Stable Diffusion

Stable Diffusion is a state-of-the-art deep learning model designed for generating high-quality images from textual descriptions. It utilizes a latent diffusion process, a type of generative model that iteratively refines noise to produce detailed and realistic images. This model is highly efficient, scalable, and capable of running on consumer-grade hardware, making it widely used for applications in art generation, content creation, and research. Developed with open accessibility in mind, Stable Diffusion empowers developers and creators to explore generative AI while maintaining adaptability for various use cases.

Apolo utilizes Stable Diffusion Web UI by AUTOMATIC1111 and StableStudio UI to make inference/have a UI playground.

Stable Diffusion WebUI is an intuitive, browser-based interface designed for generating and managing AI-generated images using Stable Diffusion. It offers customizable workflows, advanced image settings, and integrations for various Stable Diffusion models, enabling users to create and edit images with ease.

StableStudio is a web application built to enhance the Stable Diffusion experience. It provides minimalistic UI and batch inference capability.

Key Features

Feature	How the Apolo App Helps
Text‑to‑Image & Image‑to‑Image	Prompt → image or edit existing images via Web UI or REST.
Multiple UIs	Switch between AUTOMATIC1111 (power‑user) and StableStudio (lightweight).
Custom Models	Point to any Hugging Face model (stabilityai/stable-diffusion‑xl, lora/…).
GPU Presets	Pick from gpu-l4-x1, gpu-a100-x1, etc.; Apolo sets env vars & drivers.
Secrets Integration	Store HF tokens in Apolo Secrets instead of plain text.
Scalable Replicas	replica_count lets you handle bursty image queues.

Apolo Deployment

Deploy via either the Web Console UI or Apolo CLI.

Web Console UI

1 · Open the catalogue

2 · Configure the wizard

Section	Field	Example	Notes
Enable HTTP Ingress	auth = true	Adds basic‑auth to the public URL.
Resource Preset	gpu-a100-x1	1× NVIDIA A100 80 GB (or choose gpu-l4-x1, etc.).
Stable Diffusion	replica_count = 1	Increase for concurrent jobs.
Hugging Face Model	stabilityai/stable-diffusion-2-1-unclip-small	Any HF repo path.
HF Token	Secret HF_TOKEN	Needed for private or gated models.

Click Install. Wait until Status → healthy; copy the external_api host (REST) or open the Web UI at https://sd-<id>.apps.<cluster>.apolo.us.

Apolo CLI

Create a declarative YAML and push it with one command.

# stable-diffusion.yaml

template_name: "stable-diffusion"
template_version: "v25.5.0"
input:
  preset:
    name: "gpu-l4-x1"         # 1× L4 24 GB
  ingress_http:
    auth: false               # public playground
  stable_diffusion:
    replica_count: 2          # autoscale to 2 replicas
    hugging_face_model:
      model_hf_name: "stabilityai/stable-diffusion-2"
      hf_token:               # reference an Apolo Secret
        key: "HF_TOKEN"

# deploy
apolo app install -f stable-diffusion.yaml \
  --cluster <CLUSTER> --org <ORG> --project <PROJECT>

Explanation

• preset.name=gpu-l4-x1 requests one NVIDIA L4; Apolo auto‑sets CUDA_VISIBLE_DEVICES & default batch flags.

• replica_count=2 launches two pods behind a load balancer for higher QPS.

• hugging_face_model.model_hf_name points to SD‑XL‑Turbo for faster drafts.

• hf_token pulls the secret string from the HF_TOKEN key in Apolo Secrets.

Inputs / Outputs (schema v1)

Inputs

JSON Path	Default	Description
preset.name	–	GPU preset per replica.
ingress_http.auth	true	Basic‑auth gate.
stable_diffusion.replica_count	1	Number of pods.
stable_diffusion.hugging_face_model.model_hf_name	–	HF model repo.
stable_diffusion.hugging_face_model.hf_token	null	Secret or string token.

Outputs

Key	Purpose
external_api.*	Public REST endpoint (/v1/txt2img).
internal_api.*	In‑cluster endpoint for other workloads.
hf_model.model_hf_name	Echoes the loaded model.

Usage

1 · REST API

Once the app is healthy Apolo exposes auto‑generated Swagger docs at:

https://<APP_HOST>/docs

txt2img endpoint

curl -X POST \
  -H "Authorization: Basic $(apolo config show-token)" \
  -H "accept: application/json" \
  -H "Content-Type: application/json" \
  -d '{
        "prompt": "Some prompt"
      }' \

The response contains a base64‑encoded PNG under images[0].

2 · Web UIs

• AUTOMATIC1111 Web UI – the default root path (/).

• StableStudio – served at /stablestudio when enabled in the chart.

Connecting StableStudio

1. Open StableStudio and click Settings (gear icon, top right).

2. Paste the external URL of your Stable Diffusion deployment into Host URL.

3. Status should show Ready without history plugin.

4. Click Generate and enjoy! Note: The host URL is stored only in localStorage; each browser or private window needs to be configured again.

References

• StableDiffusion repository
• Stable diffusion webui Automatic repository
• StableStudio repository