☁️ Replicate Models Catalog
Pre-Curated Best-in-Class Audio AI Models - Ready to Deploy
🎯 Why Replicate Matters
Replicate has already done the hard work of identifying, hosting, and optimizing the hottest audio AI models. Their catalog represents a curated "best of" list with proven usage (millions of runs). This is our rapid prototyping playground.
⚠️ The Challenge: Cost at Scale
Replicate is perfect for MVP and validation, but costs compound quickly at scale. At 100K users generating 5 tracks/month each:
- MusicGen @ $0.02/run: $10,000/month
- ACE-Step @ $0.017/min: $8,500/month
- Total audio AI: $20,000-50,000/month
Strategy: Use Replicate to validate which models users love, then migrate those to self-hosted on Fly.io GPU or Modal.
🎵 Music Generation
Tier 1: Full Song Generation (with Vocals)End-to-end song creation with lyrics, vocals, and full instrumentation.
| Model | Creator | Runs | Description | Pricing |
|---|---|---|---|---|
| minimax/music-01 | MiniMax | 470.4K | Up to 1 min of music with lyrics & vocals matching reference styles | ~$0.03/song |
| minimax/music-1.5 | MiniMax | 22.1K | Full-length songs up to 4 mins with natural vocals & rich instrumentation | ~$0.03/song |
| lucataco/ace-step | lucataco | 80.2K | State-of-the-art music foundation model. 4 min in 20s. Lyrics support. | ~$0.017/min |
High-quality instrumentals from text prompts - the bread and butter of beat making.
| Model | Creator | Runs | Description | Pricing |
|---|---|---|---|---|
| google/lyria-2 | 49.3K | 48kHz stereo audio from text. SynthID watermarked. 30s output. | $0.0001/sec | |
| meta/musicgen | Meta | 3.2M | Generate music from prompt or melody. 300M-3.3B params. Industry standard. | ~$0.02/run |
| stability-ai/stable-audio-2.5 | Stability AI | 9.9K | High-quality music and sound from text prompts | Variable |
| riffusion/riffusion | Riffusion | 1.1M | Music via spectrogram diffusion. Real-time capable. | ~$0.039/run |
| lucataco/magnet | lucataco | 2.8K | Non-autoregressive transformer music generation | Variable |
| stackadoc/stable-audio-open-1.0 | stackadoc | - | Open source Stable Audio for short samples & sound effects | Variable |
Music generation with harmonic control - specify chord progressions and tempo.
| Model | Creator | Runs | Description |
|---|---|---|---|
| sakemin/musicgen-stereo-chord | sakemin | 3.3K | Generate stereo music restricted to chord sequences and tempo |
| sakemin/musicgen-chord | sakemin | 3K | MusicGen with chord progression input |
| sakemin/musicgen-remixer | sakemin | 18.3K | Remix existing music with MusicGen |
| pollinations/music-gen | Pollinations | - | Music generation variant |
| Model | Creator | Runs | Description |
|---|---|---|---|
| zsxkib/flux-music | zsxkib | 8.7K | Music generation with Flux architecture |
| andreasjansson/loop-gen | andreasjansson | - | Generate fixed-BPM loops from text prompts |
🗣️ Text-to-Speech / Voice Synthesis
Tier 1: Production-Grade TTSIndustry-leading text-to-speech with emotion control and voice cloning.
| Model | Creator | Runs | Description | Pricing |
|---|---|---|---|---|
| minimax/speech-02-turbo | MiniMax | 7.1M | Real-time TTS with emotion, 30+ languages, 300+ voices | $30/1M chars |
| minimax/speech-02-hd | MiniMax | 1.2M | High-fidelity TTS for voiceovers/audiobooks | $50/1M chars |
| resemble-ai/chatterbox | Resemble AI | 193.7K | Expressive speech, emotion control, instant voice cloning | Variable |
| resemble-ai/chatterbox-multilingual | Resemble AI | 5.7K | 23 languages, voice cloning, cross-language transfer | Variable |
| jaaari/kokoro-82m | jaaari | 69.5M | Lightweight 82M param TTS based on StyleTTS2 | Cheap |
Zero-shot voice cloning and specialized speech synthesis.
| Model | Creator | Runs | Description |
|---|---|---|---|
| lucataco/xtts-v2 | lucataco | 4.6M | Coqui XTTS v2 - multilingual voice cloning |
| chenxwh/openvoice | chenxwh | 80.8K | MyShell OpenVoice - zero-shot voice cloning |
| adirik/styletts2 | adirik | 132K | StyleTTS2 - style-based TTS |
| suno-ai/bark | Suno AI | 303.2K | Generates speech, music, sound effects |
| afiaka87/tortoise-tts | afiaka87 | 173K | High-quality slow TTS with voice cloning |
| lucataco/orpheus-3b-0.1-ft | lucataco | 32.7K | Orpheus 3B - Llama-based expressive TTS |
| x-lance/f5-tts | x-lance | 37.3K | F5-TTS model |
| zsxkib/dia | zsxkib | 10K | 1.6B dialogue TTS with voice cloning |
| lucataco/csm-1b | lucataco | 1.1K | Sesame CSM - conversational speech model |
| Model | Creator | Runs | Description |
|---|---|---|---|
| microsoft/vibevoice | Microsoft | - | Long-form multi-speaker podcast generation (up to 90 min, 4 speakers) |
| cjwbw/voicecraft | cjwbw | 10.7K | VoiceCraft speech editing |
| cjwbw/parler-tts | cjwbw | 2.7K | Parler TTS |
| cjwbw/seamless_communication | cjwbw | 91.9K | Meta's seamless translation + TTS |
| awerks/neon-tts | awerks | 173.1K | Neon TTS |
| minimax/voice-cloning | MiniMax | 25.4K | 10-second voice cloning |
🎤 Singing Voice & RVC
Voice conversion for creating AI covers and custom vocal performances.
| Model | Creator | Runs | Description |
|---|---|---|---|
| zsxkib/realistic-voice-cloning | zsxkib | 1.3M | Create song covers with RVC v2 AI voice |
| pseudoram/rvc-v2 | PseudoRAM | 1.3M | Speech-to-speech with RVC v2 |
| replicate/train-rvc-model | Replicate | 397.7K | Train custom RVC models |
| zsxkib/create-rvc-dataset | zsxkib | 18.6K | Create RVC dataset from YouTube |
| lucataco/singing_voice_conversion | lucataco | 1.1K | Amphion DiffWaveNetSVC |
| nateraw/autotune | nateraw | 605 | Pitch correction |
🧠 Audio Understanding
Analyze, caption, and understand audio content with AI.
| Model | Creator | Runs | Description |
|---|---|---|---|
| zsxkib/kimi-audio-7b-instruct | zsxkib | - | Kimi-Audio: speech-to-text, audio Q&A, captioning, emotion tags, voice responses |
🎛️ FlowState DAW Recommendations
Best model picks for hip-hop production workflows.
| Use Case | Best Model | Why |
|---|---|---|
| Full beat generation | ACE-Step or Music-1.5 | Fast, full songs with vocals |
| Instrumental loops | MusicGen or Lyria-2 | Proven, high quality, millions of runs |
| Chord-based backing | musicgen-stereo-chord | Control over harmony |
| Scratch vocals | Chatterbox or Orpheus | Expressive, cloneable |
| Voice cloning for hooks | OpenVoice or XTTS-v2 | Zero-shot cloning |
| Vocal covers/AI voices | RVC v2 | 1.3M+ runs, proven |
| Audio understanding | Kimi-Audio-7B | Analyze audio, Q&A |
💰 Cost Comparison (per generation)
Lyria-2
$0.003
30 sec output
ACE-Step
$0.017
60 sec output
Music-1.5
$0.03
up to 4 min
MusicGen
$0.02
30 sec output
Riffusion
$0.039
variable
Speech-02-Turbo
$0.03
per 1K chars
🔧 Self-Hosting Strategy
Replicate shows us what works. Now we need to run it cheaper.
📊 Phase 1: Validate with Replicate
Use Replicate during MVP to identify which models users actually use. Track generation counts per model type.
🎯 Phase 2: Identify Top 3
Find the 3 models that account for 80% of usage. These are candidates for self-hosting.
🖥️ Phase 3: Self-Host Winners
Deploy top models on Fly.io GPU ($0.50-2/hr) or Modal. 10-50x cost reduction at scale.
⚖️ Phase 4: Hybrid Architecture
Self-host high-volume models, use Replicate for long-tail/experimental features.
Self-Hosting Candidates (Open Source)
| Replicate Model | Open Source Version | Self-Host Cost | Savings at Scale |
|---|---|---|---|
| meta/musicgen | AudioCraft (GitHub) | ~$0.001/run | 95% |
| lucataco/ace-step | ACE-Step (GitHub) | ~$0.002/min | 88% |
| resemble-ai/chatterbox | Chatterbox (GitHub) | ~$0.0005/run | 90%+ |
| lucataco/xtts-v2 | Coqui TTS (GitHub) | ~$0.0003/run | 95%+ |
| pseudoram/rvc-v2 | RVC WebUI (GitHub) | ~$0.001/run | 90% |
Key Insight: Every popular Replicate model has an open-source version. Replicate's value is convenience, not exclusivity. At 10K+ users, self-hosting the top 3 models could save $5,000-15,000/month.
📚 Sources
- replicate.com/collections/ai-music-generation
- replicate.com/collections/text-to-speech
- replicate.com/collections/sing-with-voices
- replicate.com/lucataco/ace-step
- replicate.com/blog/minimax-text-to-speech
Research compiled December 2025. Run counts and pricing subject to change. Check Replicate for current pricing.