Your first image

This walkthrough produces the hero image you see on the home page.

Code

from strands_sana import sana_generate

result = sana_generate(
    prompt=("a majestic rubber duck wearing aviator goggles, "
            "perched on a futuristic motorcycle in a cyberpunk neon-lit city, "
            "raining, photorealistic, 8k, dramatic cinematic lighting"),
    model="sana-1.5-1.6b-1024",
    steps=20,
    seed=42,
    output_dir="./out",
)
print(result)

What happens

1. Pipeline cache miss → downloads Efficient-Large-Model/SANA1.5_1.6B_1024px_diffusers (~10 GB) on first run.
2. Loads the pipeline on best available device (cuda → mps → cpu).
3. Runs 20 denoising steps with flow_match_euler scheduler.
4. Saves PNG to ./out/sana_<timestamp>.png.

{
  "status": "success",
  "path": "./out/sana_1779038153007.png",
  "model": "sana-1.5-1.6b-1024",
  "kind": "t2i",
  "prompt": "...",
  "width": 1024,
  "height": 1024,
  "count": 1
}

Result

→ Want it 10× faster? See Sana-Sprint.