I am prefacing this with a note that I am still not generating massive amounts of AI porn anymore. However, I rebuilt my process and the new process is pretty spiffy. Use Phr00t's Qwen-Image-Edit-Rapid-AIO to generate and tweak images. If you bypass the image inputs and just put in a prompt it will do T2V, then you can take the image it creates and load it into one of the image inputs and tweak it with real language. Say "Make her breasts bigger" and it will keep everything else in the image the same, but increase the size of her breasts. Say "Make her breasts gigantic" and it will. Say "Make her breasts big enough to fill her lap" and it will. It is great!
So you can use Qwen to make "keyframes". The first image and the last image of your video you want to see. So you can take the start image of a woman, tell Qwen to make her breasts big enough to reach her bellybutton and make her shirt ripped open and you get your final frame.
Next, use Phr00t's LTX2-Rapid-Merges with his workflow to do I2V. Set the X and Y to be 768 (same as the Qwen output), set your number of frames to 241 (for 10 seconds) or 481 (for 20 seconds, but the last second will be jacked up), load in the first frame and last frame, set up your actions and dialogue, and you are off to the races. I am looking to do longer videos with small clips like that spliced together. It will take me a long time to generate one, but I want to try and do more "story" type breast expansion videos with dialogue. I may upload one or two videos as examples of what you can do. FYI, LTX-2 will try to cram all of your dialogue into the time you allow for it. So if you find your people are talking too fast, extend the video or break it into more than one clip. Again, don't try to go beyond 20 seconds or they get really crazy.
See my Mega. I have uploaded an attempt at something I am working on. Blind date with a girl who has a condition that makes her breasts swell. I need to work harder on my prompting because I suck with dialogue. Tone and emotion need help. But you can see it has a jump cut, multi-part dialogue, etc. You can really build a good "story" by splicing video clips together, using the same seed so it always gives the girl the same voice. If you use a random seed her voice can change. It is pretty good at giving people the same voice, so I guess AI has an algorithm that estimates what a person sounds like by their looks. I will also upload some examples of what you can do with Qwen.
Punchy