README.md

metadata

title: Stable Diffusion Equation Playground
emoji: 🧪
colorFrom: indigo
colorTo: green
sdk: gradio
sdk_version: 5.22.0
python_version: '3.10'
app_file: app.py
pinned: false
license: mit
models:
  - stable-diffusion-v1-5/stable-diffusion-v1-5
preload_from_hub:
  - stable-diffusion-v1-5/stable-diffusion-v1-5

Stable Diffusion Equation Playground

A ZeroGPU-ready Gradio app for learning how Stable Diffusion works inside Diffusers.

Instead of only calling pipe(prompt), the app exposes a beginner-friendly custom denoising loop:

• three prompt embeddings can be blended with simple strength sliders
• the live equation shows the exact weighted embedding blend being used
• a few Diffusers levers are exposed: seed, denoising steps, prompt guidance, and optional noise mixing
• intermediate latent snapshots show how the image emerges across denoising steps

The core idea to teach: Stable Diffusion starts from noise in VAE latent space. The prompt does not literally become half of the pixels. The prompt changes the UNet's predicted noise at every denoising step, most commonly through classifier-free guidance:

guided = negative_prediction + guidance_scale * (prompt_prediction - negative_prediction)

This app focuses on StableDiffusionPipeline checkpoints such as Stable Diffusion 1.5 because the embedding pathway is straightforward for students. SDXL uses two text encoders plus pooled prompt embeddings, so the same idea carries over, but the code is more complex.

Running

On Hugging Face Spaces, select ZeroGPU hardware in the Space settings. The image generation function is decorated with @spaces.GPU, so it requests a GPU only while a generation is running.

For local development, install the requirements, use a CUDA or MPS machine, then run:

python app.py

The default model is stable-diffusion-v1-5/stable-diffusion-v1-5. The first generation downloads the checkpoint from Hugging Face and may take a few minutes.