text-to-image generation / latent diffusion
Stable Diffusion is a latent text-to-image diffusion model that generates images from text prompts using a UNet trained in latent space with a frozen CLIP text encoder.
Solves the same problem with a different architecture or at a different layer. Expect to rewrite the integration.
Use it when you want a maintained pip-installable library with pipeline APIs for Stable Diffusion and many other diffusion models instead of the original CompVis research scripts.
🤗 Diffusers: State-of-the-art diffusion models for image, video, and audio generation in PyTorch.
Use it when you need multilingual text prompts or Apache-2.0 licensing rather than Stable Diffusion's original license, though it is no longer actively developed.
Kandinsky 2 — multilingual text2image latent diffusion model
An earlier or more limited way to do the job. Still the right call when you need something small, proven, or CPU-only.
Use it if you specifically want DALL-E2's prior/decoder architecture in PyTorch rather than latent diffusion, accepting an unofficial and unmaintained codebase.
Implementation of DALL-E 2, OpenAI's updated text-to-image synthesis neural network, in Pytorch
These projects were analysed and named stable-diffusion among their alternatives. The relationship is not symmetric — how stable-diffusion rates them is a separate judgement, made when stable-diffusion is analysed in its own right.
calls stable-diffusion “Latent text-to-image diffusion model with a huge ecosystem; choose for easy deployment, community support, and looser compute needs than DALL-E 2.”
Implementation of DALL-E 2, OpenAI's updated text-to-image synthesis neural network, in Pytorch
calls stable-diffusion “The original Stable Diffusion reference implementation; choose it for legacy SD1.x research rather than a maintained framework.”
🤗 Diffusers: State-of-the-art diffusion models for image, video, and audio generation in PyTorch.
calls stable-diffusion “Choose when you need the legacy Stable Diffusion v1 latent-diffusion pipeline/weights; lower resolution and no Sana efficiency or video support.”
SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformer