Hyperwassification LoRA
Lead Name
Lukas Kalmar / @lukaskalm
Elevator Pitch
What if there was a tool where every smol Wassie could be woven into a grand tapestry of stunning art that spreads across all the chains?
'Hyperwassification LoRA' for Stable Diffusion is the AI supported magic loom, allowing everyone to craft stunning (or stunned) Wassies, adding their unique threads to a warm, fuzzy blanket of community.
It's more than just a fun diversion too; it's a way to knit all the creative minds of the community into a cozy tapestry of shared dreams and giggles. With this, we won't just see a splash of color; everyone will be part of it, adding their own little stitches to an ever-growing masterpiece of Hyperwassification history that shall forever reverberate through the chains.
Project Description
Purpose and Objectives:
Hyperwassification LoRA is geared towards democratizing art creation within the ByWassies community. The primary goal is to provide a tool that allows everyone to generate unique Wassie content effortlessly, aligning with the ultimate aim of Hyperwassification—spreading Wassies across every blockchain. By reducing the artistic skill barrier, we empower community members to contribute visually to collaborative collections, created assets or create personalized profile pictures, enriching the community's visual identity. This is an important first step for enabling AI enhanced, customized content creation.
Technical Details:
A LoRA (Low-Rank Adaptation) training model makes it easier to train Stable Diffusion on different concepts, such as characters or a specific style. These trained models can then be exported and used by others in image generation. They are currently the most popular "plug-ins" that work in conjunction with big Stable Diffusion models like SDXL or Stable Diffusion 1.5, but are also supported by animation models like Hotshot-XL.
The preliminary Hyperwassification LoRA model, which has been trained on a modest dataset, already shows promising results. The project now aims to enhance the model by training on a carefully curated and tagged dataset to improve its flexibility, relevance and quality.
Hyperwassification LoRA extends the capabilities of Stable Diffusion, an AI for generating images from text prompts, by enabling it to generate images that depict Wassies. By referencing the LoRA model in a text prompt, users can shape the
imagery generated, enabling the creation of personalized Wassie art.
Based on Stable Diffusion XL 1.0 and utilizing open-source tools, this project emphasizes a community-centric development approach.
Community Interaction:
The proposal to release Hyperwassification LoRA as open-source, along with providing an intuitive interface for art generation, balances accessibility with user-friendliness.
Potential collaborations with existing hosting services like runpod, rundiffusion, stable, civit ai generator come to mind. Alternatively, or additionally, the development of a branded interface, be it a general-purpose interface for the Hyperwassification model or a specific byWassies project, could facilitate an easy, direct way for community members to generate Wassie art or even potentially mint it.
Milestones:
- Curate and pre-process a high-quality and varied image database of around 200-1000 images for training that is approved by artists. These images should depict Wassies of varied styles and positions and will be accompanied by text descriptions used for training.
- Train pre-release models and release them to beta-testers.
- Release of a general-purpose model capable of depicting Wassies according to the users' wishes.
- Establish a user-friendly way for community members to interact with the model. This might be an extensive tutorial, a Discord Bot, or a custom web interface.
The project anticipates reaching the third milestone within a month from its commencement, with subsequent development focused on enhancing community interaction.
How does your project add value to the ByWassies community?
Hyperwassification LoRA enriches the community by offering tools to create visual content that resonates with their identity, reflects current events, and embodies their desired style. It cultivates an environment where individuals can contribute to community endeavors like participative collections, regardless of their traditional artistic skills.
Funding Request
$4,500 USD - This includes the development, refinement, and initial community engagement for the Hyperwassification LoRA model.
Financials and Task Breakdown:
Total: $4,500
Estimated Timeline
- Month 1: Curate image database and develop a general-purpose Hyperwassification LoRA Model
- Week 1: Database collection, initial tagging & model pre-trains
- Week 2: Semi-final database for training & beta model tests
- Week 3: Database finalization, beta testing of model
- Week 4: Final Training & Release
- Post-Month 1: Establish user-friendly interaction method (Discord Bot/Tutorial for hosted Services/Branded Custom Interface) and evaluate project continuation based on community engagement and feedback.
Business Model
The core ethos of Hyperwassification LoRA is rooted in community engagement, open access, and the democratization of art creation within the ByWassies realm. Therefore, the primary proposition is to release the model publicly and free of charge. This approach not only facilitates widespread adoption and creative exploration among the community but also holds the potential to attract new users to the ByWassies ecosystem.
As such, I find the public release of this model under a non-restrictive license like GPL/CC/VPL to be of great importance.
While the direct release of the model aligns with the non-commercial spirit of the grant, the vision of further enhancing the user experience and allowing it to serve as a conduit between creative expression and blockchain integration is tempting. For example, it would allow individuals to effortlessly prompt images and mint them directly to a collection.
Although this idea is not part of this initial development grant, it can be envisioned as a perpetually open collection. It can house a growing tapestry of user-generated Wassie art. Each new creation not only stands as a testament to individual creativity but collectively, they compose an ever-evolving narrative of Hyperwassification history. This growing anthology of digital art could serve as a living archive, reflecting the community's creative journey over time.
To sustain and perhaps expand the development and maintenance of such an interface, a modest fee could be levied for minting creations to the collection. This revenue model, grounded in a small transaction fee, is designed to cover operational costs and ensure the long-term sustainability of the platform without imposing a financial burden on the community. Moreover, it aligns with the broader objective of fostering a self-sustaining ecosystem where the value generated is continually reinvested into the community.
Through this approach, Hyperwassification LoRA not only provides a robust foundation for creative expression but also explores a sustainable model that harmonizes the ethos of open access with the practical necessities of platform maintenance and development. In doing so, it exemplifies a forward-thinking approach to nurturing a vibrant, self-sustaining community of digital creators within the ByWassies ecosystem.
Deliverables
Month 1
A curated image database for training.
A general-purpose Hyperwassification LoRA Model.
Initial instructions for Model Usage helping users get the results they want.
Month 2 onward (dev and further funding needed):
- A user-friendlier method for community members to interact with the model.
This might entail a Discord Bot, a Tutorial for hosted Services, or a Branded Custom Interface.
Current Project Status
A preliminary Hyperwassification LoRA model has been trained, already showcasing promising results in generating unique Wassie art.
Showcasing the images generated so far has been met with an enthusiastic response through private and public channels. However, it's at an estimated 60% completion, with notable limitations such as image artifacts and challenges in rendering diverse poses due to a limited pose variety in the training dataset.
The current model performs well with frontal profile imagery and some meme-style poses, but to enhance its versatility and quality, further development is essential. This involves enriching the training dataset with varied poses and styles, iterative model refinement to reduce artifacts, and engaging with the ByWassies community for feedback.
The objective is to secure funding that enables dedicated development time to mature the model, improving its performance and broadening its capabilities to better serve the ByWassies community.
Other Information
Please note that I have already trained various Lora models, both public and private. My main showcase model currently is LasogetteXL. I invite you to visit the following links to check it out:
LasogetteXL - v1.1 | Stable Diffusion LoRA | Civitai
LasogetteXL Image Showcase
Adsogette - Collectible Sog Advertisement
I would also like you to consider that I already have full artist support for this project from yellowbrah. There is also of course cc0 wassies, and between these two collections of images there's already enough material to train a first model, although further variety would be welcome to wassieness.
I would love to have the team's help to find a way to incentivize more Wassie artists to help with the training database, be it through $$$$$, honor or threats of sacrifice to ze matriarch.
In all seriousness, I'd appreciate help in setting up an incentive structure, as I find it important to celebrate artists who push the envelope in this direction. B)
Should any Wassie artists be reading this, please feel free to contact me and I'll hear ya out!
Attached you will find images that have been generated with the preliminary model.
Thank you for reading! Luv u,
Lukas
Supporting Information
