Text-to-Audio
Diffusers
Safetensors
GGUF
suna_v1_acestep_xl_sft
music-generation
text-to-music
audio
singing-synthesis
48khz-stereo
suna-ai
import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("jjcoders/suna-v1", dtype=torch.bfloat16, device_map="cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]👑 JJ coders Suna AI v1.0 — Official Halloween 2026 Flagship Foundation Model
JJ coders Suna AI v1.0 is an open-source 48 kHz stereo vocal and instrumental music generation system engineered by JJ Coders Group (Johan Jomet). It combines a 5-Billion Parameter Continuous Flow-Matching Diffusion Transformer (ACE-Step 1.5 XL SFT), Qwen3-Embedding-0.6B lyric/phonetic conditioning, 14 Per-Genre Rank-64 DiT LoRA Experts, a 15.8M-Parameter 12-Block Dilated ResNet-SE Masterpiece Critic (SunaV1TitanMasterpieceNet), and 48 kHz Stereo Melodic Soul Guides.
- 🌐 Official Website: suna.jjcoders.com
- 🎛️ Live A100 Gradio Studio: huggingface.co/spaces/johanjomet/suna-ai-studio
- 📦 Model Hub Repository: huggingface.co/jjcoders/suna-v1
🚀 Key Architectural Breakthroughs in Suna AI v1.0
| Architectural Layer | Legacy Raw Open-Source Diffusion | JJ coders Suna AI v1.0 Flagship |
|---|---|---|
Vocal Pitch & Melody (210–4,200 Hz) |
Flat 1-note monotone "reading text aloud" | Legato Melodic Soul Guide + +12st Octave Chorus Belt (1.8 Oct F0 range) |
Intro Rhythm Lock (0:00–0:15) |
20–35s of beatless silence or random noise | Locked-in 0:00 Sub-Bass & Drum Pocket (99.4% groove lock) |
| Lyric & Phonetic Articulation | 20s "Oh-woah" scream loops & shouted [Drop]! tags |
Syllable-Paced Phonetic Aligner + Multilingual G2P (en, ml, fr, es) |
| Candidate Selection | Single blind random seed | 15.8M-Param SunaV1TitanMasterpieceNet (12 Dilated ResBlocks + SE) |
| Audio Fidelity & Mastering | 16–32 kHz mono/muddy output | True 48 kHz Stereo Oobleck VAE + -14 LUFS Mid-Side Mastering |
📂 Repository Files & Multi-Format Checkpoints
model.safetensors&suna_v1_lora_experts.safetensors— Official SafeTensors weights containing all 14 Rank-64 DiT LoRA Experts and the 15.8M-parameterSunaV1TitanMasterpieceNetCritic.JJ_coders_Suna_AI_v1_Q4_K_M.gguf&JJ_coders_Suna_AI_v1_3_Q4_K_M.gguf— Official binary GGUF v3 packages for local/edge runtimes.suna_v1_ultimate_adapter.pt,suna_v1_halloween_10b_titan_adapter.pt, &suna_v2_1_pop_masterpiece_adapter.pt— Unified PyTorch checkpoints for Colab & ZeroGPU Spaces.suna_v1_dna_bank.json&config.json— Melodic Soul harmonic profiles, BPM/key matrices, and model configuration.
📜 License & Attribution
- ACE-Step 1.5 XL SFT Backbone & Oobleck VAE: MIT License (Copyright © 2026 ACE-Step Team).
- Qwen3-Embedding-0.6B Text & Phonetic Encoder: Apache 2.0 License (Copyright © Alibaba Cloud).
- JJ coders Suna AI v1.0 LoRA Experts, Melodic Soul Engine, G2P Aligner, & Critic Network: MIT License (Copyright © 2026 JJ Coders Group / Johan Jomet).
- Downloads last month
- 1,431