//ARTIFICIAL INTELLIGENCE

Unleashing the power of AI with Passion and Precision.

Crystal Clear Team

  • Art Consultant-

    • I leverage my Bachelor's of Science in Art Education to provide instructive guidance on the elements and principles of design to fellow developers. This collaborative effort has led to remarkable progress in training our AI models to generate captivating artwork that holds attention and embodies aesthetic appeal.

  • Model trainer-

    • As an active member of the Crystal Clear development team, I train our Stable Diffusion models using Low-Rank Adaptation (LoRA).

    • I employ my expertise in image processing, textual inversion, and meta-text captioning to preserve image structures while effectively reducing noise. I collaborate closely with an international, multidisciplinary team to enhance the performance and visual quality of the denoising models, ensuring their suitability for real-world applications.

  • Beta tester-

    • As a beta tester, I am responsible for rigorously testing and evaluating two AI image generation models. I meticulously assess the performance, stability, and responsiveness of each model, determining the maximum stable resolution while ensuring exceptional image quality. Promptly addressing any responsiveness or image quality issues, I play a crucial role in refining and optimizing these models for optimal results.

  • Regularization image database contributor-

    • I play a key role in the compilation and curation of a diverse and extensive image database for batch normalization, actively contributing to the meticulous organization and categorization of images. My contributions enhance the performance and generalization capabilities of 2 photorealistic CC-branded image processing models.

//Crystal Clear Above Reality RC3

Personal Image Generation Showcase

“Sam the Onionman”

AI-assisted videos

Okay, look. I have strange hobbies.

What started out as a joke on my Discord server quickly turned into a second youtube channel, Sam the Onionman. Combining my love for all things cursed—be they AI generations, body-horror atrocities, or even the unprincipled ways we advertise products—Sam is the harbinger of the fever dream. Here I unapologetically produce original, so-bad-it’s-good nightmare fuel, fresh and hot in a pseudo-found-footage wrapper.

Multimedia video-bashing.

Armed with a jail-broken Large Language Model, Stable Diffusion Deforum, Spline Motion, post-process effects in nonlinear video editing (and maybe some of the most intentionally terrible concepts out there), I’m pumping out load after load of hot, messy video content! Be sure to check out:

  • Drumstik Socks (they turn you into a poultry warrior!)

  • Dromedary Cigarettes (They WILL kill you… with flavor?)

  • coming soon! The Last Breakfast (claymation horror short)

Photobashing Workflow

Photobashing Workflow

AI art is seen as cheap and easy, but even the best models need human intervention for a good final product.

There’s a lot of work that goes in behind the scenes of a polished work, and in my opinion, the best AI art is actually AI-Assisted art. I use AI as a tool, just like the pioneer photographers and photoshoppers before me.

Nicky McSandyScoot, Pixie Artificer

  1. Generate a thick pixie artificer in Stable Diffusion (SD).

  2. Using OpenPose, create a mannequin of the pose I want my character to ultimately be in.

  3. Using ControlNet, re-render the pixie into the pose. She doesn’t ever quite get there, so feed the best image back into img2img.

  4. Re-render the image using ControlNet for the pose, as well as adding a reference image in ControlNet to keep the character as close to Step 1’s image as possible.

  5. Render an adventurer for her to sit on, then import into GIMP. Erase background on pixie, reintroduce the wings from Step 1, change hair back to red, place her on the shoulder of the adventurer. Add a thigh tattoo and make her lenses blue again.

  6. Re-import image to SD and re-render at higher resolution, which smooths out any lazy image manipulation oversights from step 5 and fixes resolution differences between the pixie and the Adventurer.

Ouldbae exploring wreck of the Compass Rose

  1. Create SoftEdge map of a ship’s hold from royalty free stock.

  2. SoftEdge map of a person with a beard swimming from royalty free stock.

  3. SoftEdge map of a person in a dress swimming from royalty free stock.

  4. Map bash of a person with a beard in long flowing clothes swimming in a ship’s hold in GIMP.

  5. Low-res render in Stable Diffusion. This took multiple tries and multiple seeds.

  6. Once the right one was selected, it was upscaled for better detail and coherence.

  7. As SD can’t understand underwater scenes (due to the way it renders images through clearing up noise), I found an image of water with tiny bubbles from royalty free stock.

  8. Superimpose images in GIMP, adjust layer blending modes, and add radial gradient from the entry point (a hole in the overhead deck).