Description
Stable Diffusion
What is Stable Diffusion?
Company and Funding
| Data | Information |
|---|---|
| Company | Stability AI Ltd |
| Headquarters | London, United Kingdom |
| Founded | 2019 |
| Current CEO | Prem Akkaraju (since June 2024) |
| Valuation | $1B (October 2022) |
| Total Funding | ~$231M - $299M |
| 2024 Revenue | ~$50M - $104M |
| Employees | ~186 |
Available Models (December 2025)
Stable Diffusion 3.5 (October 2024) - Latest Generation
| Model | Parameters | Resolution | Speed | VRAM |
|---|---|---|---|---|
| SD 3.5 Large | 8.1B | 1 megapixel | Standard | ~12GB |
| SD 3.5 Large Turbo | 8.1B | 1 megapixel | 4 steps (fast) | ~12GB |
| SD 3.5 Medium | 2.5B | 0.25-2 MP | Standard | 9.9GB |
| SD 3.5 Flash | - | Variable | Very fast | Low |
Previous Models
- SDXL 1.0 (July 2023): 3.5B parameters, 1024×1024 native
- SD 2.1: Legacy model
- SD 1.5: 860M parameters, 4GB VRAM, largest ecosystem (10,000+ fine-tuned models)
Technical Architecture
- Diffusion Models: Generates images by denoising random noise
- Three text encoders: OpenCLIP-ViT/G, CLIP-ViT/L, T5-xxl
- QK-Normalization: Improves training stability
- MMDiT-X (SD 3.5 Medium): Self-attention modules in first 13 layers
Pricing and Licenses (December 2025)
Community License (Free)
- Eligibility: Individuals and organizations with < $1M annual revenue
- Includes: SD 3.5 Suite, SDXL Turbo, Stable Audio Open, Stable Fast 3D
- Use: Unlimited commercial and non-commercial
Enterprise License
- Eligibility: Organizations with > $1M annual revenue
- Price: Custom (contact sales)
- Includes: Implementation support, custom model training
Stability AI API (Credits)
| Service | Credits/Image |
|---|---|
| Stable Image Ultra | Variable |
| Stable Image Core | Affordable |
| SD 3.5 Large | ~3.7¢ |
| SD 3.5 Large Turbo | More affordable |
| SDXL 1.0 | ~1.1¢ |
| SD 1.5 | ~0.6¢ |
Third-Party Platforms
- DreamStudio: Official Stability AI web interface
- Stable Assistant: Multimodal chatbot
- ComfyUI: Node-based local interface (free)
- Automatic1111: Popular WebUI (free)
- Replicate, Hugging Face, Fireworks: Alternative APIs
Main Features
Image Generation
- Text to image from natural language
- Image to image (img2img)
- Inpainting (fill areas)
- Outpainting (expand images)
- Upscaling (increase resolution)
- Control via ControlNets
SD 3.5 Strengths
- Improved text rendering in images
- Output diversity: people with different skin tones and features
- Style versatility: 3D, photography, painting, line art
- Superior prompt adherence
- Customization: Query-Key Normalization facilitates fine-tuning
Multimodality (Stability AI Ecosystem)
- Stable Video Diffusion: Video clips from images
- Stable Video 4D 2.0 (May 2025): Dynamic multi-angle videos
- Stable Audio 2.5 (Sept 2025): Enterprise audio generation
- SPAR3D: 3D models from images in < 1 second
Hardware Requirements (Self-Hosted)
| Model | Minimum GPU | VRAM | RAM | Storage |
|---|---|---|---|---|
| SD 1.5 | GTX 1060 | 4GB | 8GB | 5GB |
| SDXL | RTX 3060 | 8GB | 16GB | 15GB |
| SD 3.5 Medium | RTX 3070 | 10GB | 16GB | 20GB |
| SD 3.5 Large | RTX 4080 | 12GB+ | 32GB | 25GB |
Integrations and Partners
Cloud Platforms
- Amazon Bedrock (AWS)
- Azure AI Foundry (Microsoft)
- NVIDIA NIM
- Hugging Face
- Replicate
Enterprise Partners
- WPP: Strategic partnership and investment (March 2025)
- Electronic Arts (EA): Co-development of gaming models
- Universal Music Group: Music creation tools
- Warner Music Group: Responsible AI for music
- HubSpot: Integration in Breeze Content Agent
- Mercado Libre: GenAds for e-commerce
Enterprise Use Cases
| Company | Application | Result |
|---|---|---|
| HubSpot | Breeze Content Agent | +150% generation capacity |
| Mercado Libre | GenAds advertising | +25% CTR |
| EA | Game assets | In development |
Open Source and Community
- Hugging Face: Downloadable models, 10,000+ fine-tuned variants
- GitHub: Inference and training code
- ComfyUI: Node interface with customizable workflows
- Civitai: Community of models and LoRAs
- Discord: Official Stability AI community
Limitations
- Does not generate harmful, violent, or explicit content (with safeguards)
- Variable quality depending on prompt specificity
- Greater output variation with same seed (by design)
- Requires powerful hardware for large models
- Enterprise license required for companies > $1M revenue
Controversies
- Getty Images: Copyright lawsuit (partial Stability AI victory in Nov 2025)
- CEO Change: Emad Mostaque resigned in March 2024
- Financial challenges: Reported in 2024, resolved with new funding
Key Features
Open-source image generation runnable locally
Stable Diffusion 3.5 with 8.1B parameters
MMDiT architecture (Multimodal Diffusion Transformer)
Improved text rendering in images
Runs on consumer hardware (from 4GB VRAM)
Text to image from natural language
Image to image (img2img) and transformations
Inpainting and outpainting
Resolution upscaling
Control via ControlNets
Over 10,000 fine-tuned models available
Free community license (<$1M revenue)
Official API with credit system
QK-Normalization for stable fine-tuning
Output diversity without extensive prompting
Multiple styles: 3D, photography, painting, line art
Stable Video Diffusion for video generation
Stable Audio 2.5 for enterprise audio
SPAR3D for 3D models in seconds
Integration with AWS Bedrock, Azure, NVIDIA NIM
Use Cases
Digital art and illustration generation
Social media content creation
Marketing material design
Concept art for games and films
Product image generation
Video game asset creation
Character and scenario design
Rapid visual idea prototyping
Photo editing and retouching
Background and texture generation
Logo and branding creation
Architectural visualization
Book and publication illustrations
Storyboarding and previsualization
Custom model training
Generative AI research
Design variation generation
Automated advertising (GenAds)
Visual educational content
NFTs and digital collectible art
User Reviews
Related AIs
ChatGPT
OpenAI
ChatGPT by OpenAI is a versatile AI assistant that excels at natural conversation, content creation, and complex problem-solving. With advanced multimodal capabilities, it processes text, voice, and images to streamline productivity and creativity.
DALL-E
OpenAI
OpenAI AI image generation system including DALL-E 3 and the new GPT-Image-1, with text-to-image, editing, inpainting capabilities and up to 4K resolution, integrated in ChatGPT and available via API.
Jasper AI
Jasper AI Inc.
AI platform for marketing content creation with personalized Brand Voice, 50+ templates, SEO integration and team collaboration. Used by 20% of Fortune 500.