Stablecog

Converts natural language text prompts into images by executing Stable Diffusion model inference on backend servers, supporting multiple model versions (including SDXL) with configurable generation parameters. The system processes prompts through a queue-based architecture that respects per-plan parallelization limits (0-4 concurrent generations), returning generated images in PNG/JPEG format within seconds to minutes depending on subscription tier and server load.

Accepts an uploaded image as input and generates new variations or style-transformed versions by conditioning Stable Diffusion's latent diffusion process on the input image features. The system preserves structural elements from the source while applying new artistic styles or modifications based on accompanying text prompts, enabling creative remixing without full regeneration from scratch.

Tracks monthly image generation quota per user account, enforcing hard limits that prevent generation requests exceeding the plan's monthly allocation. The system maintains quota state across sessions and devices, deducting credits per image generated and rejecting requests when quota is exhausted. Users can view remaining quota through the web UI or API and purchase additional credits if needed.

Provides access to multiple Stable Diffusion model checkpoints (including base models and SDXL variants) that users can select per-generation request, enabling comparison of model outputs and selection of the best-fit model for specific use cases. The system abstracts model loading and inference orchestration, allowing users to switch between models without managing local weights or CUDA environments.

Implements a monthly credit allocation system where users purchase plans (Free, Starter, Pro, Ultimate) that grant fixed monthly image generation quotas (20-12,000 images/month) and parallel generation limits (0-4 concurrent requests). The system enforces per-plan rate limiting and quota tracking, preventing overages and requiring plan upgrades or additional credit purchases for increased capacity. Credits do not roll over monthly, enforcing monthly budget cycles.

Implements differential privacy policies where free-tier generated images are stored publicly and visible to other users, while paid-tier images are stored privately and accessible only to the generating user. The system enforces this visibility policy at storage and retrieval layers, enabling commercial use only on paid plans where privacy is guaranteed.

Exposes image generation capabilities through HTTP REST endpoints that accept text prompts, image uploads, and model selection parameters, returning generated images with metadata. The API enforces per-plan rate limiting and quota tracking, rejecting requests that exceed monthly allocations or concurrent parallelization limits. Authentication uses API keys tied to user accounts, enabling programmatic access without web UI.

Supports generating multiple images in a single request (up to 4 images per batch) with concurrent execution limited by plan tier (0-4 parallel generations). The system queues requests and distributes them across available GPU resources, respecting per-plan parallelization caps to ensure fair resource allocation. Batch results are returned as a collection with individual image metadata.

Provides a browser-based interface at https://stablecog.com with text input fields for prompts, image upload for img2img, model selection dropdowns, and generation parameter controls. The UI queues requests and displays generated images in a gallery view with metadata. The interface is designed for interactive experimentation rather than production automation, with visual feedback on generation progress and quota consumption.

Generated images are delivered without watermarks or attribution requirements, and paid-tier users receive explicit commercial use rights enabling monetization, resale, or incorporation into products. The system does not embed metadata or licensing restrictions into image files, allowing full ownership and modification of generated assets.

Implements performance differentiation where free-tier generations are marked as 'slow' and paid tiers receive 'fast' generation speeds, likely through queue prioritization and GPU resource allocation. The system does not expose specific latency metrics but implies that paid users receive priority queue placement and faster GPU access, reducing wait times for generation completion.