You have a photo that feels like a song — a sunset shot that should be a warm ambient track, a neon street that should be synthwave. Until recently, you'd describe it in words and hope the AI understood. Lyria 3 lets you skip the translation: feed it the actual image and it composes around it. This guide covers how image-to-music works in Lyria, how to get the best results, and where it's still limited.
The facts come from Google's official Lyria product page and model card (linked in Sources), plus practical usage notes.
The Short Answer
Yes — Lyria 3 (and Lyria 3 Pro) accepts up to 10 reference images as input to guide music generation. Google's Lyria page describes it directly: "Upload an image and ask Lyria to transform it into a custom high-fidelity track." You upload the image, add a text prompt, and Lyria composes a track that matches the visual's mood and context.
How Lyria 3 Image-to-Music Works
The flow is simple:
- Choose your image(s). Up to 10 per generation. They set the visual-musical context — mood, color palette feel, scene energy.
- Write a text prompt. The image steers the vibe; the prompt steers the specifics: genre, tempo, instrumentation, vocals.
- Generate. Lyria 3 Pro produces a structured track up to 3 minutes; base Lyria 3 produces ~30-second clips.
- Iterate. Swap images or refine the prompt to pull the track closer to your vision.
The image isn't a magic button — it's a creative anchor. The prompt still carries the detail work, and combining both gives you a level of direction that text-only tools can't match.
A Prompt Formula for Image-to-Music
Use the same core structure as text prompts, plus a reference to what the image should contribute:
[Describe the scene / mood from the image] + [Genre] + [Tempo & rhythm] + [Instruments] + [Vocal style] + [Length]Examples that work:
- Sunset beach photo →
warm ambient sunset feel, soft lo-fi beat, 70 bpm, Rhodes piano, no vocals, 3-minute track - Neon city street photo →
synthwave night-city energy, driving bass, 110 bpm, retro keys, female vocal, full song - Mountain landscape photo →
epic cinematic build from the valley view, strings and percussion, 90 bpm, instrumental, 3 minutes
Rule of thumb: describe the feeling the image gives you (warm, tense, wide, intimate) before the genre — that's what the image can't fully communicate on its own.
Where You Can Use Image-to-Music
- Gemini app — image inputs are part of Lyria in Gemini (18+, free base tier).
- Gemini API / Vertex AI — developers, image inputs documented as accepted inputs.
- Google Vids — score video footage with matching audio.
- Credit-based studios — including the Lyria 3 Pro generator on this site, which runs the model pay-as-you-go with direct download, no Google Cloud setup.
Common Mistakes to Avoid
- One image, zero prompt. The image anchors mood, but without a prompt you leave the arrangement to chance. Always add genre/tempo/arc.
- Too many conflicting images. Ten images of different moods produce mush. Use 1–3 with consistent feel, or use the batch to explore variations.
- Expecting exact audio translation. Lyria won't extract or reproduce audio from the image — it composes matching music. Don't expect a photo of a band to produce that band's song.
- Using copyrighted character imagery. Keep inputs to your own photos and original art; output policies and content filters apply to inputs too.
Where It's Still Limited
- No audio extraction. Images are mood references, not audio sources.
- Inputs are images only — no audio upload, no voice cloning in Lyria (per the model card).
- Output fidelity varies. Google's own limitations note says: "While Lyria can create a wide range of music, we're still working on improving key capabilities. You should always carefully check that the tracks you create fit your vision." Treat the first pass as a sketch, then iterate.
FAQ
Does Lyria 3 support image to music? Yes. Lyria 3 and Lyria 3 Pro accept up to 10 reference images as input, per Google's official Lyria page and model card.
How many images can I use in Lyria 3? Up to 10 reference images per generation — use fewer (1–3) for a coherent mood, or more to explore variations.
Can Lyria 3 turn a video into music? Google's materials primarily document image inputs. Video-to-music support depends on the surface (Vids uses it for footage scoring) — check your access path's current capabilities.
Does image input cost extra? Pricing is by generation on metered surfaces, not per image — but check your access path's pricing model (Gemini free tier, API metering, or credit studios). See our pricing page for the credit model.
Is image-to-music available on Lyria 3 Pro? Yes — image inputs are documented for the Pro model as well as base Lyria 3, per the official model card.
The Bottom Line
Lyria 3's image-to-music is a genuine creative on-ramp: the image anchors the mood, your prompt shapes the track, and iteration pulls it the rest of the way. Start with 1–3 consistent images, describe the feeling before the genre, and validate on a cheap clip before spending a full-song generation. You can try the full Lyria 3 Pro model with image inputs on pay-as-you-go credits via the generator here — and our Lyria 3 Pro guide has the complete capability breakdown.
Sources
- DeepMind — Lyria model page (image-to-music): https://deepmind.google/models/lyria/
- DeepMind — Lyria 3 model card: https://deepmind.google/models/model-cards/lyria-3/
- Google — Lyria 3 Pro announcement: https://blog.google/innovation-and-ai/technology/ai/lyria-3-pro/
- Google Cloud — Lyria 3 & Lyria 3 Pro on Vertex AI: https://cloud.google.com/blog/products/ai-machine-learning/lyria-3-and-lyria-3-pro-on-vertex-ai
- Google Cloud — Ultimate prompting guide for Lyria 3 Pro: https://cloud.google.com/blog/products/ai-machine-learning/ultimate-prompting-guide-for-lyria-3-pro
Update note: image-input limits and per-surface availability evolve with model releases — verify current behavior on your access path.
