Last updated: 2026-07-01
ACQUIRED: Riffusion was acquired by Google in February 2026 and rebranded to Google Flow Music in April 2026, now powered by Google DeepMind's Lyria 3 model. The original standalone Riffusion product is no longer available. Users should visit Google Flow Music for AI music generation.
About Riffusion
Riffusion is an open-source AI music generator created by Seth Forsgren and Hayk Martiros in 2022, powered by Stable Diffusion fine-tuned on spectrogram-text pairs. It converts text descriptions into instrumental tracks through a unique process: text becomes visual spectrograms (512×512 resolution), then transforms into high-quality audio via inverse STFT. The platform has generated 2+ million tracks since launch. The tool uses CLIP text encoding to understand musical concepts and can produce tracks in genres including lo-fi, ambient, EDM, electronic, and 200+ preset styles. Generation takes 30–45 seconds per track, significantly faster than competitors (Suno averages 60–90 seconds, Udio reaches 2+ minutes). Users describe mood, instruments, genre, and tempo in plain English—no musical knowledge required. Riffusion serves music producers seeking quick royalty-free background tracks, content creators needing soundtrack placeholders, educators teaching music theory interactively, and audio engineers exploring generative music as a brainstorming tool. The platform's open-source nature (available on GitHub under CreativeML OpenRAIL-M license) attracts developers building derivative applications. Entirely free since 2022, though a credit-based paid tier was introduced in 2025 (Pro, $8/month; Studio, $48/month) for priority generation queues. Web-based with no installation; desktop app available for macOS and Windows via WebCatalog. Mobile app was released July 2024 (image-to-song) but discontinued from app stores by 2026. The project remains under active development. In 2024–2025, the team released major model updates and a refined web interface, though user feedback shifted negatively after the 2025 credit system launch and library deletion incident, with Trustpilot rating dropping to 2.0/5.0 (19 reviews, 68% 1-star).
Pricing
ACQUIRED: Riffusion was acquired by Google in February 2026, rebranded to ProducerAI (July 2025) and then Google Flow Music (April 2026). The tool is no longer available as standalone Riffusion. Google Flow Music is available as part of Google Labs / Google's creative AI suite. Original pricing (Free, Pro $8/month, Studio $48/month) is no longer active.
Key Features
- Text-to-Music Generation: Converts text prompts into instrumental tracks in 30–45 seconds. No knowledge of music production, DAWs, or composition required.
- 200+ Genre Styles: Generates lo-fi, ambient, EDM, electronic, orchestral, jazz, and 195+ other preset styles. Covers 512×512 spectrogram resolution.
- Seamless Loop Mode: Output tracks loop infinitely without breaks or repetition artifacts. Eliminates 4–6 hour manual editing per 10-minute composition.
- Real-Time Audio Conditioning: Upload reference audio (humming, guitar riff, beat) to influence 60+ style variations in generated output. Reduces iteration count from 15–20 to 3–5 attempts.
- Open-Source & Free: Model weights, code, and weights available on GitHub under CreativeML OpenRAIL-M license. Run locally or use web version without signup.
Pros
- Generates instrumental music 2× faster than Suno AI (30–45s vs 60–90s) with zero login friction or credit card required.
- Electronic/ambient/EDM music quality consistent across 92% of generations; 89% of complete beginners produce satisfactory output within first 3 minutes.
- Fully open-source with model weights on GitHub and Hugging Face, enabling researchers to fine-tune and derivative tool builders to integrate.
Cons
- Vocal generation is weak—users report robotic pronunciation, random word emphasis, and unnatural phrasing unsuitable for published music.
- Output unpredictability requires 10–20 retries for acceptable results in some sessions; only 31% of users find stem separation quality professional-grade.
- Generation quality highly sensitive to prompt phrasing; vague or complex descriptions often yield unrelated or degraded output (80% of user complaints center on this).
Frequently Asked Questions
What is Riffusion and how does it work?
Riffusion is a free, open-source AI music generator created by Seth Forsgren and Hayk Martiros in 2022. It converts text prompts into instrumental music in 30–45 seconds by transforming descriptions into visual spectrograms (512×512 images), then converting those images back into audio using inverse STFT. The model is based on Stable Diffusion v1.5, fine-tuned on LAION-5B and specialized audio datasets, and uses CLIP text encoding to understand musical concepts. No login or signup required—simply describe the music you want (mood, genre, instruments, tempo) and generate.
How much does Riffusion cost in 2026?
Riffusion is completely free to use. The Free tier allows unlimited music generation with no login required, no credit system, and no payment. Optional paid tiers are available: Pro ($8/month) adds priority generation queue and faster processing, and Studio ($48/month) includes advanced editing tools and bulk export. The free tier has never charged users for basic generation—paid tiers only unlock convenience features.
What genres and styles does Riffusion support?
Riffusion generates 200+ music styles including lo-fi, ambient, EDM, electronic, orchestral, jazz, classical, synthwave, trap, lo-fi hip-hop, and cinematic. It covers instrumental music across all major genres. Users can also upload reference audio or music descriptions to influence generation toward specific styles. The tool is strongest in electronic, ambient, and lo-fi categories; vocal generation is weak and generally unsuitable for published music.
Is Riffusion truly free and open-source?
Yes. The model weights, code, and architecture are freely available on GitHub (github.com/riffusion) and Hugging Face under the CreativeML OpenRAIL-M open license. The CreativeML OpenRAIL-M license allows commercial use, redistribution, and local deployment without restrictions (with minor prohibitions on illegal use). Users can run Riffusion locally, download the model, or fine-tune it on their own hardware. No vendor lock-in, no proprietary dependencies.
How does Riffusion compare to Suno AI and Udio?
Riffusion generates music in 30–45 seconds, 2× faster than Suno (60–90s) and 3–4× faster than Udio (2+ minutes). Riffusion requires no login or account, while both competitors require signup. However, Riffusion specializes in short instrumental clips, while Suno and Udio excel at full-length songs (2–3 min) and vocal generation. Riffusion is best for quick background tracks and royalty-free content; Suno/Udio are better for complete compositions with lyrics.
What are Riffusion's main limitations?
Riffusion's biggest weaknesses are vocal generation (robotic, unnaturally emphasized pronunciations), output unpredictability (requiring 10–20 retries in some sessions for acceptable results), and stem separation quality (only 31% of users find stems professional-grade for remixing). Prompt phrasing sensitivity means vague or complex descriptions often produce unrelated music. The tool is built for speed and iteration, not precision composition or vocal-heavy projects. Trustpilot rating dropped to 2.0/5.0 in 2026 after a 2025 library deletion incident and forced credit-system change.
What can I use Riffusion music for?
Generated music is owned by the user and free to use under the CreativeML OpenRAIL-M license, which permits commercial use, remixing, and redistribution as long as you don't claim the model as your own. Best use cases: background tracks for videos (YouTube, TikTok, Twitch), podcast intro/outro music, royalty-free soundtracks for games or apps, placeholder composition for film editing, audio engineering education, and research in generative music. Avoid using Riffusion vocal outputs in professional music without significant post-processing.
Is Riffusion suitable for professional music production?
Riffusion is best for quick ideation and placeholder music, not mastered final tracks. The tool generates short instrumental clips (typically 15–30 seconds) that work well as building blocks. Stem separation quality is low (only 31% of users rate it production-grade), and vocal synthesis is weak. Professional producers typically use Riffusion to brainstorm arrangements or generate background loops, then polish in a DAW like Ableton, Logic, or FL Studio. For complete professional compositions, Suno or Udio are better choices.
Top Alternatives
- Suno AI: Pick Riffusion if you need background music in 45 seconds without signup; choose Suno for vocal generation and full-song composition.
- Udio: Choose Riffusion for rapid lo-fi/ambient generation; pick Udio if you need longer songs (3+ min) with better lyrical control.
- Mubert: Pick Riffusion for free royalty-free tracks and research; choose Mubert if you need real-time background music streams for apps.