Exploring new video formats v2
Added 2025-07-11 17:07:08 +0000 UTCPlease read the previous post first for context.
My long-term goal is to create a "perfect" fit between narration and visuals, where there's one image for every word, and the visuals always align with the narration.
Of course, that's easier said than done. This video is that idea applied to a more "storytelling" format in which an encounter is narrated to you. I don't know what I can do with that, but it seemed cool to try.
This is a test. There are too many words and not enough images. The images don't exactly fit the sounds, but you'll get an idea of a possibility.
Again, it would be extremely helpful to get ideas and feedback. If you don't want to comment, please feel free to reach out privately in the chat.
Comments
I am pretty sure it's 99.9% impossible to create an arbitrary hypnosis script and be able to create good narration visuals. What I think is possible, though, is that if I consider what the model generates well, and think of narration that fits its capabilities. With that in mind, what "kind" of narration would be entertaining? Would it be similar to the video, featuring a scene between you, the viewer, and a woman? Would it be a JOI + scene mashup where you get instructions on what to do while the woman is executing the instructions on you as the viewer? Something else?
Sect
2025-07-12 10:50:22 +0000 UTCI think 1 image per word is a bit fast, but I'd be curious to see it if the images followed the narration. Love the concept of having the images following the narration though.
Nestle Carnation
2025-07-11 19:45:48 +0000 UTCThat's kind of the plan, but it seems almost impossible. I've tried about 5 times to get "script" -> "image prompt," but it takes everything literally, and the image quality isn't great compared to when I focus on the image. The new idea that comes to mind is to consider visuals and script simultaneously, or at least incorporate a small part of the hypnosis script into the visuals. For example, in the next black chamber audio, the script gives you a headset that "arousing images" over your eyes in one sentence. I should probably breach the technological challenge with creative storytelling. That's my conclusion for now. At the same time, I chip away at it. But yeah, if I go the word route, I'll do something like you said, have an image not even per 2 lines, but per word, literally for each word of a script, generate a prompt
Sect
2025-07-11 19:23:06 +0000 UTCI wonder if you could automate the image generation for this. e.g. to generate a prompt for every 2 line of hypno text. The prompt would probably be something like: 1) {some preamble for the context describing what's on background, the girl and the point of view} + 2) {previous 2 lines of hypno text for the AI just for additional context} + 3) {current 2 line of text which we need to illustrate}
progamoto
2025-07-11 18:17:55 +0000 UTC
