カオスマップに戻る
Stable Diffusion

Stable Diffusion

Overview

Stability AI is the creator of Stable Diffusion, the pioneer of open-source image generation. Since releasing the first Stable Diffusion in 2022, they have led the community-driven image generation ecosystem. Their latest generation is Stable Diffusion 3.5, released in 2024.

Latest: Stable Diffusion 3.5

Three variants targeting different use cases and hardware:

ModelParamsUse CaseNotes
SD 3.5 Large8.1BProfessional1 megapixel, highest quality
SD 3.5 Large Turbo8.1B (distilled)Fast generation4-step generation
SD 3.5 Medium2.5BConsumerRuns on consumer GPU

Key Features

1. Professional Quality (SD 3.5 Large)

8.1B flagship. 1-megapixel resolution for professional use.

2. Fast Generation (SD 3.5 Large Turbo)

Distilled variant generating high-quality images in just 4 steps while maintaining prompt adherence.

3. Consumer Hardware (SD 3.5 Medium)

Improved MMDiT-X architecture. Runs on consumer GPUs (RTX 3060 and above).

4. Query-Key Normalization

New technology improving customizability and prompt adherence.

5. TensorRT Optimization

Up to 2.3x faster on NVIDIA RTX GPUs, 40% less VRAM.

6. Stable Video Diffusion

Short-clip generation from static images.

License

Stability AI Community License:

  • ✅ Commercial use allowed
  • ✅ Non-commercial use allowed
  • ✅ Fine-tuning / customization freely allowed
  • ⚠️ Enterprises with $1M+ annual revenue need Enterprise license

Strengths

  • Pioneer of open-source image generation
  • Commercial use allowed
  • Runs on consumer GPUs (SD 3.5 Medium)
  • Highly customizable (LoRA / fine-tuning)
  • 2x faster with TensorRT
  • Active community (Hugging Face, Civitai)

Weaknesses

  • Business instability (leadership changes in 2024)
  • Weaker text rendering than Midjourney
  • Limited video models
  • Weak text models (image-focused)

Official Resources

公式サイト