Particle Examples
A. Introduction
A.1 The Architecture of Local Generative AI
The visual examples presented in this gallery are the product of local generative inference, a workflow that shifts the power of high-end image synthesis from cloud providers directly onto personal hardware. Operating AI locally provides absolute creative sovereignty, allowing for the deep fine-tuning of parameters and models without the constraints of external API filters or subscription tiers.
A.2 The Ecosystem: Pinokio, ComfyUI, and Stable Diffusion
The engine behind these particle simulations is Stable Diffusion, orchestrated through ComfyUI. Unlike standard interfaces, ComfyUI utilizes a node-based graph architecture, allowing the user to map out the exact path a mathematical tensor takes - from initial noise to final high-fidelity pixels.
To manage this complex environment, i utilize Pinokio. It acts as a specialized browser and autonomous orchestrator that simplifies the installation of the myriad dependencies required for AI, such as Python virtual environments, Git repositories, and CUDA drivers. It essentially turns the terminal-heavy task of setting up AI into a streamlined, "one-click" local experience.
A.3 Processing Power: GPU and VRAM
At the heart of this process is the GPU (Graphics Processing Unit). Generative AI is not a CPU-heavy task; it requires the massive parallel processing capabilities of a video card. The VRAM (Video RAM) is where the "latent space" of the models - like Dreamshaper 8 - is loaded. These models are essentially multi-gigabyte mathematical maps of human visual concepts. By utilizing custom upscalers we can transform these latent mathematical distributions into the sharp, intricate particle structures seen below.
A.4 Used Environment
| Hardware | Specs |
|---|---|
| CPU | Intel Xeon W-2145 @ 3.7 GHz, 8 Cores |
| GPU | NVIDIA Quadro RTX 4000, 8GB GDDR6 VRAM, 2304 Cuda Cores, 288 Tensor Cores, 43T RTX Ops, 7.1 TFLOPS FP32 Performance |
| HDD | 500 GB System SSD, 16TB Data HDD, 13TB Data SSD/NVMe |
| RAM | 96 GB 2.6 GHz ECC |
A.5 Used Workflow
All images are using the 4xOpscaler_general.pt upscaler as well as ImageAddNoise nodes before and after the upscaler.
A.5.1 Glossary
| Node | Section | Description |
|---|---|---|
| Load Checkpoint | Loading Models | Initializes the foundational Stable Diffusion model, containing the weights for the UNet, CLIP, and VAE |
| Load Upscale Model | Loading Models | Loads a specialized neural network (e.g., ESRGAN or SwinIR) designed specifically for enhancing resolution and detail |
| CLIP Text Encode (Prompt) Positive | User Prompt | Translates your natural language description into mathematical vectors that guide the AI toward specific visual concepts |
| CLIP Text Encode (Prompt) Negative | User Prompt | Defines concepts, styles, or artifacts that the model must actively avoid during the generation process |
| Empty Latent Image | AI Generation | Defines the initial resolution and aspect ratio in "latent space" - a compressed mathematical environment where the noise is processed |
| KSampler | AI Generation | The engine of the workflow; it iteratively removes noise from the latent image based on the model’s training and your prompt guides |
| VAE Decode | Decoding | Converts the processed latent representation back into a pixel-based image that is viewable by humans |
| ImageAddNoise | Post Processing | Introduces controlled grain to the image to provide the upscaler with high-frequency data, preventing the "over-smoothed" look during enlargement |
| Upscale Image (using Model) | Post Processing | Processes the image through the selected upscale model to increase pixel density by a factor while reconstructing lost details |
| ImageAddNoise | Post Processing | Applies a final layer of film grain or noise to the high-resolution output to unify the texture and eliminate artificial "plastic" digital artifacts |
| Save Image | Memorize Result | Commits the final rendered pixel data to the local storage disk |
B. Overview
1 Pirly Curve
| Model | Positive Prompt | Negative Prompt | Latent Image Width | Latent Image Height | Sampler Seed | Sampler Steps | Sampler CFG-Scale | Sampler Denoise |
|---|---|---|---|---|---|---|---|---|
| dreamshaper_8.safetensors https://civitai.com/models/4384/dreamshaper | macro black particles form waveforms and white particles form background | text | 512 | 512 | 1106882758745727 | 35 | 9.0 | 1.0 |
2 Aflok
| Model | Positive Prompt | Negative Prompt | Latent Image Width | Latent Image Height | Sampler Seed | Sampler Steps | Sampler CFG-Scale | Sampler Denoise |
|---|---|---|---|---|---|---|---|---|
| dreamshaper_8.safetensors https://civitai.com/models/4384/dreamshaper | macro black particles form waveforms and white particles form background | text | 512 | 512 | 1106882758745728 | 35 | 9.0 | 1.0 |
3 Chigo
| Model | Positive Prompt | Negative Prompt | Latent Image Width | Latent Image Height | Sampler Seed | Sampler Steps | Sampler CFG-Scale | Sampler Denoise |
|---|---|---|---|---|---|---|---|---|
| dreamshaper_8.safetensors https://civitai.com/models/4384/dreamshaper | macro black particles form waveforms and white particles form background | text | 512 | 512 | 1106882758745729 | 35 | 9.0 | 1.0 |
4 Astrix
| Model | Positive Prompt | Negative Prompt | Latent Image Width | Latent Image Height | Sampler Seed | Sampler Steps | Sampler CFG-Scale | Sampler Denoise |
|---|---|---|---|---|---|---|---|---|
| dreamshaper_8.safetensors https://civitai.com/models/4384/dreamshaper | macro black particles form waveforms and white particles form background | text | 512 | 512 | 1106882758745730 | 35 | 9.0 | 1.0 |
5 Sin Star
| Model | Positive Prompt | Negative Prompt | Latent Image Width | Latent Image Height | Sampler Seed | Sampler Steps | Sampler CFG-Scale | Sampler Denoise |
|---|---|---|---|---|---|---|---|---|
| dreamshaper_8.safetensors https://civitai.com/models/4384/dreamshaper | macro black particles form waveforms and white particles form background | text | 512 | 512 | 1106882758745731 | 35 | 9.0 | 1.0 |
6 Avufre
| Model | Positive Prompt | Negative Prompt | Latent Image Width | Latent Image Height | Sampler Seed | Sampler Steps | Sampler CFG-Scale | Sampler Denoise |
|---|---|---|---|---|---|---|---|---|
| dreamshaper_8.safetensors https://civitai.com/models/4384/dreamshaper | macro black particles form waveforms and white particles form background | text | 512 | 512 | 1106882758745732 | 35 | 9.0 | 1.0 |
7 Torxy
| Model | Positive Prompt | Negative Prompt | Latent Image Width | Latent Image Height | Sampler Seed | Sampler Steps | Sampler CFG-Scale | Sampler Denoise |
|---|---|---|---|---|---|---|---|---|
| dreamshaper_8.safetensors https://civitai.com/models/4384/dreamshaper | macro black particles form waveforms and white particles form background | text | 512 | 512 | 1106882758745733 | 35 | 9.0 | 1.0 |
8 Squid
| Model | Positive Prompt | Negative Prompt | Latent Image Width | Latent Image Height | Sampler Seed | Sampler Steps | Sampler CFG-Scale | Sampler Denoise |
|---|---|---|---|---|---|---|---|---|
| dreamshaper_8.safetensors https://civitai.com/models/4384/dreamshaper | macro black particles form waveforms and white particles form background | text | 512 | 512 | 1106882758745734 | 35 | 9.0 | 1.0 |
9 Baru Gat
| Model | Positive Prompt | Negative Prompt | Latent Image Width | Latent Image Height | Sampler Seed | Sampler Steps | Sampler CFG-Scale | Sampler Denoise |
|---|---|---|---|---|---|---|---|---|
| dreamshaper_8.safetensors https://civitai.com/models/4384/dreamshaper | macro black particles form waveforms and white particles form background | text | 512 | 512 | 1106882758745736 | 35 | 9.0 | 1.0 |
10 Wav Colaps
| Model | Positive Prompt | Negative Prompt | Latent Image Width | Latent Image Height | Sampler Seed | Sampler Steps | Sampler CFG-Scale | Sampler Denoise |
|---|---|---|---|---|---|---|---|---|
| dreamshaper_8.safetensors https://civitai.com/models/4384/dreamshaper | macro black particles form waveforms | text | 512 | 512 | 1106882758745737 | 35 | 9.0 | 1.0 |
11 Starb
| Model | Positive Prompt | Negative Prompt | Latent Image Width | Latent Image Height | Sampler Seed | Sampler Steps | Sampler CFG-Scale | Sampler Denoise |
|---|---|---|---|---|---|---|---|---|
| dreamshaper_8.safetensors https://civitai.com/models/4384/dreamshaper | macro black particles form waveforms | text | 512 | 512 | 1106882758745738 | 35 | 9.0 | 1.0 |
12 Undergon
| Model | Positive Prompt | Negative Prompt | Latent Image Width | Latent Image Height | Sampler Seed | Sampler Steps | Sampler CFG-Scale | Sampler Denoise |
|---|---|---|---|---|---|---|---|---|
| dreamshaper_8.safetensors https://civitai.com/models/4384/dreamshaper | macro black particles form waveforms | text | 512 | 512 | 1106882758745739 | 35 | 9.0 | 1.0 |
13 Head
| Model | Positive Prompt | Negative Prompt | Latent Image Width | Latent Image Height | Sampler Seed | Sampler Steps | Sampler CFG-Scale | Sampler Denoise |
|---|---|---|---|---|---|---|---|---|
| dreamshaper_8.safetensors https://civitai.com/models/4384/dreamshaper | macro black particles form wavetables | text | 512 | 512 | 1106882758745742 | 35 | 9.0 | 1.0 |
14 Flover
| Model | Positive Prompt | Negative Prompt | Latent Image Width | Latent Image Height | Sampler Seed | Sampler Steps | Sampler CFG-Scale | Sampler Denoise |
|---|---|---|---|---|---|---|---|---|
| dreamshaper_8.safetensors https://civitai.com/models/4384/dreamshaper | macro black particles form wavetables | text | 512 | 512 | 1106882758745742 | 35 | 9.0 | 1.0 |
15 Dont
| Model | Positive Prompt | Negative Prompt | Latent Image Width | Latent Image Height | Sampler Seed | Sampler Steps | Sampler CFG-Scale | Sampler Denoise |
|---|---|---|---|---|---|---|---|---|
| dreamshaper_8.safetensors https://civitai.com/models/4384/dreamshaper | macro black particles | text | 512 | 512 | 1106882758745750 | 35 | 9.0 | 1.0 |