrous.

• 14 Feb, '20
  Karim

  The game would be HUGE but not an impossible task..

• 31 Mar, '20
  World Conqueror

  Perhaps if the AI generated simple and interchangeable illustrations.

• 28 Apr, '20
  Ugh don’t know what to right here

  Oh yeah and you t would be a command like /picture or like a setting that you could turn on or off and it would do a illustration at the top of the screen at the beginning it would be blank but the more you go into the story it would get more and more detailed.

• 28 Apr, '20
  Nikita R. Merged

  you can create an animation from diologs for example: "I took the sword and shield in the altar of the dungeon" there is a picture of the dungeon then our character and then the way he takes the sword and shield. it may be difficult but you can pre download images that represent the altar and the sword for example and the AI will use them to generate animation
  
  (Sorry this is google translator and messages for developers)

• 30 Apr, '20
  N

  I hope that this happens since it'll be pretty cool

• 08 May, '20
  Liyar Merged

  I'think it well be too nice with images!

• 09 May, '20
  Gonçalo Merged

  That will be amazing! I hope they plan it but you have to considerate that it will be very hard to do that kind of stuff,but what do i know about A.I?
  :)

• 19 May, '20
  Devon Admin

  "A.I generated images?" (suggested by Seth on 2020-04-24), including upvotes (24) and comments (3), was merged into this suggestion.

• 19 May, '20
  Devon Admin

  "Story Pictures" (suggested by andyman11 on 2020-05-02), including upvotes (6) and comments (0), was merged into this suggestion.

• 20 May, '20
  Devon Admin

  "Drawing mode" (suggested by Layla Rose on 2020-02-17), including upvotes (1) and comments (1), was merged into this suggestion.

• 06 Jun, '20
  Devon Admin

  "Image generation. Background and characters" (suggested by Kandji on 2020-06-06), including upvotes (1) and comments (0), was merged into this suggestion.

• 07 Jun, '20
  Ichibi

  This would be insanely difficult and demanding on your system. Even transforming a pre-existing image takes a high end gpu, so this would greatly reduce compatibility especially with mobile platforms.

• 16 Jun, '20
  tagoya

  How about you add your own image? You could create your own image and add to the context of the story.

• 17 Jun, '20
  Sentrys

  Like Tagoya said. You could always just add a feature for the scripts or worldbuilding section, where whenever a person, thing, or item is mentioned the program shows you an image that you uploaded beforehand. Otherwise, peepz, just use Ascii art. There are converters out there that allow you to convert images to ascii if your willing to go through the effort. It'll look more like DF.

• 18 Jun, '20
  Frosthail

  This could be possible with a new technology introduced called "StyleGAN2". At least for characters.

• 20 Jun, '20
  Oleg.

  NSFW Pictures.. Mhm.. Seems like my imagination going on the next level of lewdness..

• 24 Jun, '20
  Devon Admin

  "Images flash up of things your character does or witnesses" (suggested by Lukasthewookuspookus on 2020-06-23), including upvotes (1) and comments (0), was merged into this suggestion.

• 03 Jul, '20
  Devon Admin

  "put images" (suggested by Sebastian Escobar on 2020-07-02), including upvotes (2) and comments (0), was merged into this suggestion.

• 03 Jul, '20
  Devon Admin

  "Make the idea give detailed images based on the description of your stories" (suggested by Thegeneral74 on 2020-07-02), including upvotes (2) and comments (0), was merged into this suggestion.

• 07 Jul, '20
  nichsu32 Merged

  please comment

• 07 Jul, '20
  Esile Merged

  Comment: I do not understand

• 09 Jul, '20
  Devon Admin

  "generated image and summon new command too" (suggested by nichsu32 on 2020-07-07), including upvotes (1) and comments (2), was merged into this suggestion.

• 11 Jul, '20
  Usotsuki

  What i'd like is a crossover to Artbreeder, where you can create Portraits or Landscapes with genes (Blond Hair Slider, Black Hair Slider, Sex Slider, Age Slider etc.) and get a few random sample who range for hideous monster to absolutely stunning.
  If the AI could convert your characters looks into slider ranges and then create a few pictures for you where you can pick what you like. That would be quite cool.

• 14 Jul, '20
  flarn2006

  @Joaquin Phoenix: I'm not sure I see the issue; if I'm understanding correctly this is about AI generating the images, meaning there's no actual victim being abused to create them. And who knows? This harmless outlet might even satisfy someone enough that they don't feel the need to actually hurt someone, when they otherwise might.

• 18 Jul, '20
  Guedez

  As much as I like the idea, I know very well that only nightmares awaits us: https://imgur.com/a/5dKuG0v

• 22 Jul, '20
  Uddin

  I thing this is bad idea. I like "text based" game because i only show text. So, i can do whatever i want without saw it. Can you imagine if you do something cursed and the ai show you that imahe and you cant erase this memory from your mind? If you make this feature, i hope you have button to disable it. Thank for your hard work. Have a nice days :)

• 23 Jul, '20
  throwawayexperimentation Merged

  If you crowd-sourced images and image labeling, you could construct a model that learns how to generate images from text, and then feed that image back into itself during generation. You could use it to give GPT-2/3 'visualization skills'.

• 23 Jul, '20
  Devon Admin

  "Add ImageGPT so we can visualize stories like a picture book." (suggested by Yoshi on 2020-07-23), including upvotes (1) and comments (1), was merged into this suggestion.

• 23 Jul, '20
  Devon Admin

  "Some Photos or something like this" (suggested by Anonimous on 2020-07-23), including upvotes (1) and comments (0), was merged into this suggestion.

• 26 Jul, '20
  yes

  yes yes yes yes

• 26 Jul, '20
  Exodus

  If would LOVE this feature to be included.
  It would make all my game experience truly complete.
  I'd definitely pay for this.

• 31 Jul, '20
  Shredder

  If the images were taken from Google and it used one of those jpg to ASCII generators this could work.

• 04 Aug, '20
  Devon Admin

  "generate picture from action background+caractere(+foes)" (suggested by Basileus on 2020-08-03), including upvotes (1) and comments (0), was merged into this suggestion.

• 09 Aug, '20
  Nobody

  It's cool 'cause it can provide more imagination to the stories, and whether the picture is 18+ or not can depend on whether you turn on NSFW or not so the age problem would not be a bid deal though. Beside, if AI can respond the player by their command, so providing the picture won't be too hard as the providing picture also depends on the player's command (although looking at the accuracy of the AI Generator right now is kinda sucks lol )

• 11 Aug, '20
  YoonbumIsGay

  I hope they are disabled for NSFW content,but it's gonna be great especially if it's not for prime content

• 12 Aug, '20
  Devon Admin

  "AI pulls images related to text" (suggested by John Dole on 2020-08-11), including upvotes (1) and comments (0), was merged into this suggestion.

• 12 Aug, '20
  Devon Admin

  "one graphic interface per image" (suggested by Naguivaï on 2020-08-12), including upvotes (1) and comments (0), was merged into this suggestion.

• 15 Aug, '20
  CyborgJiro

  Cool idea but how would that be effectively implemented? Id prefer them to keep working on making the writing better. But.. i won't complain if they somehow figured this out.

• 16 Aug, '20
  Devon Admin

  "Pictures to storylines" (suggested by Jasper on 2020-08-15), including upvotes (1) and comments (0), was merged into this suggestion.

• 16 Aug, '20
  Fuguer

  AI Dungeon + Deep Dream Images would be world changing, but probably way too computationally expensive for now.

• 17 Aug, '20
  jackomix

  I think it'd be best to have an image library. There could be two layers, background and foreground. Let's you enter a cave, some system tries to guess what location you're in and sets the background accordingly. Then if you sit on a chair, it'll set the foreground to a picture of a chair. Generating pictures is something that's only starting to be tackled let alone something that's practical enough to use. An AI or even a system to pick images from a library would be way easier.

• 19 Aug, '20
  Devon Admin

  "images of certain things with the ability to turn off images" (suggested by red_demon098 on 2020-08-18), including upvotes (2) and comments (0), was merged into this suggestion.

• 23 Aug, '20
  Malcovitch

  Personally I think it's better without images. Engages the imagination much more!

• 23 Aug, '20
  Saturnis

  I vote for ASCII representations of things

• 01 Sep, '20
  Liburia

  This would be immense for immersion.

• 04 Sep, '20
  Devon Admin

  "Teach Ai Binary to Ascii translation in order to develop images" (suggested by Nachofluid on 2020-09-04), including upvotes (1) and comments (0), was merged into this suggestion.

• 08 Sep, '20
  cocoyoc Merged

  i know the request for AI to generate images has been talked about a lot. what if it could happen BUT if you use that feature, it cannot be submitted to the public gallery?

• 08 Sep, '20
  Devon Admin

  "pictures for stories than cannot be published?" (suggested by cocoyoc on 2020-09-08), including upvotes (0) and comments (0), was merged into this suggestion.

• 11 Sep, '20
  Hellolol

  Yes! I agree!

• 17 Sep, '20
  Bruhman125

  there should be a switch that enables or disables NSFW pictures, it would be off for default.

• 01 Oct, '20
  Kasparus

  If they add the generation of a picture from the context of a text, then it will be a technological breakthrough, a game that gives absolutely everything, this game will have a huge future.

• 03 Oct, '20
  Devon Admin

  "Generate related images based on the text input and output." (suggested by Nobody. on 2020-10-02), including upvotes (1) and comments (0), was merged into this suggestion.

• 03 Oct, '20
  Yakuza_Suske

  It would be a very huge addition to the AI.

• 08 Oct, '20
  RyQueezy

  Devon, WE NEED THIS

• 11 Oct, '20
  Zero

  I don't think GANs are advanced enough today to "The ai adding pictures to whatever is currently happening"(this is the most advanced afaik: https://www.technologyreview.com/2020/09/25/1008921/ai-allen-institute-generates-images-from-captions/ ), but I think it would be possible to add backgrounds and perhaps character pictures with artbreeder.

• 17 Oct, '20
  Lxanr

  Consider this https://twitter.com/shivkanthb/status/1302039696692789249

• 17 Oct, '20
  Devon Admin

  Eh, that example is only for human faces, this suggestion is much more broad

• 19 Oct, '20
  ranji

  Is it really possible to generate image from text? I have seen some AI things on the internet that generate images from text, but all of them generate nonsense/weird stuff. However, it would be a good idea (if possible.)

• 19 Oct, '20
  CubicPulsar

  As long as it can’t generate NSFW images.

• 20 Oct, '20
  CubicPulsar

  How about some text from an image?

• 20 Oct, '20
  Angular

  This is a very good suggestion. I hope we eventually get this.

• 21 Oct, '20
  adriene

  I agree. I want to know what the AI thinks is happening so I can make changes/etc

• 21 Oct, '20
  Devon Admin

  "An option to choose whether or not you want pictures generated by the text" (suggested by Mr. Ryry on 2020-10-21), including upvotes (1) and comments (0), was merged into this suggestion.

• 29 Oct, '20
  Luke

  One that may be computationally less expensive is to start with some pre generated landscapes based on your world-land-town description. Slowly add a moderated library of tagged (possibly generated) images based on real life and scenes similar to magic the gathering/scifi/whatever landscapes. It probibly wouldnt take long to get thousands of varieties with the wierd or otherwise broken generated images downvoted.
  Just a starting place, once that feels like its working, move on to portraits, etc...

• 02 Nov, '20
  Tommy

  Damn i can picture falcon punching somebody as Nartuo

• 08 Nov, '20
  alesz

  They'd probably get a Turing Award if they managed to do it. This tech doesn't exist yet.

• 09 Nov, '20
  yae

  I don't think AI generated images will work as it requires quite a bit of power and images from the internet chosen BY the AI probably won't either as the results may vary but that could still be viable. You could possibly feed a certain selection of images of your choosing and assign them by certain words or categories and then use something akin to the world system where you use tags and whenever a word is mentioned in the story that shares the same tag or context, the AI will use one of those images you fed it earlier, assuming you added one. Or just stick to letting it scramble the internet for content, maybe a booru ( ͡° ͜ʖ ͡°)

• 27 Nov, '20
  Nationality

  Please no. All the creepy nsfw this would generate would attract all kinds of unwanted attention from corporations and governments, who would demand restrictions on AI output.

• 11 Dec, '20
  AWizardOrSumthin

  I know I'm not exactly bringing anything new to the debate with my first point so I'll keep it quick, lewd. So here is my next point, how exactly? What would it even look like? If you come up with an interesting concept you need to elaborate. I don't want to sound like I'm tooting my own horn but I suggested Character logs, an easy concept to grasp with a concept image to go with it, I think we can all agree this is good. Your suggestion just brings up questions and you don't answer any of them. (https://aidungeon.featureupvote.com/suggestions/138451/character-logs)

• 18 Dec, '20
  Gabusan

  I would rather be able to upload a picture of my choice to stories/scenarios I publish.

• 27 Dec, '20
  Skrubtimer

  I think if an AI would train on just one specific artstyle at first and then gradually support more and maybe custom art styles the AI could be trained with, generating images may work.

• 31 Dec, '20
  RogueLikely

  I'm pretty sure there is already a good deal of AI generated portraits character creators, like that site thispersondoesnotexist.com for example. I'd think implementing an tool that created AI generated faces from pre-determined details would be possible. Like age, hair color, eye color/shape etc.

• 31 Dec, '20
  Jacob

  Would be an ambitious project. But if the dev could generate a renpy type picture story line that would be amazing.
  
  You could technically extract the pictures from as many renpy games as possible and then feed those to the AI in relation to certain text and then have the AI generate pictures to text occordingly.
  
  But yes would take a ton of work, but would be amazing.

• 03 Jan
  Beta tester

  @Joaquin Phoenix
  
  The solution could be to use a pattern recognition system, which recognizes parts of the body as "Google images" and "FotoForensics" do (which blocks the scan of images with certain parts of the body.)

• 06 Jan
  E

  Add support of "DALL·E: Creating Images from Text" to upper subscription tier.

• 06 Jan
  flarn2006

  This suggestion suddenly seems a LOT more likely to be doable in the near future...

• 07 Jan
  EaglePhntm

  oh boy this would be great and well i think AI wouldnt be able to find deep web pictures anyway since they dont really appear at any search engine nor their content does so you need to have the links to those websites specifically. I dont think thats a worry.

• 07 Jan
  EaglePhntm

  the images could be inaccurate or odd though since its an AI, odder than some of the stories generated maybe lol

• 08 Jan
  Somebody

  Please investigate the following research paper and look at the examples on the following website.
  
  https://openai.com/blog/dall-e/
  
  This site describes an AI-based technique for generating a photo (often a photorealistic photo) based on a textual description. It is surprisingly effective. For instance, one of the text descriptions is "an illustration of a baby daikon radish in a tutu walking a dog", and the software is successfully able to generate an image from this. The technique described on this page seems like a natural fit for AI Dungeon.