DecentraModels
Decentralized AI model marketplace using Walrus and Dynamic - ending arbitrary censorship of AI
Project Description
DecentraModels is a comprehensive platform designed to solve the critical censorship crisis plaguing AI creative communities like Civitai, where 291.4 million downloads and 3.2 million users face increasing arbitrary content removal and account suspensions.
The Problem The AI model sharing ecosystem is dangerously centralized, with platforms like Civitai implementing increasingly restrictive policies that have resulted in: 586,800 reports leading to 730,000 images removed and 4,500 models deleted in 2024 alone Users losing access to months of creative work overnight with no appeals process Community fragmentation as creators scatter to smaller, unreliable platforms Payment processor pressure forcing removal of legitimate AI art content
The Solution DecentraModels leverages Walrus decentralized storage to create a censorship-resistant alternative that provides:
Core Features: Permanent Model Storage: AI models (LoRAs, checkpoints, embeddings) stored as immutable blobs on Walrus with cryptographic integrity verification Community-Driven Discovery: Advanced model categorization with trending algorithms, user ratings, and collaborative filtering Creator Tools: One-click uploads to Walrus, detailed model documentation, and crypto-based monetization without payment processor dependencies Decentralized Training: P2P network for model fine-tuning with 1,247+ GPUs across 342 nodes, supporting FLUX, Illustrious XL, NoobAI XL, and other popular base models
Technical Architecture: Smart Model Organization: Models categorized into AI Lab Models (official releases), Community Checkpoints (popular variants), and Trending sections with progressive loading Real-Time Updates: Dynamic content discovery with "NEW" badges for models under 7 days old and weekly growth tracking for trending algorithms Mobile-Optimized Interface: Responsive design with compact model cards (64px height) and efficient progressive disclosure
How it's Made
Deleted User deadbeef xarestotoe Do Not Disturb
Deleted User deadbeef [TOF] — 6/11/25, 10:32 PM I deleted a message of a broken link Deleted User deadbeef [TOF] — 6/11/25, 10:51 PM Very weird Image Deleted User deadbeef started a call that lasted an hour. — 6/12/25, 1:30 AM Deleted User deadbeef [TOF] — 7/20/25, 4:32 PM Image Image https://sdmntprcentralus.oaiusercontent.com/files/00000000-6538-61f5-8358-821b4f59d702/raw?se=2025-07-20T21%3A15%3A46Z&sp=r&sv=2024-08-04&sr=b&scid=08d99320-7e5c-533e-b5af-37a4ec0e98f8&skoid=24a7dec3-38fc-4904-b888-8abe0855c442&sktid=a48cca56-e6da-484e-a814-9c849652bcb3&skt=2025-07-20T01%3A22%3A35Z&ske=2025-07-21T01%3A22%3A35Z&sks=b&skv=2024-08-04&sig=c%2Ba%2Bo6%2B1FtDKYJrRSD1Y57Ib1aDCY43A9P4MncqHT8Y%3D [ [ { "name": "Mid-Century Sofa", "price": 799, "properties": { Expand message.txt 6 KB bottxrnif — 7/20/25, 7:38 PM https://cdn.klingai.com/bs2/upload-kling-api/3045058478/image2video/ChDGTWh2CYoAAAAAAOHKWg-0_raw_video_1.mp4 Deleted User deadbeef [TOF] — 7/20/25, 9:00 PM https://screen.studio/share/CufkXF4F Google Chrome — Screen Studio Google Chrome — Created and shared with Screen Studio Google Chrome — Screen Studio bottxrnif — 7/26/25, 6:36 PM Image bottxrnif — 8/12/25, 5:06 PM Image Image Image Image Image Deleted User deadbeef [TOF] — 8/12/25, 5:17 PM ok Image Image Image Deleted User deadbeef [TOF] — 8/12/25, 5:24 PM Image Image Image Deleted User deadbeef [TOF] — 8/12/25, 5:37 PM Image bottxrnif — 8/12/25, 5:44 PM question 5 find il t not vt show me graph for 1 Deleted User deadbeef [TOF] — 8/12/25, 5:49 PM Image Image Image Image Image bottxrnif — 8/12/25, 6:07 PM 1d is wrong Deleted User deadbeef [TOF] — 8/12/25, 6:10 PM Seems right to me y(t)=R(t-4)-R(t-5)-R(6-t)+R(7-t)-1 Right? Must be it Image Got it? Deleted User deadbeef [TOF] — 8/12/25, 6:19 PM Figure out what this means Image I think this is fine Right? I made too many mistakes But this is correction Are you there? These are all the answers to 1a, 1b, 1c The other answers were all hallucinations Deleted User deadbeef [TOF] — 8/12/25, 6:27 PM 1d is R(t-4)-R(t-5)-R(6-t)+R(7-t)-1 Deleted User deadbeef [TOF] — 8/13/25, 7:15 PM ok You missed a call from bottxrnif that lasted a few seconds. — 8/13/25, 7:15 PM Deleted User deadbeef started a call that lasted an hour. — 8/13/25, 7:15 PM Deleted User deadbeef [TOF] — 8/13/25, 7:15 PM https://civitai.com/models/1865359?modelVersionId=2111234 Wan 2.2 SamDoesArts T2I LoRA - v1.0 | WanVideo-22-T2V-A14B LoRA | C... text to image in the style of sam yang (samdoesarts) https://huggingface.co/lopkiloinm/wan22samdoesarts/tree/main trigger phrase: in the style of s... bottxrnif — 8/13/25, 7:32 PM ltc1q3kw4mjxp6g2yhmezturra0jgt38mg98pyds5dz Deleted User deadbeef [TOF] — 8/13/25, 7:45 PM https://fal.ai/models/fal-ai/wan/v2.2-a14b/text-to-image/lora/playground?fromTraining=874e62a1-b844-45b3-af3f-b7dcfbcc6c7a Login to fal.ai | Access 200+ Generative AI Models Login to fal.ai to access FLUX, video generation, and 200+ AI models. Free tier available. Start building with generative AI APIs today. in the style of samdoesarts bottxrnif — 8:54 AM Technology Stack DecentraModels is built as a modern web application using a carefully selected tech stack optimized for performance and user experience:
Frontend Framework: React 18 with TypeScript for type-safe component development Vite for lightning-fast development and optimized production builds TailwindCSS for consistent, utility-first styling with custom design system Dynamic for login
Blockchain Integration: Dynamic Labs SDK for multi-chain wallet authentication (Ethereum, Solana support) Wagmi & Viem for Ethereum interactions and smart contract integration Walrus SDK (planned) for decentralized storage operations
Data Management: CSV-based model database with real-time loading and caching mechanisms Progressive data loading - initial display of 26 models (6 trending + 8 lab + 12 community) with lazy loading for performance Smart categorization system distinguishing between verified AI lab models and community-created checkpoints
Model Verification System: The platform exclusively uses verified, real AI models sourced from legitimate platforms, eliminating the common problem of "hallucinated" or non-existent models. This includes: FLUX.1 (Dev, Schnell) from Black Forest Labs Stable Diffusion variants from Stability AI Community favorites like Pony Diffusion V6 XL (250M+ downloads) Real training options based on fal.ai research
Performance Optimizations: Time to first action: Reduced from 30-45 seconds to under 10 seconds through smart initial loading Compact UI design: Model cards reduced from 200px+ to 64px height with hover expansion Intelligent search: Multi-field search across name, platform, category, and specialty with debounced input
Decentralized Training Network: Unlike centralized alternatives, DecentraModels implements a P2P training system that: Provides realistic hardware requirements ("Any GPU with 6GB+ VRAM") instead of unrealistic specs Offers 11 different training types (LoRA, DreamBooth, model merging) for various use cases Shows real network statistics with GPU distribution visualization
Smart Model Organization Algorithm: The platform implements a sophisticated categorization system that automatically sorts models by source (lab vs community), applies trending algorithms based on weekly growth metrics, and uses progressive disclosure to prevent information overload.
Walrus Integration Strategy: Each model is stored as an immutable blob with unique cryptographic IDs, providing both permanent availability and tamper-proof verification. The metadata and community interactions are managed via Sui smart contracts, creating a hybrid architecture that combines decentralized storage with efficient query capabilities.
Mobile-First Responsive Design: The interface adapts from desktop grid layouts to mobile-optimized lists with touch-friendly interactions, ensuring a ccessibility across all device types with consistent performance.
The platform represents a complete reimagining of AI model sharing infrastructure, moving from centralized platforms vulnerable to censorship toward a truly decentralized, community-owned ecosystem that preserves the creative freedom essential to AI art communities. Collapse message.txt 4 KB DecentraModels is a comprehensive platform designed to solve the critical censorship crisis plaguing AI creative communities like Civitai, where 291.4 million downloads and 3.2 million users face increasing arbitrary content removal and account suspensions.
The Problem The AI model sharing ecosystem is dangerously centralized, with platforms like Civitai implementing increasingly restrictive policies that have resulted in: 586,800 reports leading to 730,000 images removed and 4,500 models deleted in 2024 alone Users losing access to months of creative work overnight with no appeals process Community fragmentation as creators scatter to smaller, unreliable platforms Payment processor pressure forcing removal of legitimate AI art content
The Solution DecentraModels leverages Walrus decentralized storage to create a censorship-resistant alternative that provides:
Core Features: Permanent Model Storage: AI models (LoRAs, checkpoints, embeddings) stored as immutable blobs on Walrus with cryptographic integrity verification Community-Driven Discovery: Advanced model categorization with trending algorithms, user ratings, and collaborative filtering Creator Tools: One-click uploads to Walrus, detailed model documentation, and crypto-based monetization without payment processor dependencies Decentralized Training: P2P network for model fine-tuning with 1,247+ GPUs across 342 nodes, supporting FLUX, Illustrious XL, NoobAI XL, and other popular base models
Technical Architecture: Smart Model Organization: Models categorized into AI Lab Models (official releases), Community Checkpoints (popular variants), and Trending sections with progressive loading Real-Time Updates: Dynamic content discovery with "NEW" badges for models under 7 days old and weekly growth tracking for trending algorithms Mobile-Optimized Interface: Responsive design with compact model cards (64px height) and efficient progressive disclosure Decentralized AI model marketplace using Walrus and Dynamic - ending arbitrary censorship of AI bottxrnif bottxrnif
Technology Stack DecentraModels is built as a modern web application using a carefully selected tech stack optimized for performance and user experience:
Frontend Framework: React 18 with TypeScript for type-safe component development Vite for lightning-fast development and optimized production builds TailwindCSS for consistent, utility-first styling with custom design system Dynamic for login
Blockchain Integration: Dynamic Labs SDK for multi-chain wallet authentication (Ethereum, Solana support) Wagmi & Viem for Ethereum interactions and smart contract integration Walrus SDK (planned) for decentralized storage operations
Data Management: CSV-based model database with real-time loading and caching mechanisms Progressive data loading - initial display of 26 models (6 trending + 8 lab + 12 community) with lazy loading for performance Smart categorization system distinguishing between verified AI lab models and community-created checkpoints
Model Verification System: The platform exclusively uses verified, real AI models sourced from legitimate platforms, eliminating the common problem of "hallucinated" or non-existent models. This includes: FLUX.1 (Dev, Schnell) from Black Forest Labs Stable Diffusion variants from Stability AI Community favorites like Pony Diffusion V6 XL (250M+ downloads) Real training options based on fal.ai research
Performance Optimizations: Time to first action: Reduced from 30-45 seconds to under 10 seconds through smart initial loading Compact UI design: Model cards reduced from 200px+ to 64px height with hover expansion Intelligent search: Multi-field search across name, platform, category, and specialty with debounced input
Decentralized Training Network: Unlike centralized alternatives, DecentraModels implements a P2P training system that: Provides realistic hardware requirements ("Any GPU with 6GB+ VRAM") instead of unrealistic specs Offers 11 different training types (LoRA, DreamBooth, model merging) for various use cases Shows real network statistics with GPU distribution visualization
Smart Model Organization Algorithm: The platform implements a sophisticated categorization system that automatically sorts models by source (lab vs community), applies trending algorithms based on weekly growth metrics, and uses progressive disclosure to prevent information overload.
Walrus Integration Strategy: Each model is stored as an immutable blob with unique cryptographic IDs, providing both permanent availability and tamper-proof verification. The metadata and community interactions are managed via Sui smart contracts, creating a hybrid architecture that combines decentralized storage with efficient query capabilities.
Mobile-First Responsive Design: The interface adapts from desktop grid layouts to mobile-optimized lists with touch-friendly interactions, ensuring a ccessibility across all device types with consistent performance.
The platform represents a complete reimagining of AI model sharing infrastructure, moving from centralized platforms vulnerable to censorship toward a truly decentralized, community-owned ecosystem that preserves the creative freedom essential to AI art communities.
