AI music from an image: Lyria 3.5 image_url mood still on Sume

What image_url does

The Music Router request has a required prompt of 1 to 5,000 characters and an optional image_url: a public HTTPS image, or null to clear. It acts as a mood still that accompanies the text. The default model is sume/music-auto, which routes to Lyria 3.5, and lyria-3.5 and lyria-3-pro can be named.

fal's Lyria 3.5 page lists the price as $0.1 per generation and describes MP3 output with lyrics carrying section markers.

A request

Send it to POST /v1/music-router/generate with an Idempotency-Key. There is no duration field, so say the length you want in the prompt.

{
  "prompt": "Warm lo-fi piano with soft rain, about 30 seconds",
  "image_url": "https://example.com/mood.jpg",
  "model": "sume/music-auto"
}

What it costs and returns

The Sume price is $0.125, the fal list price of $0.10 plus the Sume margin.

Getting better results

• Pick a still with a clear mood: a rainy window, a neon street, a sunlit kitchen.
• Keep the prompt concrete about instruments and tempo, since the image sets mood, not structure.
• Do not send negative_prompt; a non-empty value returns negative_prompt_unsupported.
• Send duration and you get a 400, so state length in words and trim with a split afterward.

Making a family of tracks

Because there is no seed field in the request, treat each generation as a fresh take. To build a family of related tracks, keep the same image and a stable stem in the prompt, and vary one thing at a time: tempo, lead instrument, or length. Tag each with metadata so you can find the keeper.

At $0.125 per generation, ten variations cost $1.25.

Where it fits

Use it when you have a product or scene image and want a bed that matches it, such as a still from a video you are about to cut. Generate, listen, and keep the one that fits. For fixed-length edits, see exact length music. Pricing is on the API pricing page.