Text-to-Image Pipeline with Stable Diffusion

Prerequisites: Python 3.10+; a CUDA-capable GPU (8GB+ VRAM recommended); comfort with pip and basic Python.

Targeted versions: Diffusers (current release, mid-2026); PyTorch matched to your CUDA version; Stability AI's SDXL 1.0 and Stable Diffusion 3.5 Large checkpoints — verified against Hugging Face and Stability AI documentation, mid-2026.

Diffusers is Hugging Face's official library for running diffusion pipelines, and it is the common backbone behind almost every open text-to-image tool you'll encounter, including ComfyUI and Automatic1111 under the hood. You'll build a small pipeline that takes a text prompt and produces an image, then extend it with negative prompts, a fixed seed for reproducibility, and an img2img refinement pass.

You'll work with two current Stability AI checkpoints: SDXL 1.0 (stabilityai/stable-diffusion-xl-base-1.0), the workhorse general-purpose model with a strong optional refiner stage, and Stable Diffusion 3.5 Large (stabilityai/stable-diffusion-3.5-large), a newer multimodal-diffusion-transformer model with noticeably better prompt adherence and text rendering. Both are loaded through the same DiffusionPipeline / AutoPipeline interface, so switching between them later is mostly a one-line change.