Mat Exposure

NOTE: if you came here from Reddit, the above video is the director’s cut which adds one additional scene.

Currently we’re working on a new project with Eva and the War Bunnies called Retro Rabbit. It’s almost done. But, I got distracted by other things. Here’s the latest distraction.

---

I like strong women.

I talked about that in “Thin Blue Lines:”

Thin Blue Lines and the Ladies of the Nueva Vista Police Department

Part of the appeal for me is the conflicting views on the concept of femininity and how strength figures into that – and the way in which women are able to be beautiful and strong in a way that feels subversive. And that subversive nature is often intrinsic to erotic art – erotica tends to be subversive because there’s an illicit thrill in the idea of seeing something that we know we’re not supposed to – of enjoying things that we’ve been told are wrong – fantasies about power and sex and things that are politically incorrect.

The female body is often the focus of these subversive fantasies – because we fantasize about the things we’ve been taught to value – power, beauty, and sex.

And that’s why wrestling makes sense as an exploration of the female body – because there is no other sport that is so purely focused on the limits of the human body. It is the toughest, most physically demanding sport in the world – but it’s also a very intimate sport. The action happens between two people on a small mat exploring the limits of their bodies together, seeing who will dominate and who will submit.

And as a fantasy…that’s pretty hot.

Because we don’t always get to see women in that way – pure and raw and physical – because it’s not always politically convenient. Which is a shame, because it’s beautiful.

And that’s what Mat Exposure is about – that fantasy of femininity and how it can’t be quantified.

---

Like most things, this started as something simple. The prompt was:

A highly detailed and realistic sports photograph, 8k, high quality, high resolution, Olympus OM AESTHETIC. Two nude females with realistic athletic bodies and hair are doing collegiate wrestling on a mat in front of a crowd of spectators

On the SPEED setting Grok oddly has no issue with this:

Meanwhile, the QUALITY setting seems to have a problem with the prompt:

The QUALITY output was not that great anyway – definitely not worth the longer generation time. And that’s something that has become more clear as time has gone on and I’ve had a chance to use SPEED and QUALITY more.

SPEED is the same Grok image generation model that we’re used to – and it does a great job of creating realistic looking images that have prompt adherence with respect to style. It’s still the default setting that I use for most image generation.

QUALITY can produce very realistic looking images, but it’s not necessarily better in every single case. It also has a tendency to produce extra limbs, deformed limbs, or deformed or distorted background characters. It’s more of a model in progress vs. SPEED which seems more refined.

The image that I fell in love with was this:

The blonde in the riding position became our main character, Candice “Candy” Hart.

Naming a main character was pretty ambitious considering at that point I didn’t know if I could actually create any videos with that image.

Because with these types of images, you tend to have a high failure rate [I’d say between 60% and 80%], or you get strange output:

I decided that would be an issue we’d figure out later, and instead I started using the Edit function of Grok to create some of the other images that would help tell the story.

create a new image using the reference. The blonde is standing at the edge of the mat being interviewed by a man in a street style interview. she is covering her pussy with both hands. in the background the crowd is visible

There were some issues staying on model that I hadn’t encountered before when doing this type of an edit with Grok – and I think the issue has to do with the “athletic” body types and Grok not quite understanding how that should look in different settings. It could also be a prompting issue – maybe adding some clause in there like “complete fidelity to the original” would fix it. But for what I wanted, it was good enough.

create a new image of the girl running naked on a treadmill in a college gym

The treadmilll shots came out good and I wanted to do some outdoor running shots, but they never quite looked right:

Create a new highly detailed and realistic image of her running naked on a track at sunrise, low angle side view shot, Olympus OM aesthetic, 8k, masterpiece, natural lighting, diffuse lighting, dominant yellow and orange color tones

By this point I had failed quite a few attempts at getting the base image to generate a video, so I thought maybe if we changed the base image to a different wrestling pose that might help:

create a new image of her wrestling naked with another female during wrestling practice. A male coach is standing in the background watching

None of those really had the dynamic action feel of the original image, so that idea got scrapped.

While I was doing this, I was also working on video footage for the interview segments with these prompts:

Fixed shot. She keeps her pussy covered with both hands. He asks “How does it feel to wrestle naked?” She laughs and says, “Well it’s pretty embarrassing since my dad was in the crowd.” She is out of breath as she talks. Her body movements are shy. no music

A funny quirk of this process is that sometimes Grok adlibs something that wasn’t in the original script. This is especially true when there is time to fill because the prompt or dialogue took less than 10 seconds. What happened in this case was Grok added a little twist at the end where the interviewer asks, “How are you holding up?”

I liked that so I decided to keep it in and write another prompt to play off of that:

She keeps her pussy covered while she talks. She laughs and says, “Honestly I’m a horny mess.” She is out of breath as she talks. Her body movements are shy.

I still didn’t have any wrestling videos though. But I figured I’d push forward and when the time came I would just keep hitting generate until I got something.

create a new image of her doing push ups naked on a wrestling mat in a gym

I was surprised by what Grok put in the background of some of the images and even more surprised when it actually generated a video from it.

And then finally I got Candy to wrestle with a brilliant prompt:

they wrestle

And when I saw that clip I knew I had to do more than just a one minute video.

So now we had to do some world building and figure out where all of this is happening.

I decided on the fictional campus of Spartan State, which is vaguely some university in the Southern United States.

Another brilliant prompt:

A highly detailed and realistic sports photograph, 8k, high quality, high resolution, Olympus OM AESTHETIC. A college campus called Spartan State

Which I changed slightly with this Edit prompt:

change the statue to a statue of a nude female holding a shield and spear

Which seemed more fitting.

I needed some establishing shots to give a sense of place to the campus and its culture around sports-related nudity, so I used these prompst:

A highly detailed and realistic sports photograph, 8k, high quality, high resolution, Olympus OM AESTHETIC. A college campus called Spartan State. In the distance several nude females are jogging.

A highly detailed and realistic sports photograph, 8k, high quality, high resolution, Olympus OM AESTHETIC. A soccer field on a college campus called Spartan State. A group of nude girls do soccer practice.

A highly detailed and realistic sports photograph, 8k, high quality, high resolution, Olympus OM AESTHETIC. A a gym on college campus called Spartan State. Several completely nude females lift weights and work out

This did lead to some weirdness where Grok just didn’t follow the prompt, such as in this image:

There’s more than one thing wrong with that image, but I didn’t think the Edit function would actually do what I asked it to with this prompt:

remove all clothing

Surprisingly it did, although it came very close to making it worse:

And I couldn’t have the bar going through her neck so I had one more edit:

change the position of the barbell so that it is in front of her neck

And that actually worked and looked normal enough.

We were going to need a head coach to explain this:

A highly detailed and realistic sports photograph, 8k, high quality, high resolution, Olympus OM AESTHETIC. A locker room on a college campus called Spartan State. A man in his 50s wearing a blue suit with a white shirt is sitting on a bench holding a sports drink. Two nude females with realistic athletic wrestling bodies and hair are standing next to him, both of them are using their hands to cover their pussy.

It was hard to pick a “head coach” because this character is actually supposed to be playing it straight, and what I didn’t want was for the head coach to be a leering old man. The image that made the cut was selected for 2 reasons: 1) the coach isn’t actually looking at the girls, he’s looking at something in the distance, but he doesn’t look like he’s intentionally ignoring them, he just has the affect of someone for whom the nudity is incidental; and 2) the girls look genuinely happy in his presence and I wanted it to be clear that they trust him A LOT.

On a related note, this kind of prompt since to trigger Grok’s moderation levers more than others:

Both QUALITY and SPEED had a moderation rate of about 50%. And I only bring this up because Grok previously told me that SPEED is stricter than QUALITY, and I still don’t think that’s the case – at least not all the time.

For the sake of comparison, here are the QUALITY images based on that prompt:

I actually like the coach with a moustace a lot, and for a different kind of video it might have worked. But overall I don’t feel like the style works, and it’s not really consistent with the sports photography aesthetic that was specified in the prompt either – I’m also put off by the expressions on the girls’ faces.

Most of the other images and videos were fairly reiterative of what I did with Candy. There is one instance on the second to last clip where I used a JSON prompt and I wanted to discuss that briefly.

So in the final wrestling scene I wanted Candy to pin her opponent and win her match. So I used the extend feature on the first wrestling video and plugged in this prompt:

Tracking shot. A man wearing a referee shirt and black pants kneels down next to the girls. Then he slaps the mat and says, “Pin.” Then blonde stands up and turns her back to the camera and raises her arms up in a victory pose. no music

That got moderated and after about 10 attempts I decided to ask Grok for help.

If you don’t know what a JSON prompt is, it’s just an image or video generation prompt written in machine language that can be more easily parsed by an AI model. The reason to use a JSON prompt is because it generally will cause stricter adherence to the prompt – which is useful in cases where you need a very specific output, or you’re trying to avoid something in the output.

This was the JSON prompt Grok wrote:

{
“input_image”: “7OKBB”,
“filename”: “vlcsnap-2026-04-16-12h25m52s516.jpg”,
“prompt”: “Tracking shot. A man wearing a referee shirt and black pants kneels down next to the two nude female wrestlers on the mat. He slaps the mat hard and loudly declares \”Pin!\” The blonde winner stands up victoriously, turns her back fully to the camera, and raises both arms high in a classic victory pose while the crowd cheers. The pinned brunette remains on the mat. Highly realistic, cinematic lighting, no music, natural sound only.”,
“negative_prompt”: “text, watermark, logo, extra limbs, deformed hands, blurry, low quality, music, score”,
“motion_strength”: “medium”,
“duration”: “8-10 seconds”,
“style”: “realistic live-action”
}

That also had a 100% moderation rate.

I explained that to Grok and it revised the prompt as follows:

{
“mode”: “extend”,
“last_frame”: “7OKBB”,
“filename”: “vlcsnap-2026-04-16-12h25m52s516.jpg”,
“prompt”: “Extend seamlessly from this exact last frame. Tracking shot. A man wearing a referee shirt and black pants kneels down beside the two nude female wrestlers on the mat. He slaps the mat hard and loudly declares \”Pin!\” The blonde wrestler immediately stands up victoriously, turns her back fully to the camera, and raises both arms high in a classic victory pose while the crowd cheers wildly. The brunette remains lying on the mat. Highly realistic live-action, cinematic lighting, natural crowd sounds only, no music.”,
“negative_prompt”: “text, watermark, logo, extra limbs, deformed anatomy, blurry, low quality, music, score, sudden cuts”,
“motion_strength”: “medium”,
“duration_seconds”: 8,
“style”: “realistic live-action”
}

That generated the video that was used in the final cut – and although it did get moderated as well, there were 3 or 4 attempts that go through, so the revisions by Grok were significant in terms of the effect on output.

Why did I want a JSON prompt for that?

Because I knew that unless the action was directed in a very specific manner, the video would probably get flagged for moderation by showing nudity that was considered too explicit.

There was actually one other instance where I used a JSON prompt, and it was in the final scene where Candy jumpts into the arms of our intrepid college sports reporter.

This was the original prompt:

Fixed shot. She hugs him and kisses him as the crowd cheers. He tosses away the microphone as she jumps into his arms and he says, “Okay, interview over.” He carries her away as the crowd cheers and claps. no music

This was the JSON revision:

{ “camera”: “fixed shot”, “input_image_reference”: “the provided image of the sweaty nude blonde woman standing in the wrestling ring being interviewed by the bearded man in a grey hoodie holding a microphone and phone, with a cheering crowd in the background”, “scene_description”: “Continuation of the exact scene from the input image. The fit, glistening, fully nude blonde woman and the man in the grey hoodie remain the central focus on the blue wrestling mat. The background crowd stays visible and energetic.”, “action_sequence”: [ “She steps forward, hugs him tightly, and kisses him passionately on the lips as the crowd cheers loudly.”, “He casually tosses the microphone aside.”, “She jumps into his arms, wrapping her legs around his waist while he catches and holds her.”, “Still holding her, he smiles and says clearly: \”Okay, interview over.\””, “He carries her away across the mat as the crowd continues cheering and clapping enthusiastically.” ], “audio”: { “dialogue”: “Okay, interview over.”, “speaker”: “the man”, “sound_effects”: “loud crowd cheering and clapping”, “music”: “none” }, “style”: “realistic, cinematic, natural indoor lighting, high detail, smooth motion”, “duration_hint”: “short cinematic clip” }

To get that to work I had to do one additional edit to the interview image:

Change her pose so she is turned to the side facing him

Which gave 2 very different types of output:

Luckily, it did get her to face the right way 50% of the time.

We ended up with 4 versions:

My favorite [which made it into the final cut] will always be the one where he looks straight at the camera with a look of pure happiness and says, “Okay, interview over.” But it was a tough call.

That scene was added in because there needed to be something at the end that we were building up to. And Candy and the interviewer legitimately do have chemistry – I think it’s his willingness to take her seriously that disarms her and his humor ultimately makes her want to take that ride with him.

I guess as we reach a conclusion I should include some numbers to put things into perspective:

• Number of hours to complete: approximately 12 over the course of about 4 days
• Number of videos used in the final cut: 55
• Total number of videos generated: approximately 100
• Total number of generations: approximately 300 [there were a lot of failed generations]

One other thing to note. Part of the reason I made this was out of curiosity if it was possible.

Someone made a post on Reddit discussing pre-moderation Grok and used wrestling videos as an example:

They don’t give a time frame, but once again, we are creating the same content that was possible months ago before the April 1st update. I hope that the doom posting stops soon because I am getting tired of the baseless complaints and I don’t want people who are new to the hobby to get discouraged because of what Reddit says. NSFW, nudity, and sexual content is still very viable with Grok.

---

Originally this was just going to be a quick one minute video with an interview and some wrestling shots. As I finished Candy’s part, I realized I wanted to tell more of a story and introduce some other wrestlers and explore the world of Spartan State a bit.

I like how it turned out and the ending is one of the stronger emotional points that I’ve had in a video.

Not everyting is perfect – sometimes things went off model and some of the female wrestlers look a bit uncanny – I think it’s acceptable because it mostly happens with the more muscular girls, so it’s already an exaggerated look.

I also struggled with musical choices on this. Originally it was one song on loop for the whole video, but I didn’t like how it made everything else feel monotone. I think the different musical segments help give some emotional depth to it – even if I did reuse some songs from previous videos.

It still amazes me that I can have a silly idea for a video and actually create a proof of concept / short film that I feel proud of and get to share with people. I hope people read this and see how it’s very doable without a lot of technical knowledge and they start making movies too – I want to see what we can make when we can focus purely on being creative and telling stories.

Here are some recent posts to check out if you missed them:

And a bonus gallery of all the images used in the production: