🚀 Model Overview: Unleashing the Power of Imagination! 🌠

Introducing FFUSION AI - a groundbreaking tool for image generation and transformation, crafted around the cutting-edge Latent Diffusion Model.

Dive into the surreal world of FFUSION Ai, powered by Stable Diffusion 2.1, and let your favorite prompts transform into captivating works of art. Effortlessly weave your ideas with mesmerizing effects, immersing your audience in a world where imagination knows no bounds.

Developed by: Idle Stoev, Source Code Bulgaria, Praesidium CX & BlackSwan Technologies

Model type: Diffusion-based text-to-image generation model

Language(s): English

License: CreativeML Open RAIL++-M License

🔬 Intended Use: From Research to Artistry 🎨

FFUSION AI is a multi-faceted tool that shines in various applications. Primarily envisioned for research, FFUSION AI has potential to: Examine the limitations and inherent biases in generative models. Unleash the artist within, aiding in creative processes or artistic endeavours. Reinvent educational or creative utilities with AI-driven innovations. Propel the research in the fascinating domain of generative models. However, it's crucial to note that certain uses of FFUSION AI are strictly prohibited, as outlined below.

🚫 Forbidden Use: Setting Boundaries for Safe AI 🛑

We've borrowed the principles from the Stable Diffusion v2.1 model card, which apply equally to Fusion AI alpha, beta, and final releases. We strictly prohibit the use of this model for generating or spreading images intended to incite hostility or alienation. This includes content that is foreseeably disturbing, distressing, offensive, or stereotype-propagating.

# Out-of-Scope Use:

Since this model isn't designed to create factual representations of people or events, such usage is deemed out-of-scope.

# Misuse and Malicious Use:

Utilizing the model to create content that inflicts harm upon individuals is deemed misuse. This includes:

Generating content that belittles, dehumanizes, or otherwise harms individuals or their environments, cultures, religions, etc.
Deliberately promoting or disseminating discriminatory content or harmful stereotypes.
Impersonating individuals without consent.
Generating sexual content without viewer consent.
Spreading mis- and disinformation.
Illustrating extreme violence and gore.
Distributing copyrighted or licensed material against its usage terms.
Modifying copyrighted or licensed material against its usage terms.

Our policy, adopted from the principles of the Stable Diffusion v2.1 model card, ensures the responsible use of Fusion AI beta and final releases. We expressly prohibit the utilization of our model for generating or distributing images that might incite hostility or exclusion. This includes:

Content that is distressing, offensive, or perpetuates harmful stereotypes.
Misuse or malicious use that harms individuals or communities, including creating demeaning or harmful representations, or promoting discriminatory content.
Using the model for impersonation without consent or creating non-consensual explicit content.
Generating or spreading mis- and disinformation, violent, gory imagery, or violating copyright terms.

🔭 Model Limitations and Bias: Acknowledging Imperfections 🌐

While our model leaps toward the future of AI-driven creativity, it's essential to recognize its current limitations:

The quest for perfect photorealism and creative surrealism continues.
Rendering legible text remains a challenge.
Even more complex tasks, such as depicting "A red cube on top of a blue sphere in the middle of the ocean in a desert" may pose difficulty (but still processable).
Human figures, particularly faces, may not be accurately generated.

Version Releases

We are excited to unveil the following versions:

Version 512 Beta – LiTE, MiD BFG model variations:

FFUSION.ai-512-beta-BFG-build.0401.safetensors
FFUSION.ai-512-beta-LiTE-build.0201.safetensors
FFUSION.ai-512-beta-MiD-build.0401.safetensors

Version 768 Alpha - BaSE, FUSION, FFUSION:

BaSE and FUSION models will soon come with enhanced training capabilities including LoRa, LyCORIS, Dylora & Kohya-ss/sd-scripts. More information will be revealed upon release.

di.FFUSION.ai-v2.1-768-BaSE-alpha-preview.safetensors

FUSION AI Text Encoders:

di.FFUSION.ai-tXe-FXAA: Trained on "121361" images. Enhance your model's quality and sharpness using the pre-trained Unet.
di.FFUSION.ai-tXe-fX: Trained on "211924" images. Amplify your model's surrealism and effects.

Environmental Impact

Our dedication to sustainable development is reflected in the model's carbon footprint. The CO2 emissions, calculated using the Machine Learning Impact calculator, stand at 124.95 kg for a total of 1190 hours of usage with an A100 PCIe 40GB GPU.

Hardware Type: A100 PCIe 40GB

Hours used: 1190

Cloud Provider: CoreWeave & Runpod (official partner)

Compute Region: US Cyxtera Chicago Data Center - ORD1 / EU - CZ & EU - RO

Carbon Emitted (Power consumption x Time x Carbon produced based on the location of the power grid): 124.95 kg of CO2 emitted.
Power consumption x Time x Carbon Produced Based on the Local Power Grid: 250W x 1190h = 297.5 kWh x 0.42 kg eq. CO2/kWh = 124.95 kg eq. CO2
Local Hardware Storage 4x16TB Raid5 WD Gold Optimizer: AdamW & Dadaptation

This model card was written by: Idle Stoev and is based on the Stability AI - Stable Diffusion 2.1 model card. Models: