There are several versions of Stable Diffusion,
and each version has different performance, resolution, and architecture.
The major differences between versions are summarized as follows.
1. STABLE Diffusion 1.x
- SD 1.3, SD 1.5
- Resolution: Native 512 x 512
- Based on latent diffusion model
- Relatively light computational requirements
- There are many different community models (LoRA, DreamBooth, etc.).
- SD 1.5 offers improved quality over SD 1.4
2. Stable Diffusion 2.x
- SD 2.0, SD 2.1
- Resolution: Native 768 x 768
- Improve text recognition accuracy using OpenClip
- Capable of creating more detailed images than SD 1.x
- There were many complaints about NSFW filtering becoming stronger
in version 2.0, which was eased in 2.1.
3. Stable Diffusion XL (SDXL)
- SDXL 1.0, SDXL Turbo
- Resolution: Default 1024 x 1024 (significantly higher resolution than 1.x)
- More powerful text-to-image interpretation and improved image quality
- Multi-prompt support (allows for more sophisticated prompt interpretation)
- Improved natural facial and finger expressions
- SDXL Turbo is a version with increased real-time creation speed.
Summary of key differences by version
version | native resolution | understand text | image quality | special feature |
SD 1.4 / 1.5 | 512×512 | commonly | commonly | Most widely used, custom models galore |
SD 2.0 / 2.1 | 768×768 | Improved | improved | OpenCLIP application, NSFW filter controversy |
SDXL 1.0 | 1024×1024 | greatly improved | very excellent | Improved face and hand expressions, large model |
SDXL Turbo | 1024×1024 | greatly improved | very excellent | Significantly improved speed, real-time creation possible |
댓글 없음:
댓글 쓰기