Image To Song
Turns a photo into a song up to four minutes long: a vision-language model reads the image, and ACE-Step v1 3.5B generates the track from that reading plus your genre, mood, instrument and vocal choices.
Overview
Here a picture takes the place of the lyrics box. Wiro built this page around ACE-Step v1 3.5B, the open music model from ACE Studio and StepFun. A vision-language model first writes song lyrics from the uploaded image, and ACE-Step then generates the song from those lyrics together with the tags you select. You never type a word of the song yourself.
The two kinds of input do different jobs. Your tag selections go into every run exactly as chosen, so genre, instruments, mood and voice stay fixed while the picture changes.
What you can build
- A sung track for a product shot, a travel photo or a piece of artwork, with lyrics written from the image.
- One track per slide in a photo slideshow, or per image in a series of social posts.
- A "make a song from my photo" button in a gallery or camera app.
- Lyric sketches for a mood board: run each reference image with the same tags and compare the lyrics each one produces.
Inputs
- Image: upload a file or paste a public image URL. The prefilled example is a photo of a green woodland.
- Tags (required, several picks allowed in each list): genre, such as Celtic, K-pop or bluegrass; instrument, such as sitar, harp or drum machine; mood; voice, from male and female to soprano or child; and vocal timbre, such as warm or breathy.
- Length and seed: 1 to 240 whole seconds (one minute unless changed), with seed 42 preset.
- Advanced section: the sampler controls of ACE-Step v1, including step count (60 by default), guidance strength (15 by default), the guidance method and a granularity scale.
Outputs
Each run returns an MP3 and a text file with the lyrics written from your image. On this page the woodland photo is shown next to a one-minute stereo example at 48 kHz.
Recommended settings
When you compare pictures, hold everything else constant: same tags, same seed, same duration. Any difference you then hear comes from the lyrics written from each image.
Pick mood and genre to fit the scene. A stormy photo run with the default happy and Pop tags is still generated with those tags, so a darker result needs darker tags too.
The sampler defaults match those of the official ACE-Step demo. If a track has audible artifacts, raise the granularity scale; the demo describes higher values as a way to reduce them.
Limitations
- You cannot edit the lyrics written from your image before the song is generated, and there is no field for your own lyrics.
- The image only shapes the lyrics. Genre, instruments, mood and voice come from your tags alone.
- The weak points on the ACE-Step v1 card apply here too: wide variation between seeds and durations, rare instruments that may not render well, coarse vocals and weaker results in some genres.
Safety & compliance
ACE-Step v1 carries the Apache 2.0 license, and its authors rule out unlicensed copyrighted output, harmful content and passing AI music off as human work. They also ask that AI involvement be disclosed. If your image shows a real person, have their consent before uploading it.
API quick start
Run image-to-song-ACE-Step-v1-3.5B with a single API call.
{
"genre": "Pop",
"instrument": "guitar",
"mood": "happy",
"gender": "female"
}curl -X POST "https://api.wiro.ai/v1/Run/ace-step/image-to-song-ace-step-v1-3-5b" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"genre": "Pop",
"instrument": "guitar",
"mood": "happy",
"gender": "female"
}
JSONOther models you might like
Explore similar music generation models.



