Ai Generated content

Collapse
X
 
  • Time
  • Show
Clear All
new posts
  • LoonPopper
    Senior Member
    • Aug 2011
    • 772

    #46
    Originally posted by frankfrank

    So, to somebody, even a watermelon could be porn, then?

    Watermelon fetishism appears in several forms, ranging from the sexualized crushing of the fruit (often by female bodybuilders) to the use of hollowed-out melons for sexual pleasure. It also includes online content focused on "watermelon inflation" art, where characters are depicted as swelling to resemble the fruit. [1, 2, 3, 4, 5, 6]
    • Crushing Trend: A notable subset, popularized by individuals like Kortney Olson, involves crushing watermelons between the thighs, which is often consumed as a strength or fetish video. [1, 2, 3, 4]
    • "Watermelon Sex" Trend: This involves using the fruit as a, often, improvised sexual aid. Such videos are popular on some social media platforms. [1, 2]
    • DeviantArt Watermeloninflation: A niche digital art genre focusing on characters inflating or transforming into watermelons. [1, 2]
    • Metaphorical Use: In some contexts, particularly in Reddit discussions, the watermelon emoji (🍉) or the word "melon" is used as slang for breasts. [1]
    For those looking for information on these topics, the VICE article provides context on the "freaky" side of these videos. Discussions on Reddit provide user-driven commentary on the subject. [1, 2]

    Blow, Baby, Blow...when will it pop nobody knows!

    Comment

    • O'Dragon
      Senior Member
      • Jan 2024
      • 298

      #47
      I am at the moment tinkering with z image turbo and qwen models and have to agree it’s definitely difficult if you have a very specific vision in mind.
      So I’m using a mixture of generating images directly in comfyui and then using special workflows to assemble different images together to get the precise characters and backgrounds.

      been starting with videos by using ltx and image to video to animate the images. Working okayish as long as you do t want specific camera movements.
      It kinda works for some things but getting a character to hump a balloon or do some stuff to it I guess you def need to do motion transfer workflows where you use a real video as reference for the motion.

      I guess then you can create something quite unique and also aesthetically pleasing.

      when I got something finished I can post it here if you are interested.
      Its mainly NonPop with a gay male focus so far as I am missing such content a lot actually but if I find something that works well I’m planning on expanding that to pop and also female characters.

      but could take a while still as I have to find the time and muse inbetween my job and weekends to work on that stuff.

      Comment

      • Helium Guy
        Senior Member
        • Dec 2019
        • 256

        #48
        ChatGPT came up with a design for my Balloon Barn.
        Attached Files

        Comment

        • CleaLooner
          Junior Member
          • Dec 2023
          • 6

          #49
          I trained a few motion LORAs for Wan 2.2 (a video model you can run at home). Not that it probably matters to the haters, but the training material is 95% from footage I shot myself with paid models years ago Anyway, here's a short compilation of test clips. Sound was added later (wan 2.2 doesn't generate audio): https://drive.google.com/file/d/1mVH...ew?usp=sharing. Switch to 720p for better quality.

          I put the b2p & stomp-pop models up on civitai: https://civitai.com/user/Asteria83 (not an ad - these are free to download if anyone wants to try them out). I have sit-pop and hug-pop models that I haven't put up yet, but I'm just looking into LTX 2.3 now so maybe I'll just retrain everything for that.

          Would *love* to chat with anyone interested in this (DM me!).

          Comment

          • thepopper90
            Member
            • Feb 2024
            • 36

            #50
            Originally posted by CleaLooner
            I trained a few motion LORAs for Wan 2.2 (a video model you can run at home). Not that it probably matters to the haters, but the training material is 95% from footage I shot myself with paid models years ago Anyway, here's a short compilation of test clips. Sound was added later (wan 2.2 doesn't generate audio): https://drive.google.com/file/d/1mVH...ew?usp=sharing. Switch to 720p for better quality.

            I put the b2p & stomp-pop models up on civitai: https://civitai.com/user/Asteria83 (not an ad - these are free to download if anyone wants to try them out). I have sit-pop and hug-pop models that I haven't put up yet, but I'm just looking into LTX 2.3 now so maybe I'll just retrain everything for that.

            Would *love* to chat with anyone interested in this (DM me!).
            This video is really good. It is probably the best AI generated balloon popping "effect" I have ever seen. How did you generate it? Can you run it at home on a regular PC? I have tried to generate images locally and haven't had much success.

            Comment

            • CleaLooner
              Junior Member
              • Dec 2023
              • 6

              #51
              Originally posted by thepopper90

              This video is really good. It is probably the best AI generated balloon popping "effect" I have ever seen. How did you generate it? Can you run it at home on a regular PC? I have tried to generate images locally and haven't had much success.
              Thanks! Yeah, these were generated on my PC with comfyui and the wan 2.2 video model. I think the absolute bare minimum VRAM required is 8GB. The initial learning curve is a little steep, but there are tons of great tutorials out there.

              Wan 2.2 barely understands the concept of balloon popping on its own. I put up a sample earlier in this thread, but here is a compilation of a bunch of my early experiments: https://drive.google.com/file/d/1YlM...ew?usp=sharing. Balloons kind of break/shatter lol.

              Luckily you can train it with new "concepts" (props, people, actions/motions) since you have full control running it at home. These are the LoRAs I mentioned in my last post.

              As for how I generated these exactly, I created a starting image (e.g. woman in a photo studio with balloons strewn about, etc) - you can use nano banana, or a local image model in comfyui like z-image. Then I used Wan 2.2 to turn the starting image into a video. Using the trained LoRAs, I can just prompt it with the training keyword, like "The woman does a stomp2pop on the green balloon, then laughs and points to the camera".

              Comment

              • thepopper90
                Member
                • Feb 2024
                • 36

                #52
                Originally posted by CleaLooner

                Thanks! Yeah, these were generated on my PC with comfyui and the wan 2.2 video model. I think the absolute bare minimum VRAM required is 8GB. The initial learning curve is a little steep, but there are tons of great tutorials out there.

                Wan 2.2 barely understands the concept of balloon popping on its own. I put up a sample earlier in this thread, but here is a compilation of a bunch of my early experiments: https://drive.google.com/file/d/1YlM...ew?usp=sharing. Balloons kind of break/shatter lol.

                Luckily you can train it with new "concepts" (props, people, actions/motions) since you have full control running it at home. These are the LoRAs I mentioned in my last post.

                As for how I generated these exactly, I created a starting image (e.g. woman in a photo studio with balloons strewn about, etc) - you can use nano banana, or a local image model in comfyui like z-image. Then I used Wan 2.2 to turn the starting image into a video. Using the trained LoRAs, I can just prompt it with the training keyword, like "The woman does a stomp2pop on the green balloon, then laughs and points to the camera".
                I have tried generating some images locally with fooocus and it didn't turn out quite as I expected. Mostly I think because the models I used were not trained with the correct material.

                For exemple, when I tried to generate a picture of a woman sitting on top of a large balloon, the deformation of the balloon would be incorrect. The texture of the balloon would also be strange.

                I never played with AI video creation but I watched some videos online and usually the pop has this shattering effect like you had on your first experiments. Your videos from the other post have a lot more natural behavior of the pop. I believe that is because you trained your LoRA with proper inputs.

                How did you train your own LoRAs? I'm really interested in this subject, although I don't have much time to work on it. If you’d be willing to send me the material you read to do this, I’d be grateful. Feel free to send me on my DMs or something.

                I'm not at home right now, so I can't try the software. Do you believe it would be possible to run this on a Radeon MI50 with 16GB of VRAM? I tried some LLM and the picture creation on fooocus and it handled it.

                Comment

                • CleaLooner
                  Junior Member
                  • Dec 2023
                  • 6

                  #53
                  Originally posted by thepopper90

                  I have tried generating some images locally with fooocus and it didn't turn out quite as I expected. Mostly I think because the models I used were not trained with the correct material.

                  For exemple, when I tried to generate a picture of a woman sitting on top of a large balloon, the deformation of the balloon would be incorrect. The texture of the balloon would also be strange.

                  I never played with AI video creation but I watched some videos online and usually the pop has this shattering effect like you had on your first experiments. Your videos from the other post have a lot more natural behavior of the pop. I believe that is because you trained your LoRA with proper inputs.

                  How did you train your own LoRAs? I'm really interested in this subject, although I don't have much time to work on it. If you’d be willing to send me the material you read to do this, I’d be grateful. Feel free to send me on my DMs or something.

                  I'm not at home right now, so I can't try the software. Do you believe it would be possible to run this on a Radeon MI50 with 16GB of VRAM? I tried some LLM and the picture creation on fooocus and it handled it.
                  I haven't kept up with Radeon support in modern AI tools. A quick search looks like it may be somewhat supported now? If so, 16GB is great!

                  Yeah, to get really good results with both video AND image generation, you typically need to use LoRAs. For example, there are a few popular balloon blowing LoRAs for nearly all the popular image gen models. You typically won't get super realistic "balloon physics" from any model 'out of the box', even the major paid ones. But I've seen some great images posted in this thread, so you can definitely get lucky.

                  I've also gotten excellent results by mixing in additional reference images in nano banana pro ("The model is sitting on the balloon, which is deforming similar to the balloon in the second image", etc). You might try that, keeping in mind it's all (very fun) experimentation I'll try to attach a couple of images I did in nano banana pro today with a consistent model and a B2P reference image.

                  Happy to DM more detailed info/tutorials re: the training.
                  Attached Files

                  Comment

                  • Guest

                    #54
                    Something that, so far, no AI has managed to do convincingly, is create symmetrical balloons with necks - like old Qualatex, the shape of a light bulb. Every once in a while it gives a good result, but normally, no matter how detailed the prompts, the balloons in the picture are nowhere near 16 inch, look distorted, or the neck even appears in a different place on the balloon (which is quite a bizarre sight). If anyone already has found reliable prompts for this, please let me know!

                    Comment

                    • Helium Guy
                      Senior Member
                      • Dec 2019
                      • 256

                      #55
                      Keep On Inflatin'

                      Inspired by many conversations about balloons, wonder, and never forgetting to look up. Created with the creative assistance of ChatGPT 5.5.
                      Attached Files

                      Comment

                      • Helium Guy
                        Senior Member
                        • Dec 2019
                        • 256

                        #56
                        This is the start of Images of my Balloon Sanctuary - A History of Joy

                        Inspired by many conversations about balloons, wonder, and never forgetting to look up. Created with the creative assistance of ChatGPT 5.5.
                        Attached Files

                        Comment

                        • ArchibaldR
                          Member
                          • May 2021
                          • 62

                          #57
                          I'm a massive sucker for advertising balloons and if you write your prompts in just the right way, you'll get some amazing results when using Google Gemini. I'm not the kind of guy that's pursuing porn-ish content through AI, i'm really trying to generate images that could happen IRL if the world wouldn't try to be so environment friendly. Here are some of my favourite results, really curious about what you guys think about these kind of images:
                          Attached Files
                          Last edited by ArchibaldR; 04-07-2026, 18:24.

                          Comment

                          • TooBigToFail
                            Member
                            • Feb 2019
                            • 30

                            #58
                            We should start an AI content Discord. InflateRUs is very hostile to AI content (even for mockups)

                            Comment

                            • TooBigToFail
                              Member
                              • Feb 2019
                              • 30

                              #59
                              Originally posted by The1
                              Something that, so far, no AI has managed to do convincingly, is create symmetrical balloons with necks - like old Qualatex, the shape of a light bulb. Every once in a while it gives a good result, but normally, no matter how detailed the prompts, the balloons in the picture are nowhere near 16 inch, look distorted, or the neck even appears in a different place on the balloon (which is quite a bizarre sight). If anyone already has found reliable prompts for this, please let me know!
                              I've had great success with Grok

                              Attached Files

                              Comment

                              • kahuna09
                                Senior Member
                                • Jul 2013
                                • 126

                                #60
                                Originally posted by ArchibaldR
                                I'm a massive sucker for advertising balloons and if you write your prompts in just the right way, you'll get some amazing results when using Google Gemini. I'm not the kind of guy that's pursuing porn-ish content through AI, i'm really trying to generate images that could happen IRL if the world wouldn't try to be so environment friendly. Here are some of my favourite results, really curious about what you guys think about these kind of images:
                                this is some stunning content wow super nice.

                                Comment

                                Working...