AI Photo Forge

Generates personalized AI-rendered portraits of users through a Telegram bot interface, likely using a fine-tuned diffusion model or generative API (e.g., Stable Diffusion, DALL-E) triggered via Telegram's Bot API. The bot handles user authentication, session management, and image delivery directly within Telegram's chat interface, eliminating the need for external web dashboards. Architecture likely involves a Telegram bot webhook or polling mechanism that queues generation requests to a backend GPU cluster or third-party image generation service.

Analyzes user-provided reference images or initial portrait requests to extract facial features, style preferences, and identity markers, then conditions the generative model to produce portraits that maintain consistent likeness across multiple generations. This likely involves face detection (using OpenCV, MediaPipe, or a custom CNN), embedding extraction (via a face recognition model like FaceNet or ArcFace), and prompt augmentation that encodes facial characteristics into the diffusion model's latent space or as CLIP embeddings.

Queues image generation requests from Telegram users and processes them asynchronously on a backend GPU cluster or third-party API, then notifies users via Telegram message or callback when images are ready. Uses a message queue (likely Redis, RabbitMQ, or cloud task queue) to decouple request ingestion from generation processing, preventing Telegram API timeouts on long-running operations. Implements polling or webhook callbacks to deliver completed images back to users without blocking the bot.

Provides users with preset or custom controls to vary the artistic style, pose, lighting, and composition of generated portraits without requiring manual prompt engineering. Likely implemented via a button menu or slash command interface that maps user selections to pre-crafted prompt templates or LoRA (Low-Rank Adaptation) model weights that modify the base diffusion model's behavior. May support style categories like 'oil painting', 'anime', 'professional headshot', 'fantasy', etc., each with associated model weights or prompt augmentations.

Tracks user sessions, stores generation history, and manages user preferences within a persistent database (likely PostgreSQL or MongoDB). Associates each Telegram user ID with a session record containing reference images, style preferences, previous generations, and usage statistics. Enables users to retrieve past portraits, manage their generation quota, and maintain consistent personalization across multiple bot interactions without re-uploading reference images.

Implements per-user generation quotas and rate limits to control API costs, prevent abuse, and manage GPU resource allocation. Tracks user generation counts in a database or cache (Redis), enforces daily/monthly limits, and returns quota-exceeded errors when users exceed their allowance. May implement tiered quotas (free users get 5 generations/month, paid users get unlimited) and rate limiting (max 1 generation per 30 seconds) to prevent API spam and ensure fair resource distribution.

Provides a conversational, button-based interface within Telegram for users to initiate portrait generation, select styles, view results, and manage settings without typing commands. Uses Telegram's InlineKeyboardMarkup and callback query system to create interactive menus that guide users through the generation workflow. Buttons trigger backend actions (style selection, generation initiation, history retrieval) and update the Telegram message with results or new options.