AI and endosomatophilia

[mastergts]

Midjourney (MJ) 7 alpha is out. I thought it would be nice to talk about it since this thread began with images from MJ3.



After using GPT 4o image gen, MJ7 is underwhelming. Do diffusion based image generators like MJ have a future or should companies start moving to transformer based image generators like GPT 4o image gen?

If you folks think OpenAI's censorship is excessive, MJ is off the charts. I got a rejection for wanting a zip on a wetsuit.

In today's post we're going to compare inside the stomach shots of MJ7, Reve and Hailuo Minimax. We already looked at GPT 4o image gen in an earlier post

Hailuo Minimax

Spoiler: show

This service is known more for its video generation but it recently made an image service available. This is my favourite image generator at the moment for inside the stomach shots. It's worth noting that it makes great inside the stomach videos too.

362377760338223113.png

364482070820409346.png

364482070820409345.png

These all look amazing was wondering what your prompts for the Hailuo inside stomach shots were?

[Nez]

These all look amazing was wondering what your prompts for the Hailuo inside stomach shots were?

The two prompts I used were:

Code: Select all

A lady in a white wetsuit inside a human stomach. She is trying to find a way out.

Code: Select all

lady in a white wetsuit with zip down the front, exploring the inside of a stomach, remnants of food being digested, pink moist walls, darkness, surreal, imaginative, anatomical setting, soft lighting, detailed textures

[KawaiiXaii]

I found it works really well to add custom dialog after an image is made.

There's lots of great image editing programs. You can crop, add text, and make small edits. This works well even if you don't have experience with drawing.

It's also good to "frankenstein" panels together. With an AI-generated page, there might be a panel that doesn't work here, but works great somewhere else.

That's not just an endosoma thing -- even in mainstream film and videos, leaving some things to the audience's imagination is a powerful technique.

Also, thank you for the reply Nez! That was a good tip with "imply when you can't show".

[Nez]

I don't know if this counts for endosomatophilia, but does anyone know what the best prompts to use for AI zoom-in videos?

There's a new service called Higgsfield that lets you zoom in to the mouth. Has some other cool camera controls that could be useful for endosoma story telling. You get two free generations per month.

https://higgsfield.ai/motion/7351a8ad-9 ... baf293d588

[zeslota]

I don't know if this counts for endosomatophilia, but does anyone know what the best prompts to use for AI zoom-in videos?

There's a new service called Higgsfield that lets you zoom in to the mouth. Has some other cool camera controls that could be useful for endosoma story telling. You get two free generations per month.

https://higgsfield.ai/motion/7351a8ad-9 ... baf293d588

Just as a quick test, I ended up using the model to do a zoom in transition. The image I used is Mizuhara Misono as she appears in MIMK-102 Fornication Teacher's Seibu Seikatsu Guidance Record Megumi Fujimiya Edition, and I followed the mouth zoom with a snip from the Body Story episode "Spreading Menace, using the clips where the (originally male) person swallows and the camera travels down into the stomach (though be warned the original episode focused on salmonella and food poisoning)

Anyways here's how it turned out https://mega.nz/file/qj42kBJQ#8fe56DNHiQbJoH8njTaxL9jazLIabXLRH6TF9tfcmeU

[mastergts]

I don't know if this counts for endosomatophilia, but does anyone know what the best prompts to use for AI zoom-in videos?

There's a new service called Higgsfield that lets you zoom in to the mouth. Has some other cool camera controls that could be useful for endosoma story telling. You get two free generations per month.

https://higgsfield.ai/motion/7351a8ad-9 ... baf293d588

Just as a quick test, I ended up using the model to do a zoom in transition. The image I used is Mizuhara Misono as she appears in MIMK-102 Fornication Teacher's Seibu Seikatsu Guidance Record Megumi Fujimiya Edition, and I followed the mouth zoom with a snip from the Body Story episode "Spreading Menace, using the clips where the (originally male) person swallows and the camera travels down into the stomach (though be warned the original episode focused on salmonella and food poisoning)

Anyways here's how it turned out https://mega.nz/file/qj42kBJQ#8fe56DNHiQbJoH8njTaxL9jazLIabXLRH6TF9tfcmeU

Nice video, also a fellow JAV man! There's definitely a lot of gorgeous big JAV mouths that will be perfect to use for this.

[mastergts]

These all look amazing was wondering what your prompts for the Hailuo inside stomach shots were?

The two prompts I used were:

Code: Select all

A lady in a white wetsuit inside a human stomach. She is trying to find a way out.

Code: Select all

lady in a white wetsuit with zip down the front, exploring the inside of a stomach, remnants of food being digested, pink moist walls, darkness, surreal, imaginative, anatomical setting, soft lighting, detailed textures

Nez thank you so much, that's very helpful!

[zeslota]

Nice video, also a fellow JAV man! There's definitely a lot of gorgeous big JAV mouths that will be perfect to use for this.

Thanks, was unsure how well it would look considering there's no higher quality footage for Body Story unfortunately.

And yeah, JAV is a bit of an untapped well for endo edits. Funnily enough, MIMK-102 and the hentai series it was adapted from contains a lot of endo stuff within the girls' wombs, which was kept for the live action at the mangaka's insistence. https://mega.nz/file/C6pXwDIC#7-BW8PAYmPPEWOarQChfjkzurqFrRNQoiuCnfIIs5yk

Unfortunately, it's built off of hypnosis and NTR so your mileage may vary on enjoyment.

[zeslota]

By sheer accident, I was somehow able to get Higgsfield to smoothly transition from mouth to throat for this https://mega.nz/file/qrJTEQiK#U1LufqWcoc12HnB1YKGk0p9UJ-l0bqfhLJjxCxq6aOw

[Nez]

Many have praised Kling 2.0 so I decided to purchase some credits and give it a go. Here are the results:

https://youtu.be/PJR17I2vWO4
https://youtu.be/SReerGp14A0

I burned a lot of credits trying to get growth and shrinking to work. I had to resort to using first frame and last frame in Kling 1.6 -- even then, the model tries to walk characters backwards and forwards or makes them walk up stairs to make characters appear larger or smaller.

This is my first go at doing video work so please go easy on me. I found capcut to be an amazing app for video editing. So intuitive.

[ProfessorGecko]

Another AI video site I've been trying out is Wan. The model itself is free and open source, but there's also a website you can use: https://wan.video/

I've been unsuccessful with trying to get it to do a proper mouth or nose zoom-in, but I do think it's also a good option for AI-generated videos.

For the most part, I've been getting my "quick endo stuff" through JanitorAI, a Character.ai alternative that I've used a lot for chatbot roleplay: https://janitorai.com/

[mastergts]

Many have praised Kling 2.0 so I decided to purchase some credits and give it a go. Here are the results:

https://youtu.be/PJR17I2vWO4
https://youtu.be/SReerGp14A0

I burned a lot of credits trying to get growth and shrinking to work. I had to resort to using first frame and last frame in Kling 1.6 -- even then, the model tries to walk characters backwards and forwards or makes them walk up stairs to make characters appear larger or smaller.

This is my first go at doing video work so please go easy on me. I found capcut to be an amazing app for video editing. So intuitive.

Great work!! The first video I love the part where the passengers are walking up inside that beautiful big asian mouth. The second video I love the production and zoom to mouth shot! Keep up the great work!

[mastergts]

Another AI video site I've been trying out is Wan. The model itself is free and open source, but there's also a website you can use: https://wan.video/

I've been unsuccessful with trying to get it to do a proper mouth or nose zoom-in, but I do think it's also a good option for AI-generated videos.

For the most part, I've been getting my "quick endo stuff" through JanitorAI, a Character.ai alternative that I've used a lot for chatbot roleplay: https://janitorai.com/

I'm wondering whether a LORA model would be best to get the effect, running locally.

[Nez]

Great work!! The first video I love the part where the passengers are walking up inside that beautiful big asian mouth. The second video I love the production and zoom to mouth shot! Keep up the great work!

Thanks for the feedback! It's helpful to know what people are enjoying. The passenger stairs was a test of Kling 2.0's camera control. I was actually surprised that it rendered the passengers moving into the mouth because I was expecting it to get censored. I was bit disappointed about the smoothness in the giantess's face but I guess that's to be expected since the starting image had that airbrush illustrated look.

The Boss Lady video clip was something I made with left over Kling credits. In fact, once I got going, I was starting to struggle with having not enough credits.I used Higgsfield to pull off the mouth shot. It's the best available at the moment, in my opinion. Here's another quick one: https://www.youtube.com/shorts/PQWiGtwOtT0

I feel a bit dirty uploading the above YouTube short because it was as simple as uploading an image and clicking "generate". It reinforces the "AI art is low effort" narrative but I do think it's a valuable tool when incorporating it in storytelling.

[ProfessorGecko]

I've been trying some of Hailuo's different camera panning presets, and when I figured out that you can combine more than one camera pan, I knew I had to experiment with it.

I mostly combined the "Push In" and "Zoom In" preset options, and they go for nice close-ups, but not right enough for a mouth, nose, or eye zoom-in.

Also, I'm trying out start to end frames in Wan. I think zoom-ins could work with that AI model using a starting frame and an ending frame along with a good enough prompt, but so far I'm just getting cursed stuff.

Just thought those observations would be nice to share in case if anyone wants to try out Hailou or Wan.