HotPixl AI artworks (2025)

19/09/2024

All the AI image generation models are trained to give you an image you ask for.

They work using a step methodology. They create an image and then use that image as the start of the next run, improving it each iteration.

The early images are raw, shadows of what you ask for but have a unique feeling, like waking up and not being able to see clearly yet.

I have two images for this post, first had twenty steps and the second had three. They're both nice but I prefer the second, it's etherial in a way that a finished image really never is.

11/09/2024

It’s always a great idea to start something new right before you move. Actually, scratch that — it's a terrible idea.

It's been five months, almost to the day, since my last post.

In my defense, I've been busy: moving, working on our old house, moving, dealing with projects at the new house, and, of course, moving some more. Then there were two vacations, endless moving, sorting out electrical quirks at the new place, rebuilding my network, and getting everything back up and running.

On the bright side, I managed to miss the whole Stable Diffusion 3 license controversy.

I’m jumping back in just after Flux1's release — perfect timing, right? I'll share more about what makes Flux1 unique another time. For now, here’s the prompt used to generate these images with the flux1-dev-fp8 and flux1-schnell models:

"A beautiful landscape with a young girl facing the viewer, holding a sign that reads, 'I do not exist,' with a setting sun in the background, her face glowing in golden hour hues."

Those familiar with AI-generated images will see what makes this model stand out, especially for portraits.

It's not perfect but it's doing two things particularly well without any workarounds that are difficult with many other models.

08/04/2024

Last four fun eclipse images for the day. Hope everyone had a great day

08/04/2024

Second set of eclipse images for today.

08/04/2024

My first set of eclipse images for today.

Each batch will be similar with minor changes to the seed or a mid batch model switch.

Looking at the weather AI might be the only eclipse we see today.

08/04/2024

The week I explored the ReActor face swap node within ComfyUI, showcasing the versatility and fun of digital manipulation by integrating my own face into four distinct images.

There are several face swapping techniques available, each with unique approaches to either refine the final visual output or steer the image generation process to closely mirror a specified source image. Among these, the ReActor face swap node stands out for its user-friendly interface, making it an accessible tool for achieving impressively accurate results with minimal effort.

I'm goin to look into the various IP Adapter nodes this week, exploring the potential they hold for more advanced face swapping.

04/04/2024

Apologies for the brief hiatus; my setup has recently undergone an upgrade with the arrival of a second, nVidia 3090 video card. The installation wasn't straightforward—it required custom-designed 3D-printed components to integrate into my unique open air case. Now, these cards alone have a considerable appetite for power, drawing around 500 watts just for rendering tasks, and pushing the overall power consumption of my system to nearly 700 watts.

I'll delve into the intricacies of my machine in an upcoming post, detailing the deviations from my original idea and discussing how these alterations have influenced performance. It's a topic I think will interest some.

For the image creation today, I opted for a solitary checkpoint—a single model approach despite having amassed an 500GB collection of various models. This process has been enlightening, underscoring that each model presents its own unique challenge. Although these models share a common origin, unlocking their potential requires a nuanced understanding of specific settings critical for each individual model.

I retained the same prompt and seed for consistency in the images I produced today. However, tweaking certain parameters—such as the sampler configuration, which determines the degree of influence the prompt has on the image; the number of steps, which refine the transition from random noise to a clear image; the sampler model, introducing unique computational methodologies for image generation; and the scheduler, which orchestrates the noise reduction process—can dramatically alter the outcome. The first image I'm sharing represents the culmination of finding the right settings for this model for my desired outcome, whereas the second reveals the starting point using a previously successful setup.

The creative directive for both images was to conjure a "intricate stylized 3D silver, gold, and copper wire sculpture of a full-bodied, long-haired Dachshund, ensconced within a round walnut wood frame, featuring a single letter 'E' in a style that exudes luxury and elegance, rendered in the highest definition."

01/04/2024

In today's exploration, I delve into the fascinating world of checkpoints within the realm of generative models, focusing specifically on their pivotal role in the development and refinement of images. The foundation for this study is the Stable Diffusions XL base model, which serves as the starting point for each transformation. Checkpoints, as I will demonstrate, are not mere stages but crucial junctures where the model undergoes refinement, each adjustment bringing a unique flavor to the resulting images.

To maintain consistency and minimize variability, my approach involved keeping the prompt and seed constant throughout the experiments. The prompt I employed was meticulously chosen to evoke a specific atmosphere and style, reminiscent of the renowned works of John Singer Sargent. It read:

"Create an image in the style of John Singer Sargent of a woman sitting against a tree, immersed in a book, while overlooking a tranquil pond. The woman should exude a sense of quiet serenity and contentment, with the sun casting long shadows on her face as she gazes at the peaceful landscape. The tree, tall and imposing, reaches towards the sky, while the pond reflects, mirroring the woman's inner peace."

To enrich the visual narrative and infuse the images with distinctive stylistic elements, I utilized two specialized Loras. For those unfamiliar, "Loras" refers to layers of refinement algorithms designed to enhance the generative model's output in specific ways. The first Lora was trained explicitly in the style of John Singer Sargent, ensuring that the images bore the hallmark elegance and sophistication of his artistry. The second Lora, referred to as a "fog Lora," was employed to subtly introduce an element of mist into the background, adding a layer of atmospheric depth and mystique to the scenes.

This exploration not only showcases the transformative power of checkpoints and Loras in shaping the aesthetics of generative art but also highlights the delicate balance between technological precision and artistic inspiration.

31/03/2024

Today, I delve into the intricacies of two pivotal elements in image generation: the prompt and the seed. These components play crucial roles in shaping the outcome of the images we create.

**Prompts:** At the heart of image generation, prompts act as the guiding light. They are essentially the descriptions or instructions we provide to the AI. Intriguingly, prompts can be bifurcated into two types: positive and negative. The positive prompts articulate what we desire in the image—detailing the elements we wish to include. Conversely, negative prompts serve as boundaries, specifying what we aim to exclude from our visual creation. This dual approach allows for a nuanced control over the image generation process, enabling creators to steer clear of unwanted elements while emphasizing desired features.

**Seed:** The concept of the seed introduces a layer of unpredictability and uniqueness. A seed is a random number that dictates the starting point of the image generation algorithm. In the realm of computing, generating a truly random number is a complex feat, yet it's fundamental to achieving diversity and spontaneity in the creation of images. The seed ensures that each image, even with identical prompts, possesses its own distinct character and composition.

In a practical exploration, I initiated the journey with four images, each born from a unique prompt crafted through conversational interaction with ChatGPT. Despite the uniformity in the request, the varied prompts led to the generation of distinctly unique images, highlighting the AI's interpretative flexibility.

Following this, I embarked on a different experiment with the last five images, maintaining a constant prompt but altering the seed for each generation. This process unveiled the profound impact of the seed on the outcome. Among these, one image emerged as notably nonsensical—a testament to the unpredictable nature of using random sequences in image generation. It underscores the element of surprise and exploration inherent in leveraging randomness, where the unexpected can emerge from the algorithmic depths.

Through these explorations, the interplay between the prompt and the seed emerges as a dance of control and chance, guiding the creative process in generating digital art. It's a reminder of the delicate balance between direction and serendipity in the realm of AI-assisted creativity.

30/03/2024

My inaugural post will delve into a personal compilation of works from my favorite books and ideas. In this first post I'm eager to merge images from Norman Maclean's "A River Runs Through It" with my own unique perspective. The book, which was made popular by the 1992 film adaptation, is a testament to Maclean's storytelling prowess. I wholeheartedly recommend reading the book, for it offers a more immersive experience than any film could ever provide.

HotPixl AI artworks

19/09/2024

11/09/2024

08/04/2024

08/04/2024

08/04/2024

08/04/2024

04/04/2024

01/04/2024

31/03/2024

30/03/2024

Address

Website

Alerts

Contact The Business

Shortcuts

Share