Stability API

Converts natural language text prompts into images using Stable Diffusion models via REST API endpoints. The implementation accepts structured JSON payloads containing prompt text, negative prompts, and generation parameters (steps, guidance scale, seed), then routes requests through Stability's inference infrastructure which performs diffusion-based image synthesis. Supports multiple model versions (SDXL, SD3, etc.) with automatic model selection or explicit specification.

Accepts an existing image as input along with a text prompt and applies Stable Diffusion conditioning to transform the image while preserving structural elements based on a strength parameter (0-1 scale). The API encodes the input image into latent space, applies diffusion steps conditioned on both the image and prompt, then decodes back to pixel space. Strength parameter controls how much the original image influences the output: 0.0 preserves the original, 1.0 ignores it entirely.

Supports generation of images in multiple aspect ratios and resolutions (e.g., 512x512, 768x768, 1024x1024, 1024x576, 576x1024, etc.) through API parameters. The implementation adapts the diffusion model to generate images at specified dimensions without cropping or padding, enabling direct generation of images optimized for specific use cases (mobile, desktop, print, social media).

Provides specialized model variants trained on specific visual domains (photography, illustration, 3D rendering, anime, etc.) that can be selected to influence generation style without explicit style prompting. The API routes requests to domain-specific models based on selection, enabling consistent aesthetic output aligned with training data characteristics.

Exposes generation capabilities through RESTful HTTP endpoints with standardized JSON request/response payloads, authentication via API keys, and consistent error handling. The implementation follows REST conventions with POST endpoints for generation requests, GET endpoints for status/results, and structured error responses with detailed error codes and messages.

Enables selective image editing by accepting an image, a binary mask indicating regions to modify, and a text prompt describing desired changes. The API applies diffusion only to masked regions while keeping unmasked areas unchanged, using the prompt to guide content generation in those regions. Mask is typically provided as a grayscale image where white (255) indicates regions to inpaint and black (0) indicates regions to preserve.

Extends images beyond their original boundaries by accepting an image and specifying expansion parameters (left, right, top, bottom pixels), then generating new content that seamlessly blends with the original image edges. The implementation analyzes edge context and uses diffusion conditioning to synthesize plausible extensions that maintain visual coherence with the original image content and a provided prompt.

Increases image resolution (typically 2x, 4x, or custom factors) while enhancing detail and reducing artifacts using neural upscaling models. The API accepts an image and upscaling factor, applies learned upsampling that reconstructs high-frequency details, and returns a higher-resolution version. Implementation uses diffusion-based or super-resolution neural networks trained on high-quality image pairs.

Generates short video clips (typically 4-25 seconds) from text descriptions using video diffusion models. The API accepts a text prompt and optional parameters (duration, aspect ratio, seed), then applies temporal diffusion to generate frame sequences that form coherent video. Implementation extends image diffusion to the temporal domain, ensuring frame-to-frame consistency and smooth motion.

Provides access to multiple Stable Diffusion model variants (SDXL, SD3, SD1.5, specialized models) through a unified API interface with explicit model selection via request parameters. The implementation maintains a registry of available models with metadata (capabilities, performance characteristics, pricing), routes requests to appropriate inference endpoints, and handles version deprecation/updates transparently.

Exposes detailed control over diffusion process parameters including guidance scale (0-35), step count (10-150), sampler algorithm selection (DDIM, Euler, Heun, DPM++, etc.), seed specification for reproducibility, and model-specific parameters. The API accepts these parameters in request payloads and applies them during inference to enable precise control over generation quality, speed, and consistency.

Supports submitting multiple generation requests in batch mode with asynchronous processing and webhook callbacks or polling for results. The API accepts batch payloads containing multiple prompts/images, queues them for processing, and returns job IDs for tracking. Results are delivered via webhook callbacks or retrieved through polling endpoints, enabling efficient processing of large image volumes without blocking.

Accepts negative prompt text that specifies content to exclude from generated images, using inverse conditioning during diffusion to suppress unwanted elements. The API applies negative prompts as guidance signals that push the generation away from specified concepts, enabling fine-grained control over what should NOT appear in outputs alongside positive prompts.