BLOG

Bathroom Selfie

by GymDreams
Taking a selfie in the bathroom before going to the gym.
Bathroom Selfie, with Virile Reality in Stable Diffusion

Selfie Series

The Selfie is a commentary about selfies seen all over Instagram. It is a technical challenge, made to see if I could replicate that same look and feel from selfies that are casually snapped, and as such with a figure situated in a surrounding that’s not tidied up. AI is great at making perfect things, so creating surroundings that are natural and messy poses its unique sets of challenges. This series originally started as a joke, but was generally very well received. I also tend to get so many DMs every time I post them because most people don’t read description and they always asked if it’s me. Short answer: no, but the first two me in this series is based on someone special in my heart.

The Selfies series literally started as a joke — I made some renders as a response to the selfies I saw on Instagram. I posted one of them on my IG story, and got so many messages asking me if it’s me. Then a friend on IG (who also doesn’t know what I look like), asked me if it’s real, and suggested that I post to my feed instead because the thought that it’s really good. So I found some extra renders that I did and now it’s closing in at 900 likes after a week.

For that first set, I wanted to make it very clear that it’s not real, so I rendered it in Virile Fusion, which gives very photorealistic renders but that are clearly an illustration when you look closely at the colors and textures. Like a photorealistic 3d render. Here’s an image from the first set:

Selfie in the Bedroom, created with Stable Diffusion and Virile Fusion.
Selfie in the Bedroom, created with Stable Diffusion and Virile Fusion.

Now that we have established that these are not real, I want to try to render them closer to photography. For this second set, I picked Virile Reality, which is a checkpoint model made by the same guy behind Virile Fusion. As a photorealistic model, the body hair will render a bit more realistic than the first set. Where it didn’t fully fool you is probably the lighting in the room itself. It‘s lacking a bit of texture and dirt. These are things that I probably could add in Photoshop in post, but I didn’t do it because that wouldn’t truly show what the model would natively do. In the future, as this series develop, I will try to use my Photoshop-fu to add those necessary details.

As for the bathroom, it’s another popular location that people take selfies in. There’s a very good reason why — first, there’s a mirror. Second, and I think that it‘s fairly important, it’s that the reflective surfaces inside a bathroom often provide lighting qualities similar to a soft box, so it rendesr indirect lights very well. If you don’t believe me, just snap some photos the next time you go to the loo. The only unfortunate artifacts are the toilets everywhere, but I think that by now everyone is used to seeing them so it’s not so strange anymore.

For the framing, I tried a few different ones — both including the face and excluding the face. Often people take photos without their face to post on dating apps, and it’s almost a signature of these bathroom selfies, and that’s why I also made some of them that way.

Text prompts in Stable Diffusion, with Virile Reality 2.0, 30 steps, DPM++ SDE Karras, CFG 5, 512x512, vae-ft-mse-840000-ema-pruned.vae, Denoising 0.5, ADetailer: mediapipe_face_full, Hires upscale: 1.5 (768x768), hires steps: 15, R-ESRGAN 4x+. Post: Gigapixel 4x HQ (3072x3072), Adobe Lightroom color correction, Adobe Photoshop Beta AI (hand fix).

Images

Set 1

Bathroom Selfie, with Virile Reality in Stable Diffusion
Bathroom Selfie, with Virile Reality in Stable Diffusion
Bathroom Selfie, with Virile Reality in Stable Diffusion
Bathroom Selfie, with Virile Reality in Stable Diffusion
Bathroom Selfie, with Virile Reality in Stable Diffusion
Bathroom Selfie, with Virile Reality in Stable Diffusion

Set 2

Bathroom Selfie, with Virile Reality in Stable Diffusion
Bathroom Selfie, with Virile Reality in Stable Diffusion
Bathroom Selfie, with Virile Reality in Stable Diffusion
Bathroom Selfie, with Virile Reality in Stable Diffusion
Bathroom Selfie, with Virile Reality in Stable Diffusion
Bathroom Selfie, with Virile Reality in Stable Diffusion

Set 3

Bathroom Selfie, with Virile Reality in Stable Diffusion
Bathroom Selfie, with Virile Reality in Stable Diffusion
Bathroom Selfie, with Virile Reality in Stable Diffusion
Bathroom Selfie, with Virile Reality in Stable Diffusion
Bathroom Selfie, with Virile Reality in Stable Diffusion
Bathroom Selfie, with Virile Reality in Stable Diffusion
  • Steps: 30
  • Sampler: DPM++ SDE Karras
  • CFG scale: 5
  • Size: 512x512
  • Model hash: 63f0231527
  • Model: virileReality_v20
  • VAE: vae-ft-mse-840000-ema-pruned.vae
  • Denoising strength: 0.5
  • Version: a844a83
  • Token merging ratio: 0.5
  • Token merging ratio hr: 0.5
  • Parser: Full parser
  • ADetailer model: mediapipe_face_full
  • ADetailer confidence: 0.3
  • ADetailer dilate/erode: 32
  • ADetailer mask blur: 4
  • ADetailer denoising strength: 0.4
  • ADetailer inpaint only masked: True
  • ADetailer inpaint padding: 32
  • ADetailer version: 23.7.6
  • Hires upscale: 1.5 (768x768)
  • Hires steps: 15
  • Hires upscaler: R-ESRGAN 4x+
  • Lora hashes: "muscle2: 58b17ca61429"
  • Post: Gigapixel HQ 4x (3072x3072)
  • Color Correction: Adobe Lightroom
  • Post Retouch: Adobe Photoshop
  • Post Inpaintg: Adobe Photoshop Beta AI Generative Layers (hand / finger fixes)