Add Audio8 TTS engine integration

LICENSE

@@ -44,6 +44,7 @@ The project code is MIT. Model weights carry their own licenses:
   Echo-TTS             CC-BY-NC-SA-4.0                                           No
   Fish Audio S2 Pro    Fish Audio Research License                               No
   Dots TTS             Apache-2.0                                                Yes
+  Audio8 TTS           Apache-2.0                                                Yes
   DramaBox             LTX-2 Community License                                   Conditional
   OmniVoice            Apache-2.0                                                Yes
   MOSS-TTS             Apache-2.0                                                Yes

PROJECT_INDEX.md

@@ -17,7 +17,7 @@
 **Key architectural rules:**
 - Chunking happens in the **processor**, not the adapter (`generate_single()` on adapter = raw single call)
 - Runtime routing happens through `ModelLoadConfig.runtime_mode` + `runtime_profile`, not ad-hoc subprocess calls
-- Shared runtime workers are currently used for fragile engine families such as VibeVoice, Qwen3-TTS / ASR, Granite forced alignment, and Higgs Audio 2. Engines that support the modern stack run natively in the main Transformers 5 environment.
+- Shared runtime workers are currently used for fragile engine families such as VibeVoice, Qwen3-TTS / ASR, Granite forced alignment, Higgs Audio 2, and Audio8 TTS. Engines that support the modern stack run natively in the main Transformers 5 environment.
 - YAML (`docs/Dev reports/tts_audio_suite_engines.yaml`) is source of truth for engine doc tables → run `python3 scripts/generate_engine_tables.py --readme` to regenerate
 - Auxiliary YAML (`docs/Dev reports/tts_audio_suite_aux_models.yaml`) is source of truth for helper/post-process model docs → run `python3 scripts/generate_aux_model_docs.py`
 - All models download to `ComfyUI/models/TTS/<model-name>/`
@@ -25,7 +25,7 @@
 
 ## Engines
 
-19 engines follow the pattern above:
+20 engines follow the pattern above:
 
 | Engine | Adapter | Processor | SRT Processor | Engine Node |
 |--------|---------|-----------|---------------|-------------|
@@ -44,13 +44,14 @@
 | Echo-TTS | `echo_tts_adapter.py` | `nodes/echo_tts/echo_tts_processor.py` | `echo_tts_srt_processor.py` | `echo_tts_engine_node.py` |
 | Fish Audio S2 Pro | `fish_audio_s2_adapter.py` | `nodes/fish_audio_s2/fish_audio_s2_processor.py` | `fish_audio_s2_srt_processor.py` | `fish_audio_s2_engine_node.py` |
 | Dots TTS | `dots_tts_adapter.py` | `nodes/dots_tts/dots_tts_processor.py` | `dots_tts_srt_processor.py` | `dots_tts_engine_node.py` |
+| Audio8 TTS | `audio8_tts_adapter.py` | `nodes/audio8_tts/audio8_tts_processor.py` | `audio8_tts_srt_processor.py` | `audio8_tts_engine_node.py` |
 | DramaBox | `dramabox_adapter.py` | `nodes/dramabox/dramabox_processor.py` | `dramabox_srt_processor.py` | `dramabox_engine_node.py` |
 | OmniVoice | `omnivoice_adapter.py` | `nodes/omnivoice/omnivoice_processor.py` | `omnivoice_srt_processor.py` | `omnivoice_engine_node.py` |
 | MOSS-SoundEffect v2 | `moss_soundeffect_v2_adapter.py` | — | — | `moss_soundeffect_v2_engine_node.py` |
 | RVC | — | `engines/rvc/` | — | `rvc_engine_node.py` |
 
 **Engine implementations live in:**
-- `engines/chatterbox/`, `engines/chatterbox_official_23lang/`, `engines/f5tts/`, `engines/higgs_audio/`, `engines/higgs_audio_v3/`, `engines/vibevoice_engine/`, `engines/step_audio_editx/`, `engines/cosyvoice/`, `engines/qwen3_tts/`, `engines/qwen3_asr/`, `engines/moss_tts/`, `engines/moss_soundeffect_v2/`, `engines/granite_asr/`, `engines/echo_tts/`, `engines/fish_audio_s2/`, `engines/dots_tts/`, `engines/dramabox/`, `engines/omnivoice/`, `engines/rvc/`
+- `engines/chatterbox/`, `engines/chatterbox_official_23lang/`, `engines/f5tts/`, `engines/higgs_audio/`, `engines/higgs_audio_v3/`, `engines/vibevoice_engine/`, `engines/step_audio_editx/`, `engines/cosyvoice/`, `engines/qwen3_tts/`, `engines/qwen3_asr/`, `engines/moss_tts/`, `engines/moss_soundeffect_v2/`, `engines/granite_asr/`, `engines/echo_tts/`, `engines/fish_audio_s2/`, `engines/dots_tts/`, `engines/audio8_tts/`, `engines/dramabox/`, `engines/omnivoice/`, `engines/rvc/`
 
 ## Documentation Files
 
@@ -145,13 +146,14 @@
 - `launcher.py` - runtime bootstrap, venv creation, Windows toolchain env setup
 - `session.py`, `protocol.py` - JSONL worker transport and message protocol
 - `bootstrap.py` - shared runtime bootstrap helpers
-- `vibevoice_proxy.py`, `qwen3_tts_proxy.py`, `qwen3_asr_proxy.py`, `higgs_audio_proxy.py` - parent-process proxies
-- `workers/` - worker subprocess entrypoints for VibeVoice, Qwen3-TTS, Qwen3-ASR/aligner, Higgs Audio
+- `vibevoice_proxy.py`, `qwen3_tts_proxy.py`, `qwen3_asr_proxy.py`, `higgs_audio_proxy.py`, `audio8_tts_proxy.py` - parent-process proxies
+- `workers/` - worker subprocess entrypoints for VibeVoice, Qwen3-TTS, Qwen3-ASR/aligner, Higgs Audio, and Audio8 TTS
 - Current shared legacy T4 runtime profile is reused by:
   - VibeVoice / Kugel
   - Qwen3-TTS
   - Qwen3-ASR and Granite's optional Qwen forced aligner
   - Higgs Audio 2
+  - Audio8 TTS
 
 ### Audio (`utils/audio/`)
 - `processing.py` - Tensor manipulation, normalization, format conversion

README.md

@@ -23,7 +23,7 @@ Subtitle workflows are still a core focus: the suite can transcribe to SRT, rebu
 
 <!-- ENGINE_COMPARISON_START -->
 
-## Quick Engine Comparison — 19 Engines
+## Quick Engine Comparison — 20 Engines
 
 | Engine | Languages | Size | Key Features |
 |--------|-----------|------|--------------|
@@ -41,6 +41,7 @@ Subtitle workflows are still a core focus: the suite can transcribe to SRT, rebu
 | **Echo-TTS** | 🇺🇸 | ~5.3GB + ~1.8GB | Diffusion-based (~30s best), Force Speaker KV (speaker drift control) |
 | **Fish Audio S2 Pro** | 🌐 80+ languages | ~10.3GB / ~8.0GB | Free-form sub-word emotion/prosody tags, Native multi-speaker and multi-turn dialogue with dynamic speaker references |
 | **Dots TTS** | 🇺🇸​🇨🇳​🇩🇪​🇪🇸​🇫🇷​🇮🇹 +13 | ~6GB | Official auto language detect / language control, SOAR and MeanFlow distilled variants |
+| **Audio8 TTS** | 🌐 11 recommended languages | ~2.39 GiB | Reference-free TTS + zero-shot/cross-lingual cloning, 44.1kHz output with sampling or greedy decoding |
 | **DramaBox** | 🇺🇸 | ~16.4GB | Expressive scene prompting and stage directions, Native and SRT-aware duration targeting |
 | **OmniVoice** | 🌐 600+ languages | ~3.7GB | Inline non-verbal tags and pronunciation overrides, Reference-free voice design |
 | **MOSS-TTS** | 🇺🇸​🇨🇳​🇩🇪​🇪🇸​🇫🇷​🇮🇹 +18 | ~8.5GB tokenizer + ~6.1GB/17GB/18GB model | Reference-free voice design with MOSS-VoiceGenerator, Native 1-5 speaker TTSD dialogue |
@@ -1527,6 +1528,7 @@ For offline/manual setup:
 | Granite ASR | `ComfyUI/models/TTS/granite_asr/` | ✅ | Granite ASR models; plus adds native diarization/timestamps, optional Qwen forced aligner reused lazily for timestamps/SRT fallback |
 | Echo-TTS | `ComfyUI/models/TTS/echo-tts-base/` | ✅ | ~7.1GB total (base + dac); CC-BY-NC-SA |
 | Dots TTS | `ComfyUI/models/TTS/dots_tts/` | ✅ | Official base / soar / mf checkpoints with tokenizer, vocoder, speaker encoder |
+| Audio8 TTS | `ComfyUI/models/TTS/audio8_tts/Audio8-TTS-Preview-0.6b/` | ✅ | ~2.39 GiB official 0.6B Preview checkpoint; shared Transformers 4.57.3 runtime required because main 5.10.2 collapses voice cloning |
 | DramaBox | `ComfyUI/models/TTS/dramabox/DramaBox/` | ✅ | ~16.4GB download; fast mode roughly 24GB VRAM; experimental FP8 peaks on RTX 4090: staged ~15.1GB allocated, sequential ~11.7GB allocated / ~12.4GB reserved; conditional LTX-2 Community License |
 | Fish Audio S2 Pro | `ComfyUI/models/TTS/fish_audio_s2_pro/` | ✅ | Official BF16 or optional community FP8 checkpoint; the official checkpoint can be quantized on load with BNB INT8/NF4; main T5 environment with process teardown for Clear VRAM; Fish Audio Research License |
 | OmniVoice | `ComfyUI/models/TTS/omnivoice/` | ✅ | Official OmniVoice model. Voice cloning in this suite requires explicit reference text. |

docs/Dev reports/tts_audio_suite_engines.yaml

@@ -1339,6 +1339,77 @@ engines:
       speed_performance: { supported: "partial", notes: "Moderate; mf variant is faster" }
       reference_free_tts: { supported: true, notes: "(default speaker)" }
 
+  - id: audio8_tts
+    name: Audio8 TTS
+    models: "Audio8 TTS Preview 0.6B"
+    size: "~2.39 GiB"
+    license: "Apache-2.0"
+    commercial: true
+    language_summary_full: "11 recommended languages (including Cantonese)"
+    language_summary_compact: "🌐 11 recommended languages"
+
+    capabilities:
+      tts: true
+      srt: true
+      vc: false
+      asr: false
+      training: false
+
+    runtime_isolation:
+      default_mode: "shared_runtime"
+      main_environment: false
+      shared_runtime: true
+      dedicated_runtime: false
+      runtime_profile: "vibevoice_transformers4_shared"
+      notes: "Required for feature-complete Audio8 inference: main Transformers 5.10.2 collapsed the official voice-clone demo to 0.325s, while the shared Transformers 4.57.3 runtime generated 10.495s for the published 10.54s demo"
+
+    readme_key_features:
+      - "Reference-free TTS + zero-shot/cross-lingual cloning"
+      - "44.1kHz output with sampling or greedy decoding"
+
+    special_features:
+      - "44.1kHz reference-free TTS and zero-shot/cross-lingual voice cloning"
+      - "Exact reference transcript required when voice cloning"
+      - "Greedy or sampled decoding with temperature, top-p, top-k, seed, and token-limit controls"
+      - "Suite text chunking, character switching, and SRT timing/assembly"
+      - "No explicit language selector; multilingual behavior is text-driven"
+      - "No native emotion/style/speed controls, streaming, timestamps, dialogue/multi-speaker mode, or presets"
+      - "Not compatible with Voice Designer: the official model has no instruction-conditioned voice-design mode"
+      - "Unified TTS/SRT integration only; no special node or suite training integration"
+
+    model_sources:
+      - component: "Audio8 TTS Preview 0.6B"
+        source_name: "Audio8/Audio8-TTS-Preview-0.6b"
+        source_url: "https://huggingface.co/Audio8/Audio8-TTS-Preview-0.6b"
+        size: "~2.39 GiB"
+        auto_download: true
+        notes: "Official 0.6B Preview checkpoint with bundled 44.1kHz codec; suite uses the shared Transformers 4.57.3 runtime because voice cloning is not compatible with main Transformers 5.10.2"
+
+    languages:
+      en: { supported: true, flag: "🇺🇸", notes: "" }
+      zh: { supported: true, flag: "🇨🇳", notes: "(Chinese; upstream separately recommends Cantonese, folded into this matrix row)" }
+      de: { supported: true, flag: "🇩🇪", notes: "" }
+      es: { supported: true, flag: "🇪🇸", notes: "" }
+      fr: { supported: true, flag: "🇫🇷", notes: "" }
+      it: { supported: true, flag: "🇮🇹", notes: "" }
+      ja: { supported: true, flag: "🇯🇵", notes: "" }
+      ko: { supported: true, flag: "🇰🇷", notes: "" }
+      pl: { supported: true, flag: "🇵🇱", notes: "" }
+      nl: { supported: true, flag: "🇳🇱", notes: "" }
+
+    features:
+      voice_cloning: { supported: true, notes: "Zero-shot and cross-lingual; reference audio must be paired with its exact transcript" }
+      reference_transcript: { requirement: conditional }
+      native_multi_speaker: { supported: false, notes: "No native dialogue or multi-speaker interface; suite character switching generates speakers as separate segments" }
+      voice_conversion: { supported: false, notes: "No separate voice-conversion mode" }
+      asr_transcribe: { supported: false, notes: "" }
+      emotion_control: { supported: false, notes: "No native emotion, style, or speed controls" }
+      native_long_form: { supported: false, notes: "(uses suite text chunking and SRT assembly)" }
+      community_finetunes: { supported: false, notes: "No community model or upstream SFT/training integration in the suite" }
+      vram_efficient: { supported: "partial", notes: "~1.75 GiB model-plus-codec parameter memory at fp16/bf16 or ~3.5 GiB at float32/CPU, before KV cache, activations, and framework overhead; shared runtime reuses the main environment's PyTorch" }
+      speed_performance: { supported: "partial", notes: "Live shared-runtime validation: 6.36s reference-free audio in 39.3s including load; 9.99s cloned audio in 36.7s warm; exact cache reuse in 56ms" }
+      reference_free_tts: { supported: true, notes: "Reference audio is optional" }
+
   - id: dramabox
     name: DramaBox
     models: "DramaBox 3.3B"
@@ -1903,8 +1974,9 @@ table_notes:
   reference_transcript: >-
     Conditional means the transcript is required only for the specific mode:
     CosyVoice3 zero-shot, Qwen3-TTS full Base cloning, or MOSS-TTSD cloned-speaker
-    dialogue. Higgs Audio 2, Higgs Audio v3, and Dots TTS accept matching text
-    when provided but do not require it.
+    dialogue, and Audio8 TTS zero-shot/cross-lingual cloning. Higgs Audio 2,
+    Higgs Audio v3, and Dots TTS accept matching text when provided but do not
+    require it.
   language_support:
     - "**CosyVoice3 Chinese**: Includes 18+ dialects (Cantonese, Sichuan, Dongbei, Shanghai, etc.)"
     - "**Higgs Audio 2**: Trained on EN, ZH (Mandarin), KO, DE, ES (English majority) - 10M hours AudioVerse dataset"
@@ -1912,6 +1984,7 @@ table_notes:
     - "**IndexTTS-2**: Trained on 55K+ hours - ZH, EN, JA primary."
     - "**MOSS-TTS**: Official list includes ZH, EN, DE, ES, FR, JA, IT, HU, KO, RU, FA, AR, PL, PT, CS, DA, SV, EL, TR; HU/FA/CS are not separate columns in this matrix."
     - "**OmniVoice**: Official model supports 600+ languages; this matrix only shows the suite's comparison subset."
+    - "**Audio8 TTS**: The Preview recommends Cantonese, Chinese, Dutch, English, French, German, Italian, Japanese, Korean, Polish, and Spanish. It has no explicit language selector; language behavior is text-driven. Cantonese is folded into the Chinese row in this matrix."
     - "**RVC**: Language-agnostic voice conversion/post-processing; it is marked supported for every language row."
     - "Qwen3-TTS: Supports pt-BR only with instruction (voice design or harcoded voices). Base can't do instructions, so it will always ouput pt-PT."
 
@@ -1980,6 +2053,10 @@ readme_model_download_table:
     primary_model_path: "ComfyUI/models/TTS/dots_tts/"
     auto_download: "✅"
     notes: "Official base / soar / mf checkpoints with tokenizer, vocoder, speaker encoder"
+  - engine: "Audio8 TTS"
+    primary_model_path: "ComfyUI/models/TTS/audio8_tts/Audio8-TTS-Preview-0.6b/"
+    auto_download: "✅"
+    notes: "~2.39 GiB official 0.6B Preview checkpoint; shared Transformers 4.57.3 runtime required because main 5.10.2 collapses voice cloning"
   - engine: "DramaBox"
     primary_model_path: "ComfyUI/models/TTS/dramabox/DramaBox/"
     auto_download: "✅"
@@ -2423,6 +2500,22 @@ model_layouts_markdown: |
   - Native sample rate is 48kHz.
   - Main-environment support works on Transformers 5; on Windows, `normalize_text` falls back to no-op if `WeTextProcessing` is unavailable.
 
+  ## Audio8 TTS
+
+  ```text
+  ComfyUI/models/TTS/audio8_tts/
+  └── Audio8-TTS-Preview-0.6b/
+      └── (13 required official runtime files, including codec weights)
+  ```
+
+  Notes:
+
+  - The official `Audio8/Audio8-TTS-Preview-0.6b` checkpoint is about 2.39 GiB and includes the 44.1kHz neural codec.
+  - The 0.6B model plus codec use about 1.75 GiB for parameters at fp16/bf16 or about 3.5 GiB at float32/CPU, before KV cache, activations, and framework overhead.
+  - The suite uses the existing `vibevoice_transformers4_shared` runtime with Transformers 4.57.3. Main Transformers 5.10.2 is not compatible with Audio8 voice cloning: the official demo collapsed to 0.325s, while the shared runtime generated 10.495s for the published 10.54s demo.
+  - Live shared-runtime validation produced 6.36s of reference-free audio in 39.3s including model load, 9.99s of cloned audio in 36.7s warm, and exact cache reuse in 56ms.
+  - The engine has unified TTS and SRT support only: no Voice Designer support, special node, training integration, native streaming/timestamps/dialogue, or built-in presets.
+
   ## OmniVoice
 
   ```text

docs/ENGINE_COMPARISON.md

@@ -18,6 +18,7 @@
 | **Echo-TTS**       | Main      | echo-tts-base + fish-s1-dac-min           | ~5.3GB + ~1.8GB | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | CC-BY-NC-SA-4.0          | Diffusion-based (~30s best), Force Speaker KV (speaker drift control)                    | 1                                                                                        |
 | **Fish Audio S2 Pro** | Main      | S2 Pro 4B / FP8                           | ~10.3GB / ~8.0GB | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | Fish Audio Research License | Free-form sub-word emotion/prosody tags, Native multi-speaker and multi-turn dialogue with dynamic speaker references, Zero-shot voice cloning, Optional per-segment custom character switching, Configurable 4K-32K native context with reduced KV-cache VRAM, Optional community FP8 weight-only checkpoint with BF16 activations, Optional on-the-fly BitsAndBytes INT8/NF4 for the official checkpoint | 80+ languages                                                                            |
 | **Dots TTS**       | Main      | dots.tts-base, dots.tts-soar, dots.tts-mf | ~6GB         | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | Apache-2.0               | Official auto language detect / language control, SOAR and MeanFlow distilled variants   | 19                                                                                       |
+| **Audio8 TTS**     | Shared    | Audio8 TTS Preview 0.6B                   | ~2.39 GiB    | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | Apache-2.0               | 44.1kHz reference-free TTS and zero-shot/cross-lingual voice cloning, Exact reference transcript required when voice cloning, Greedy or sampled decoding with temperature, top-p, top-k, seed, and token-limit controls, Suite text chunking, character switching, and SRT timing/assembly, No explicit language selector; multilingual behavior is text-driven, No native emotion/style/speed controls, streaming, timestamps, dialogue/multi-speaker mode, or presets, Not compatible with Voice Designer: the official model has no instruction-conditioned voice-design mode, Unified TTS/SRT integration only; no special node or suite training integration | 11 recommended languages (including Cantonese)                                           |
 | **DramaBox**       | Main      | DramaBox 3.3B                             | ~16.4GB      | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | LTX-2 Community License  | Expressive scene prompting and stage directions, Native and SRT-aware duration targeting, Official duration-aware long-form chunking with scene-prefix preservation, Optional 10-second zero-shot voice reference, CFG negative prompt with per-segment switching, Explicit generation/reference durations, CFG rescale control, and optional Perth watermark, Experimental staged and sequential strategies for lowering peak VRAM; official LTX FP8-cast storage, Optional official torch.compile path | 1                                                                                        |
 | **OmniVoice**      | Main      | OmniVoice                                 | ~3.7GB       | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | Apache-2.0               | Inline non-verbal tags and pronunciation overrides, Reference-free voice design, 600+ language support, Upstream long-form chunk orchestration | 600+                                                                                     |
 | **MOSS-TTS**       | Main      | Local 1.7B, Delay 8B v1.5/1.0, VoiceGenerator 1.7B, SoundEffect 8B v1, TTSD 8B | ~8.5GB tokenizer + ~6.1GB/17GB/18GB model | ✅ | ✅ | ❌ | ❌ | ✅ | ✅ | Apache-2.0               | Reference-free voice design with MOSS-VoiceGenerator, Native 1-5 speaker TTSD dialogue, 31-language generation with MOSS-TTS-v1.5, Prompt-only sound-effect generation with MOSS-SoundEffect v1, Long-form generation (TTSD/Delay), Duration token hint, Local/Delay/TTSD variants, Initial integrated LoRA training workflow (Delay 8B) | 24                                                                                       |

docs/FEATURE_COMPARISON.md

@@ -2,22 +2,22 @@
 
 ## Feature Comparison Matrix
 
-| Feature                      | F5-TTS | ChatterBox | ChatterBox 23L | VibeVoice | Higgs Audio 2 | Higgs Audio v3 | IndexTTS-2 | CosyVoice3 | Qwen3-TTS | Granite ASR | Step Audio EditX | Echo-TTS | Fish Audio S2 Pro | Dots TTS | DramaBox | OmniVoice | MOSS-TTS | MOSS-SoundEffect v2 | RVC |
-|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
-| **TTS**                      | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
-| **SRT**                      | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
-| **Voice Conversion**         | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| **ASR (Transcribe)**         | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
-| **Sound Effects**            | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ |
-| **Training**                 | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ |
-| **Voice Cloning**            | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ (Base model) | ❌ | ✅ | ✅ | ✅ Reference audio plus exact transcript | ✅ | ✅ Optional reference audio; configurable 3-30 second window from the beginning of the source, default 10 seconds | ✅ | ✅ | ❌ | ⚠️ (needs training) |
-| **Reference Transcript†**    | **Required** | Not used | Not used | Not used | Optional | Optional | Not used | Conditional | Conditional | N/A | **Required** | Not used | **Required** | Optional | Not used | **Required** | Conditional | Not used | N/A |
-| **Native Multi-Speaker**     | ❌ | ❌ | ❌ | ✅ (Base only, Kugel uses fallback) | ✅ | ❌ | ❌ | ❌ | ❌ | ⚠️ (Plus variant speaker attribution / diarization) | ❌ | ❌ | ✅ Native dialogue is default; optional custom mode generates each [Character] segment independently as local speaker 0 | ❌ | ❌ | ❌ | ✅ (TTSD v1.0; 1-5 speakers) | ❌ | ❌ |
-| **Emotion Control**          | ❌ | ❌ | ⚠️ (v2 tags - doesn't work) | ❌ | ⚠️ (via prompt) | ✅ (native inline tags) | ✅ (8 emotions) | ⚠️ (via instruct) | ⚠️ (via instruct) | ❌ | ✅ (14 emotions) | ❌ | ✅ Free-form inline natural-language tags | ❌ | ✅ Natural-language scene prompt and stage directions | ⚠️ (voice-design instruct + inline non-verbal tags) | ✅ (MOSS-VoiceGenerator instruction-conditioned voice design) | ❌ | ❌ |
-| **Native Long-form**         | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ Configurable 4K-32K native context; suite text chunking is bypassed | ❌ | ✅ Official duration-aware quote-group chunking; ~37s target / 45s cap | ✅ (uses upstream audio_chunk_duration / audio_chunk_threshold orchestration; bypasses suite char-based chunk splitting) | ✅ (TTSD/Delay long-form; use chunk orchestration for very long inputs) | ❌ | N/A |
-| **Community Finetunes**      | ✅ | ✅ | ✅ | ✅ KugelAudio, Hindi | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (LoRA adapter inference supported; initial integrated LoRA training support added for MOSS-TTS Delay 8B) | ❌ | ✅ |
-| **VRAM Efficient**           | ✅ | ✅ | ✅ | ⚠️ (5-18GB) | ⚠️ (9GB) | ⚠️ (~8-10GB) | ⚠️ (9-12GB) | ✅ (5.4GB) | ✅ (3-6GB) | ✅ (~4.6GB) | ⚠️ (7GB) | ⚠️ (~7GB total) | ⚠️ 8K context measured at ~15.2GB BF16, ~11.2GB FP8, ~11.2GB BNB INT8, or ~8.9GB BNB NF4; BF16 codec and activations; BNB is a load-time option for the official checkpoint | ⚠️ (main env works; 2B-class model is not lightweight) | ✅ Fast mode is ~24GB; experimental FP8 peaks on RTX 4090: staged ~15.1GB allocated, sequential ~11.7GB allocated / ~12.4GB reserved | ⚠️ (main env works; 2B-class model is not lightweight) | ⚠️ (Local 1.7B smaller; tokenizer is large) | ⚠️ (runs in the main ComfyUI environment but remains GPU-heavy) | ✅ |
-| **Speed/Performance**        | ✅ Very Fast | ✅ Fast | ✅ Fast | ⚠️ | ⚠️ | ⚠️ CUDA recommended | ⚠️ | ✅ Fast | ⚠️ | ⚠️ Moderate | ⚠️ | ✅ Fast (diffusion, realtime-capable) | ✅ Main-environment subprocess with reliable teardown; local compile measurements: ~40 it/s BF16, ~11.8 it/s NF4 at ~8.9GB VRAM, and ~3.7 it/s INT8 at ~11.2GB VRAM; quality comparison pending | ⚠️ Moderate; mf variant is faster | ⚠️ Fast favors repeated generation; staged reloads temporary components; sequential also transfers the transformer; optional block-level torch.compile accelerates repeat denoising | ✅ Fast; upstream reports sub-realtime RTF | ✅ Fast with CUDA/FlashAttention | ⚠️ (100 diffusion steps by default) | ✅ Fast |
-| **No Narrator Required**     | ❌ | ✅ (default speaker) | ✅ (default speaker) | ✅ (zero-shot / default speaker) | ✅ (basic TTS if no narrator/reference is provided) | ✅ (zero-shot) | ❌ | ✅ (cross-lingual or instruct mode) | ✅ (Base default voice or CustomVoice presets) | N/A | ❌ | ❌ | ✅ Reference audio is optional | ✅ (default speaker) | ✅ Voice reference is optional | ✅ (native default voice; instruct also works without narrator) | ✅ (direct TTS and prompt-only generation) | ❌ | N/A |
+| Feature                      | F5-TTS | ChatterBox | ChatterBox 23L | VibeVoice | Higgs Audio 2 | Higgs Audio v3 | IndexTTS-2 | CosyVoice3 | Qwen3-TTS | Granite ASR | Step Audio EditX | Echo-TTS | Fish Audio S2 Pro | Dots TTS | Audio8 TTS | DramaBox | OmniVoice | MOSS-TTS | MOSS-SoundEffect v2 | RVC |
+|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
+| **TTS**                      | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
+| **SRT**                      | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
+| **Voice Conversion**         | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| **ASR (Transcribe)**         | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
+| **Sound Effects**            | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ |
+| **Training**                 | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ |
+| **Voice Cloning**            | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ (Base model) | ❌ | ✅ | ✅ | ✅ Reference audio plus exact transcript | ✅ | ✅ Zero-shot and cross-lingual; reference audio must be paired with its exact transcript | ✅ Optional reference audio; configurable 3-30 second window from the beginning of the source, default 10 seconds | ✅ | ✅ | ❌ | ⚠️ (needs training) |
+| **Reference Transcript†**    | **Required** | Not used | Not used | Not used | Optional | Optional | Not used | Conditional | Conditional | N/A | **Required** | Not used | **Required** | Optional | Conditional | Not used | **Required** | Conditional | Not used | N/A |
+| **Native Multi-Speaker**     | ❌ | ❌ | ❌ | ✅ (Base only, Kugel uses fallback) | ✅ | ❌ | ❌ | ❌ | ❌ | ⚠️ (Plus variant speaker attribution / diarization) | ❌ | ❌ | ✅ Native dialogue is default; optional custom mode generates each [Character] segment independently as local speaker 0 | ❌ | ❌ | ❌ | ❌ | ✅ (TTSD v1.0; 1-5 speakers) | ❌ | ❌ |
+| **Emotion Control**          | ❌ | ❌ | ⚠️ (v2 tags - doesn't work) | ❌ | ⚠️ (via prompt) | ✅ (native inline tags) | ✅ (8 emotions) | ⚠️ (via instruct) | ⚠️ (via instruct) | ❌ | ✅ (14 emotions) | ❌ | ✅ Free-form inline natural-language tags | ❌ | ❌ | ✅ Natural-language scene prompt and stage directions | ⚠️ (voice-design instruct + inline non-verbal tags) | ✅ (MOSS-VoiceGenerator instruction-conditioned voice design) | ❌ | ❌ |
+| **Native Long-form**         | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ Configurable 4K-32K native context; suite text chunking is bypassed | ❌ | ❌ | ✅ Official duration-aware quote-group chunking; ~37s target / 45s cap | ✅ (uses upstream audio_chunk_duration / audio_chunk_threshold orchestration; bypasses suite char-based chunk splitting) | ✅ (TTSD/Delay long-form; use chunk orchestration for very long inputs) | ❌ | N/A |
+| **Community Finetunes**      | ✅ | ✅ | ✅ | ✅ KugelAudio, Hindi | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (LoRA adapter inference supported; initial integrated LoRA training support added for MOSS-TTS Delay 8B) | ❌ | ✅ |
+| **VRAM Efficient**           | ✅ | ✅ | ✅ | ⚠️ (5-18GB) | ⚠️ (9GB) | ⚠️ (~8-10GB) | ⚠️ (9-12GB) | ✅ (5.4GB) | ✅ (3-6GB) | ✅ (~4.6GB) | ⚠️ (7GB) | ⚠️ (~7GB total) | ⚠️ 8K context measured at ~15.2GB BF16, ~11.2GB FP8, ~11.2GB BNB INT8, or ~8.9GB BNB NF4; BF16 codec and activations; BNB is a load-time option for the official checkpoint | ⚠️ (main env works; 2B-class model is not lightweight) | ⚠️ ~1.75 GiB model-plus-codec parameter memory at fp16/bf16 or ~3.5 GiB at float32/CPU, before KV cache, activations, and framework overhead; shared runtime reuses the main environment's PyTorch | ✅ Fast mode is ~24GB; experimental FP8 peaks on RTX 4090: staged ~15.1GB allocated, sequential ~11.7GB allocated / ~12.4GB reserved | ⚠️ (main env works; 2B-class model is not lightweight) | ⚠️ (Local 1.7B smaller; tokenizer is large) | ⚠️ (runs in the main ComfyUI environment but remains GPU-heavy) | ✅ |
+| **Speed/Performance**        | ✅ Very Fast | ✅ Fast | ✅ Fast | ⚠️ | ⚠️ | ⚠️ CUDA recommended | ⚠️ | ✅ Fast | ⚠️ | ⚠️ Moderate | ⚠️ | ✅ Fast (diffusion, realtime-capable) | ✅ Main-environment subprocess with reliable teardown; local compile measurements: ~40 it/s BF16, ~11.8 it/s NF4 at ~8.9GB VRAM, and ~3.7 it/s INT8 at ~11.2GB VRAM; quality comparison pending | ⚠️ Moderate; mf variant is faster | ⚠️ Live shared-runtime validation: 6.36s reference-free audio in 39.3s including load; 9.99s cloned audio in 36.7s warm; exact cache reuse in 56ms | ⚠️ Fast favors repeated generation; staged reloads temporary components; sequential also transfers the transformer; optional block-level torch.compile accelerates repeat denoising | ✅ Fast; upstream reports sub-realtime RTF | ✅ Fast with CUDA/FlashAttention | ⚠️ (100 diffusion steps by default) | ✅ Fast |
+| **No Narrator Required**     | ❌ | ✅ (default speaker) | ✅ (default speaker) | ✅ (zero-shot / default speaker) | ✅ (basic TTS if no narrator/reference is provided) | ✅ (zero-shot) | ❌ | ✅ (cross-lingual or instruct mode) | ✅ (Base default voice or CustomVoice presets) | N/A | ❌ | ❌ | ✅ Reference audio is optional | ✅ (default speaker) | ✅ Reference audio is optional | ✅ Voice reference is optional | ✅ (native default voice; instruct also works without narrator) | ✅ (direct TTS and prompt-only generation) | ❌ | N/A |
 
-† **Reference Transcript:** Conditional means the transcript is required only for the specific mode: CosyVoice3 zero-shot, Qwen3-TTS full Base cloning, or MOSS-TTSD cloned-speaker dialogue. Higgs Audio 2, Higgs Audio v3, and Dots TTS accept matching text when provided but do not require it.
+† **Reference Transcript:** Conditional means the transcript is required only for the specific mode: CosyVoice3 zero-shot, Qwen3-TTS full Base cloning, or MOSS-TTSD cloned-speaker dialogue, and Audio8 TTS zero-shot/cross-lingual cloning. Higgs Audio 2, Higgs Audio v3, and Dots TTS accept matching text when provided but do not require it.

docs/LANGUAGE_SUPPORT.md

@@ -2,110 +2,110 @@
 
 ## Language Support by Engine
 
-| Language       | Code | F5-TTS | ChatterBox | ChatterBox 23L | VibeVoice | Higgs Audio 2 | Higgs Audio v3 | IndexTTS-2 | CosyVoice3 | Qwen3-TTS | Granite ASR | Step Audio EditX | Echo-TTS | Fish Audio S2 Pro | Dots TTS | DramaBox | OmniVoice | MOSS-TTS | MOSS-SoundEffect v2 | RVC |
-|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
-| 🇺🇸 **English** | EN   | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ Tier 1 | ✅ | ✅ Official model is English-only | ✅ | ✅ | ✅ Officially demonstrated prompt language | ✅ |
-| 🇨🇳 **Chinese** | ZH   | ❌ | ❌ | ✅ | ✅ | ✅ (Mandarin) | ✅ | ✅ | ✅ + 18 dialects | ✅ | ❌ | ✅ (Mandarin + Sichuanese, Cantonese) | ❌ | ✅ Tier 1 | ✅ (Mandarin; official also exposes YUE separately outside this matrix) | ❌ | ✅ | ✅ | ✅ Officially demonstrated prompt language | ✅ |
-| 🇩🇪 **German**  | DE   | ✅ | ✅ (×3) | ✅ | ✅ (Kugel) | ✅ | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ Tier 2 | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
-| 🇪🇸 **Spanish** | ES   | ✅ | ❌ | ✅ | ✅ (Kugel) | ✅ | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ Tier 2 | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
-| 🇫🇷 **French**  | FR   | ✅ | ✅ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ Tier 2 | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
-| 🇮🇹 **Italian** | IT   | ✅ | ✅ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
-| 🇯🇵 **Japanese** | JA   | ✅ | ✅ | ✅ | ✅ (Kugel) | ❌ | ✅ | ✅ ? | ✅ | ✅ | ✅ (4.1-2b supports Japanese; plus variant does not) | ✅ | ❌ | ✅ Tier 1 | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
-| 🇰🇷 **Korean**  | KO   | ❌ | ✅ | ✅ | ✅ (Kugel) | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ | ❌ | ✅ Tier 2 | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
-| 🇷🇺 **Russian** | RU   | ❌ | ✅ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ Tier 2 | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
-| 🇧🇷 **Portuguese** | PT   | ✅ (BR) | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ✅ (EU/BR*) | ✅ | ❌ | ❌ | ✅ Tier 2 | ✅ (Official PT tag is generic; upstream does not expose separate PT-BR/PT-PT tags and it may lean more European Portuguese than Brazilian Portuguese) | ❌ | ✅ (generic PT; official language space is much broader than this matrix) | ✅ | ❌ | ✅ |
-| 🇵🇱 **Polish**  | PL   | ✅ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
-| 🇮🇳 **Hindi**   | HI   | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ (v1.5) | ❌ | ✅ |
-| 🇪🇬 **Arabic**  | AR   | ❌ | ❌ | ✅ (Egyptian) | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ Tier 2 | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
-| 🇹🇷 **Turkish** | TR   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
-| 🇹🇭 **Thai**    | TH   | ✅ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ (v1.5) | ❌ | ✅ |
-| 🇳🇴 **Norwegian** | NO   | ❌ | ✅ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ |
-| 🇻🇳 **Vietnamese** | VI   | ❌ | ❌ | ✅ (Viterbox) | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ (v1.5) | ❌ | ✅ |
-| 🇦🇲 **Armenian** | HY   | ❌ | ✅ | ❌ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ |
-| 🇬🇪 **Georgian** | KA   | ❌ | ✅ | ❌ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ |
-| 🇩🇰 **Danish**  | DA   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ✅ |
-| 🇫🇮 **Finnish** | FI   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ (v1.5) | ❌ | ✅ |
-| 🇬🇷 **Greek**   | EL   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
-| 🇮🇱 **Hebrew**  | HE   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ (v1.5) | ❌ | ✅ |
-| 🇲🇾 **Malay**   | MS   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ (v1.5) | ❌ | ✅ |
-| 🇳🇱 **Dutch**   | NL   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ (v1.5) | ❌ | ✅ |
-| 🇸🇪 **Swedish** | SV   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ✅ |
-| 🇰🇪 **Swahili** | SW   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ (v1.5) | ❌ | ✅ |
-| 🇿🇦 **Afrikaans** | AF   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇦🇱 **Albanian** | SQ   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇮🇳 **Assamese** | AS   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇪🇸 **Asturian** | AST  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇦🇿 **Azerbaijani** | AZ   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇷🇺 **Bashkir** | BA   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇪🇸 **Basque**  | EU   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇧🇾 **Belarusian** | BE   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇧🇩 **Bengali** | BN   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇧🇦 **Bosnian** | BS   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇧🇬 **Bulgarian** | BG   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇪🇸 **Catalan** | CA   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇵🇭 **Cebuano** | CEB  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇮🇶 **Central Kurdish** | CKB  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇭🇷 **Croatian** | HR   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇨🇿 **Czech**   | CS   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇷🇺 **Eastern Mari** | MHR  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🌍 **Esperanto** | EO   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇪🇪 **Estonian** | ET   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇪🇸 **Galician** | GL   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇮🇳 **Gujarati** | GU   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇭🇹 **Haitian Creole** | HT   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇳🇬 **Hausa**   | HA   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇭🇺 **Hungarian** | HU   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇮🇩 **Indonesian** | ID   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇮🇩 **Javanese** | JV   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇮🇳 **Kannada** | KN   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇰🇿 **Kazakh**  | KK   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇷🇼 **Kinyarwanda** | RW   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇰🇬 **Kyrgyz**  | KY   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇱🇻 **Latvian** | LV   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇨🇩 **Lingala** | LN   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇱🇹 **Lithuanian** | LT   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇰🇪 **Luo**     | LUO  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇲🇰 **Macedonian** | MK   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇮🇳 **Malayalam** | ML   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇲🇹 **Maltese** | MT   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇳🇿 **Māori**   | MI   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇮🇳 **Marathi** | MR   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇲🇳 **Mongolian** | MN   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇳🇵 **Nepali**  | NE   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇫🇷 **Occitan** | OC   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇮🇷 **Persian** | FA   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇷🇴 **Romanian** | RO   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇿🇦 **Sepedi**  | NSO  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇷🇸 **Serbian** | SR   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇿🇼 **Shona**   | SN   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇸🇰 **Slovak**  | SK   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇸🇮 **Slovene** | SL   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇵🇭 **Tagalog** | TL   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇹🇯 **Tajik**   | TG   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇮🇳 **Tamil**   | TA   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇮🇳 **Telugu**  | TE   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇺🇦 **Ukrainian** | UK   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇵🇰 **Urdu**    | UR   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇨🇳 **Uyghur**  | UG   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇺🇿 **Uzbek**   | UZ   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇿🇦 **Xhosa**   | XH   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇿🇦 **Zulu**    | ZU   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇲🇼 **Chichewa/Nyanja** | NY   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇮🇳 **Eastern Punjabi** | PA   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇺🇬 **Ganda**   | LG   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇮🇸 **Icelandic** | IS   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇮🇪 **Irish**   | GA   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇩🇿 **Kabyle**  | KAB  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇨🇻 **Kabuverdianu** | KEA  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇰🇪 **Kamba**   | KAM  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇻🇦 **Latin**   | LA   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇱🇺 **Luxembourgish** | LB   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇪🇹 **Oromo**   | OM   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇦🇫 **Pashto**  | PS   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇵🇰 **Sindhi**  | SD   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇸🇴 **Somali**  | SO   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🇦🇴 **Umbundu** | UMB  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
-| 🏴 **Welsh**    | CY   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| Language       | Code | F5-TTS | ChatterBox | ChatterBox 23L | VibeVoice | Higgs Audio 2 | Higgs Audio v3 | IndexTTS-2 | CosyVoice3 | Qwen3-TTS | Granite ASR | Step Audio EditX | Echo-TTS | Fish Audio S2 Pro | Dots TTS | Audio8 TTS | DramaBox | OmniVoice | MOSS-TTS | MOSS-SoundEffect v2 | RVC |
+|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
+| 🇺🇸 **English** | EN   | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ Tier 1 | ✅ | ✅ | ✅ Official model is English-only | ✅ | ✅ | ✅ Officially demonstrated prompt language | ✅ |
+| 🇨🇳 **Chinese** | ZH   | ❌ | ❌ | ✅ | ✅ | ✅ (Mandarin) | ✅ | ✅ | ✅ + 18 dialects | ✅ | ❌ | ✅ (Mandarin + Sichuanese, Cantonese) | ❌ | ✅ Tier 1 | ✅ (Mandarin; official also exposes YUE separately outside this matrix) | ✅ (Chinese; upstream separately recommends Cantonese, folded into this matrix row) | ❌ | ✅ | ✅ | ✅ Officially demonstrated prompt language | ✅ |
+| 🇩🇪 **German**  | DE   | ✅ | ✅ (×3) | ✅ | ✅ (Kugel) | ✅ | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ Tier 2 | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
+| 🇪🇸 **Spanish** | ES   | ✅ | ❌ | ✅ | ✅ (Kugel) | ✅ | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ Tier 2 | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
+| 🇫🇷 **French**  | FR   | ✅ | ✅ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ Tier 2 | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
+| 🇮🇹 **Italian** | IT   | ✅ | ✅ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
+| 🇯🇵 **Japanese** | JA   | ✅ | ✅ | ✅ | ✅ (Kugel) | ❌ | ✅ | ✅ ? | ✅ | ✅ | ✅ (4.1-2b supports Japanese; plus variant does not) | ✅ | ❌ | ✅ Tier 1 | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
+| 🇰🇷 **Korean**  | KO   | ❌ | ✅ | ✅ | ✅ (Kugel) | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ | ❌ | ✅ Tier 2 | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
+| 🇷🇺 **Russian** | RU   | ❌ | ✅ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ Tier 2 | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ | ✅ |
+| 🇧🇷 **Portuguese** | PT   | ✅ (BR) | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ✅ (EU/BR*) | ✅ | ❌ | ❌ | ✅ Tier 2 | ✅ (Official PT tag is generic; upstream does not expose separate PT-BR/PT-PT tags and it may lean more European Portuguese than Brazilian Portuguese) | ❌ | ❌ | ✅ (generic PT; official language space is much broader than this matrix) | ✅ | ❌ | ✅ |
+| 🇵🇱 **Polish**  | PL   | ✅ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ |
+| 🇮🇳 **Hindi**   | HI   | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ | ✅ (v1.5) | ❌ | ✅ |
+| 🇪🇬 **Arabic**  | AR   | ❌ | ❌ | ✅ (Egyptian) | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ Tier 2 | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ | ✅ |
+| 🇹🇷 **Turkish** | TR   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ | ✅ |
+| 🇹🇭 **Thai**    | TH   | ✅ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ | ✅ (v1.5) | ❌ | ✅ |
+| 🇳🇴 **Norwegian** | NO   | ❌ | ✅ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ |
+| 🇻🇳 **Vietnamese** | VI   | ❌ | ❌ | ✅ (Viterbox) | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ | ✅ (v1.5) | ❌ | ✅ |
+| 🇦🇲 **Armenian** | HY   | ❌ | ✅ | ❌ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ |
+| 🇬🇪 **Georgian** | KA   | ❌ | ✅ | ❌ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ |
+| 🇩🇰 **Danish**  | DA   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ✅ |
+| 🇫🇮 **Finnish** | FI   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ | ✅ (v1.5) | ❌ | ✅ |
+| 🇬🇷 **Greek**   | EL   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ | ✅ |
+| 🇮🇱 **Hebrew**  | HE   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ (v1.5) | ❌ | ✅ |
+| 🇲🇾 **Malay**   | MS   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ (v1.5) | ❌ | ✅ |
+| 🇳🇱 **Dutch**   | NL   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ✅ | ✅ (v1.5) | ❌ | ✅ |
+| 🇸🇪 **Swedish** | SV   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ✅ |
+| 🇰🇪 **Swahili** | SW   | ❌ | ❌ | ✅ | ✅ (Kugel) | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ (v1.5) | ❌ | ✅ |
+| 🇿🇦 **Afrikaans** | AF   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇦🇱 **Albanian** | SQ   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇮🇳 **Assamese** | AS   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇪🇸 **Asturian** | AST  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇦🇿 **Azerbaijani** | AZ   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇷🇺 **Bashkir** | BA   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇪🇸 **Basque**  | EU   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇧🇾 **Belarusian** | BE   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇧🇩 **Bengali** | BN   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇧🇦 **Bosnian** | BS   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇧🇬 **Bulgarian** | BG   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇪🇸 **Catalan** | CA   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇵🇭 **Cebuano** | CEB  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇮🇶 **Central Kurdish** | CKB  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇭🇷 **Croatian** | HR   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇨🇿 **Czech**   | CS   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇷🇺 **Eastern Mari** | MHR  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🌍 **Esperanto** | EO   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇪🇪 **Estonian** | ET   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇪🇸 **Galician** | GL   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇮🇳 **Gujarati** | GU   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇭🇹 **Haitian Creole** | HT   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇳🇬 **Hausa**   | HA   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇭🇺 **Hungarian** | HU   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇮🇩 **Indonesian** | ID   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇮🇩 **Javanese** | JV   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇮🇳 **Kannada** | KN   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇰🇿 **Kazakh**  | KK   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇷🇼 **Kinyarwanda** | RW   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇰🇬 **Kyrgyz**  | KY   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇱🇻 **Latvian** | LV   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇨🇩 **Lingala** | LN   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇱🇹 **Lithuanian** | LT   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇰🇪 **Luo**     | LUO  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇲🇰 **Macedonian** | MK   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇮🇳 **Malayalam** | ML   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇲🇹 **Maltese** | MT   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇳🇿 **Māori**   | MI   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇮🇳 **Marathi** | MR   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇲🇳 **Mongolian** | MN   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇳🇵 **Nepali**  | NE   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇫🇷 **Occitan** | OC   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇮🇷 **Persian** | FA   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇷🇴 **Romanian** | RO   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇿🇦 **Sepedi**  | NSO  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇷🇸 **Serbian** | SR   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇿🇼 **Shona**   | SN   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇸🇰 **Slovak**  | SK   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇸🇮 **Slovene** | SL   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇵🇭 **Tagalog** | TL   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇹🇯 **Tajik**   | TG   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇮🇳 **Tamil**   | TA   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇮🇳 **Telugu**  | TE   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇺🇦 **Ukrainian** | UK   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇵🇰 **Urdu**    | UR   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇨🇳 **Uyghur**  | UG   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇺🇿 **Uzbek**   | UZ   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇿🇦 **Xhosa**   | XH   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇿🇦 **Zulu**    | ZU   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (polished) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇲🇼 **Chichewa/Nyanja** | NY   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇮🇳 **Eastern Punjabi** | PA   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇺🇬 **Ganda**   | LG   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇮🇸 **Icelandic** | IS   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇮🇪 **Irish**   | GA   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇩🇿 **Kabyle**  | KAB  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇨🇻 **Kabuverdianu** | KEA  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇰🇪 **Kamba**   | KAM  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇻🇦 **Latin**   | LA   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇱🇺 **Luxembourgish** | LB   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇪🇹 **Oromo**   | OM   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇦🇫 **Pashto**  | PS   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇵🇰 **Sindhi**  | SD   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇸🇴 **Somali**  | SO   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🇦🇴 **Umbundu** | UMB  | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
+| 🏴 **Welsh**    | CY   | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ (usable) | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
 
 **Notes:**
 
@@ -115,5 +115,6 @@
 - **IndexTTS-2**: Trained on 55K+ hours - ZH, EN, JA primary.
 - **MOSS-TTS**: Official list includes ZH, EN, DE, ES, FR, JA, IT, HU, KO, RU, FA, AR, PL, PT, CS, DA, SV, EL, TR; HU/FA/CS are not separate columns in this matrix.
 - **OmniVoice**: Official model supports 600+ languages; this matrix only shows the suite's comparison subset.
+- **Audio8 TTS**: The Preview recommends Cantonese, Chinese, Dutch, English, French, German, Italian, Japanese, Korean, Polish, and Spanish. It has no explicit language selector; language behavior is text-driven. Cantonese is folded into the Chinese row in this matrix.
 - **RVC**: Language-agnostic voice conversion/post-processing; it is marked supported for every language row.
 - Qwen3-TTS: Supports pt-BR only with instruction (voice design or harcoded voices). Base can't do instructions, so it will always ouput pt-PT.

docs/MODEL_DOWNLOAD_SOURCES.md

@@ -129,6 +129,12 @@ Use this as the canonical list of model repositories/links for offline setup.
 | dots.tts-soar | [rednote-hilab/dots.tts-soar](https://huggingface.co/rednote-hilab/dots.tts-soar) | ~6GB | ✅ | Official SOAR checkpoint for higher-quality zero-shot cloning |
 | dots.tts-mf | [rednote-hilab/dots.tts-mf](https://huggingface.co/rednote-hilab/dots.tts-mf) | ~6GB | ✅ | Official MeanFlow-distilled checkpoint for faster inference |
 
+## Audio8 TTS
+
+| Component | Source | Size | Auto-Download | Notes |
+|---|---|---|---|---|
+| Audio8 TTS Preview 0.6B | [Audio8/Audio8-TTS-Preview-0.6b](https://huggingface.co/Audio8/Audio8-TTS-Preview-0.6b) | ~2.39 GiB | ✅ | Official 0.6B Preview checkpoint with bundled 44.1kHz codec; suite uses the shared Transformers 4.57.3 runtime because voice cloning is not compatible with main Transformers 5.10.2 |
+
 ## DramaBox
 
 | Component | Source | Size | Auto-Download | Notes |

docs/MODEL_LAYOUTS.md

@@ -427,6 +427,22 @@ Notes:
 - Native sample rate is 48kHz.
 - Main-environment support works on Transformers 5; on Windows, `normalize_text` falls back to no-op if `WeTextProcessing` is unavailable.
 
+## Audio8 TTS
+
+```text
+ComfyUI/models/TTS/audio8_tts/
+└── Audio8-TTS-Preview-0.6b/
+    └── (13 required official runtime files, including codec weights)
+```
+
+Notes:
+
+- The official `Audio8/Audio8-TTS-Preview-0.6b` checkpoint is about 2.39 GiB and includes the 44.1kHz neural codec.
+- The 0.6B model plus codec use about 1.75 GiB for parameters at fp16/bf16 or about 3.5 GiB at float32/CPU, before KV cache, activations, and framework overhead.
+- The suite uses the existing `vibevoice_transformers4_shared` runtime with Transformers 4.57.3. Main Transformers 5.10.2 is not compatible with Audio8 voice cloning: the official demo collapsed to 0.325s, while the shared runtime generated 10.495s for the published 10.54s demo.
+- Live shared-runtime validation produced 6.36s of reference-free audio in 39.3s including model load, 9.99s of cloned audio in 36.7s warm, and exact cache reuse in 56ms.
+- The engine has unified TTS and SRT support only: no Voice Designer support, special node, training integration, native streaming/timestamps/dialogue, or built-in presets.
+
 ## OmniVoice
 
 ```text

docs/New Engines Guides/NEW_ENGINE_IMPLEMENTATION_GUIDE.md

@@ -466,7 +466,7 @@ Also to test, requirements and dependencies need to be added.
 - [ ] Pause tags work with `[pause:1.5s]`
 - [ ] Caching works (same input = cached output)
 - [ ] Model auto-download works
-- [ ] VRAM management works (model unloads)
+- [ ] VRAM management works (unload, switch to another isolated engine, then switch back with cache disabled)
 - [ ] Different parameter combinations work
 - [ ] Engine prints a standard `Settings:` summary with the active generation/load parameters
 - [ ] **Interrupt handling works** - User can stop SRT generation and it stops within ~1 segment

docs/New Engines Guides/fails_to_avoid_TTS_Engine_Implementation.md

@@ -4,7 +4,7 @@
 
 ### Import Errors
 - **Missing import**: Always add `import folder_paths` when using `folder_paths.get_temp_directory()` in node files
-- **Bundled code imports**: For complex bundled packages with internal cross-imports, add `sys.path.insert(0, impl_dir)` at top of main files instead of converting all imports
+- **Bundled code imports**: Never permanently prepend a bundled implementation directory to `sys.path` in the main ComfyUI process. Generic files such as `utils.py` can shadow suite packages and break later isolated workers. Prefer package-relative or vendor-namespaced imports; if upstream imports make path injection unavoidable, confine it to the isolated worker and load suite protocol modules before adding the vendor path.
 
 ### Audio Utility Functions
 - **Temp file creation**: Use `AudioProcessingUtils.save_audio_to_temp_file()` not `save_audio()` (doesn't exist)

engines/adapters/__init__.py

@@ -49,6 +49,19 @@ except Exception as e:
         def __init__(self, *args, **kwargs):
             raise ImportError(f"Dots TTS adapter not available: {e}")
 
+try:
+    from .audio8_tts_adapter import Audio8TTSEngineAdapter
+    AUDIO8_TTS_ADAPTER_AVAILABLE = True
+except Exception as exc:
+    AUDIO8_TTS_ADAPTER_AVAILABLE = False
+    _AUDIO8_TTS_ADAPTER_ERROR = str(exc)
+
+    class Audio8TTSEngineAdapter:
+        def __init__(self, *args, **kwargs):
+            raise ImportError(
+                f"Audio8 TTS adapter not available: {_AUDIO8_TTS_ADAPTER_ERROR}"
+            )
+
 try:
     from .dramabox_adapter import DramaBoxEngineAdapter
     DRAMABOX_ADAPTER_AVAILABLE = True
@@ -96,10 +109,10 @@ except Exception as e:
 
 __all__ = [
     'ChatterBoxEngineAdapter', 'F5TTSEngineAdapter', 'CosyVoiceAdapter', 'EchoTTSEngineAdapter',
-    'DotsTTSEngineAdapter', 'DramaBoxEngineAdapter', 'OmniVoiceEngineAdapter',
+    'DotsTTSEngineAdapter', 'Audio8TTSEngineAdapter', 'DramaBoxEngineAdapter', 'OmniVoiceEngineAdapter',
     'MossTTSEngineAdapter', 'HiggsAudioV3EngineAdapter',
     'CHATTERBOX_ADAPTER_AVAILABLE', 'F5TTS_ADAPTER_AVAILABLE', 'COSYVOICE_ADAPTER_AVAILABLE',
-    'ECHO_TTS_ADAPTER_AVAILABLE', 'DOTS_TTS_ADAPTER_AVAILABLE',
+    'ECHO_TTS_ADAPTER_AVAILABLE', 'DOTS_TTS_ADAPTER_AVAILABLE', 'AUDIO8_TTS_ADAPTER_AVAILABLE',
     'DRAMABOX_ADAPTER_AVAILABLE', 'OMNIVOICE_ADAPTER_AVAILABLE',
     'MOSS_TTS_ADAPTER_AVAILABLE', 'HIGGS_AUDIO_V3_ADAPTER_AVAILABLE',
     'MossSoundEffectV2Adapter'

engines/adapters/audio8_tts_adapter.py

@@ -0,0 +1,277 @@
+"""Adapter between suite processors and the official Audio8 TTS engine."""
+
+from __future__ import annotations
+
+import os
+from typing import Any, Dict, Optional, Tuple
+
+import torch
+
+from engines.audio8_tts.downloader import Audio8TTSDownloader
+from utils.audio.audio_hash import generate_stable_audio_component
+from utils.audio.cache import get_audio_cache
+from utils.device import resolve_torch_device
+from utils.models.factory_config import ModelLoadConfig, RUNTIME_MODE_SHARED
+from utils.voice.character_logging import resolved_character_label
+from utils.voice.reference import effective_voice_audio
+
+
+class Audio8TTSEngineAdapter:
+    """Translate unified TTS calls into Audio8's official inference API."""
+
+    SAMPLE_RATE = 44100
+
+    def __init__(self, config: Optional[Dict[str, Any]] = None):
+        self.config = dict(config or {})
+        self.audio_cache = get_audio_cache()
+        self.downloader = Audio8TTSDownloader()
+        self._last_config: Optional[ModelLoadConfig] = None
+        self._load_signature: Optional[Tuple[Any, ...]] = None
+
+    def update_config(self, new_config: Dict[str, Any]):
+        self.config = dict(new_config or {})
+
+    def _model_selection(self) -> str:
+        return str(
+            self.config.get(
+                "model_variant",
+                self.config.get(
+                    "model_name",
+                    Audio8TTSDownloader.MODEL_NAME,
+                ),
+            )
+            or Audio8TTSDownloader.MODEL_NAME
+        )
+
+    def _build_load_signature(self) -> Tuple[Any, ...]:
+        return (
+            self._model_selection(),
+            resolve_torch_device(self.config.get("device", "auto")),
+            self.config.get("dtype", "auto"),
+        )
+
+    def load_model(
+        self,
+        model_variant: str,
+        device: str = "auto",
+        dtype: str = "auto",
+    ):
+        """Resolve the organized model path and load through the unified factory."""
+        from utils.models.unified_model_interface import unified_model_interface
+
+        model_path = self.downloader.resolve_model_path(model_variant)
+        config = ModelLoadConfig(
+            engine_name="audio8_tts",
+            model_type="tts",
+            model_name=model_variant,
+            model_path=model_path,
+            device=device,
+            additional_params={"dtype": dtype},
+            runtime_mode=RUNTIME_MODE_SHARED,
+            runtime_profile="vibevoice_transformers4_shared",
+        )
+        self._last_config = config
+        return unified_model_interface.load_model(config)
+
+    def _ensure_model_loaded(self):
+        signature = self._build_load_signature()
+        if signature != self._load_signature or self._last_config is None:
+            self.load_model(
+                model_variant=self._model_selection(),
+                device=self.config.get("device", "auto"),
+                dtype=self.config.get("dtype", "auto"),
+            )
+            self._load_signature = signature
+
+    def _get_engine(self):
+        if self._last_config is None:
+            self._ensure_model_loaded()
+        from utils.models.unified_model_interface import unified_model_interface
+
+        return unified_model_interface.load_model(self._last_config)
+
+    @staticmethod
+    def _mono_waveform(waveform: Any) -> torch.Tensor:
+        audio = torch.as_tensor(waveform, dtype=torch.float32).detach().cpu()
+        if audio.ndim == 3:
+            if audio.shape[0] != 1:
+                raise ValueError(
+                    "Audio8 TTS reference audio must contain one batch item"
+                )
+            audio = audio[0]
+        if audio.ndim == 2:
+            audio = audio.mean(dim=0)
+        if audio.ndim != 1 or audio.numel() == 0:
+            raise ValueError(
+                "Audio8 TTS reference audio must be non-empty mono or "
+                "channels-first audio"
+            )
+        return audio.contiguous()
+
+    def _in_memory_reference(
+        self,
+        waveform: Any,
+        sample_rate: Any,
+    ) -> Tuple[Dict[str, Any], str]:
+        mono = self._mono_waveform(waveform)
+        sample_rate = int(sample_rate)
+        comfy_audio = {
+            "waveform": mono.unsqueeze(0),
+            "sample_rate": sample_rate,
+        }
+        processor_audio = {
+            "array": mono,
+            "sampling_rate": sample_rate,
+        }
+        return (
+            processor_audio,
+            generate_stable_audio_component(reference_audio=comfy_audio),
+        )
+
+    def _extract_voice_reference(
+        self,
+        voice_ref: Optional[Dict[str, Any]],
+    ) -> Tuple[Any, str, str]:
+        if not isinstance(voice_ref, dict):
+            return None, "", "default_voice"
+
+        reference_text = str(
+            voice_ref.get("reference_text")
+            or voice_ref.get("prompt_text")
+            or voice_ref.get("text")
+            or ""
+        ).strip()
+        reference_audio = effective_voice_audio(voice_ref)
+        if reference_audio is None:
+            return None, reference_text, "default_voice"
+
+        if isinstance(reference_audio, (str, os.PathLike)):
+            path = os.fspath(reference_audio)
+            component = generate_stable_audio_component(audio_file_path=path)
+            return path, reference_text, component
+
+        if isinstance(reference_audio, dict):
+            waveform = reference_audio.get(
+                "waveform",
+                reference_audio.get("array"),
+            )
+            sample_rate = reference_audio.get(
+                "sample_rate",
+                reference_audio.get("sampling_rate"),
+            )
+            if waveform is None or sample_rate is None:
+                raise ValueError(
+                    "Audio8 TTS in-memory reference requires waveform and sample_rate"
+                )
+            normalized, component = self._in_memory_reference(
+                waveform,
+                sample_rate,
+            )
+            return normalized, reference_text, component
+
+        if isinstance(reference_audio, (tuple, list)) and len(reference_audio) == 2:
+            normalized, component = self._in_memory_reference(
+                reference_audio[0],
+                reference_audio[1],
+            )
+            return normalized, reference_text, component
+
+        if torch.is_tensor(reference_audio):
+            normalized, component = self._in_memory_reference(
+                reference_audio,
+                voice_ref.get("sample_rate", self.SAMPLE_RATE),
+            )
+            return normalized, reference_text, component
+
+        raise TypeError(
+            f"Unsupported Audio8 TTS reference type: {type(reference_audio)}"
+        )
+
+    def generate_single(
+        self,
+        text: str,
+        voice_ref: Optional[Dict[str, Any]],
+        seed: int = 0,
+        enable_audio_cache: bool = True,
+        character_name: Optional[str] = None,
+    ) -> torch.Tensor:
+        """Generate one raw utterance; chunking remains processor-owned."""
+        text = str(text or "").strip()
+        if not text:
+            return torch.zeros(1, 0, dtype=torch.float32)
+
+        reference_audio, reference_text, audio_component = (
+            self._extract_voice_reference(voice_ref)
+        )
+        if reference_audio is not None and not reference_text:
+            raise ValueError(
+                "Audio8 TTS voice cloning requires the exact transcript of "
+                "the reference audio. Add reference text to Character Voices "
+                "or the narrator voice."
+            )
+
+        params = {
+            "model_variant": self._model_selection(),
+            "max_new_tokens": int(self.config.get("max_new_tokens", 1024)),
+            "retry_max_new_tokens": int(self.config.get("retry_max_new_tokens", 2000)),
+            "temperature": float(self.config.get("temperature", 0.8)),
+            "top_p": float(self.config.get("top_p", 0.95)),
+            "top_k": int(self.config.get("top_k", 50)),
+            "do_sample": bool(self.config.get("do_sample", True)),
+            "dtype": self.config.get("dtype", "auto"),
+            "device": resolve_torch_device(self.config.get("device", "auto")),
+            "seed": int(seed),
+        }
+
+        cache_key = None
+        if enable_audio_cache:
+            cache_key = self.audio_cache.generate_cache_key(
+                "audio8_tts",
+                text=text,
+                audio_component=audio_component,
+                reference_text=reference_text,
+                character=character_name or "narrator",
+                **params,
+            )
+            cached = self.audio_cache.get_cached_audio(cache_key)
+            if cached:
+                display_name = resolved_character_label(
+                    character_name or "narrator",
+                    voice_ref,
+                )
+                print(
+                    "💾 Using cached Audio8 TTS audio for "
+                    f"'{display_name}': '{text[:30]}...'"
+                )
+                return cached[0]
+
+        self._ensure_model_loaded()
+        audio = self._get_engine().generate(
+            text=text,
+            reference_audio=reference_audio,
+            reference_text=reference_text or None,
+            max_new_tokens=params["max_new_tokens"],
+            retry_max_new_tokens=params["retry_max_new_tokens"],
+            temperature=params["temperature"],
+            top_p=params["top_p"],
+            top_k=params["top_k"],
+            do_sample=params["do_sample"],
+            seed=params["seed"],
+        )
+        if not isinstance(audio, torch.Tensor):
+            audio = torch.as_tensor(audio, dtype=torch.float32)
+        audio = audio.detach().float().cpu()
+        if audio.ndim == 1:
+            audio = audio.unsqueeze(0)
+        if audio.ndim != 2:
+            raise RuntimeError(
+                f"Audio8 TTS returned invalid audio shape: {tuple(audio.shape)}"
+            )
+
+        if cache_key:
+            self.audio_cache.cache_audio(
+                cache_key,
+                audio,
+                audio.shape[-1] / self.SAMPLE_RATE,
+            )
+        return audio

engines/audio8_tts/__init__.py

@@ -0,0 +1,6 @@
+"""Audio8 TTS engine integration."""
+
+from .audio8_tts_engine import Audio8TTSEngine
+from .downloader import Audio8TTSDownloader
+
+__all__ = ["Audio8TTSEngine", "Audio8TTSDownloader"]

engines/audio8_tts/audio8_tts_engine.py

@@ -0,0 +1,375 @@
+"""Wrapper for the official Audio8 TTS remote-code model."""
+
+from __future__ import annotations
+
+import os
+import random
+import warnings
+from typing import Any, Dict, Optional, Tuple
+
+import torch
+
+from engines.audio8_tts.downloader import Audio8TTSDownloader
+from engines.audio8_tts.progress import build_audio8_stopping_criteria
+from utils.device import resolve_torch_device
+
+
+class Audio8TTSEngine:
+    """ComfyUI-friendly wrapper around the official Audio8 model."""
+
+    SAMPLE_RATE = 44100
+
+    def __init__(
+        self,
+        model_name: str = Audio8TTSDownloader.MODEL_NAME,
+        device: str = "auto",
+        dtype: str = "auto",
+        model_dir: Optional[str] = None,
+    ):
+        self.model_name = model_name
+        self.device = resolve_torch_device(device)
+        self.dtype_name = str(dtype or "auto").lower()
+        self.dtype = self._resolve_dtype(self.dtype_name, self.device)
+        self.model_dir = model_dir or Audio8TTSDownloader().resolve_model_path(
+            model_name
+        )
+        self._processor = None
+        self._model = None
+        self._codec = None
+
+    @staticmethod
+    def _resolve_dtype(dtype: str, device: str) -> torch.dtype:
+        normalized = str(dtype or "auto").lower()
+        valid = {
+            "bfloat16": torch.bfloat16,
+            "float16": torch.float16,
+            "float32": torch.float32,
+        }
+        if normalized not in {"auto", *valid}:
+            raise ValueError(
+                "Audio8 TTS dtype must be auto, bfloat16, float16, or float32"
+            )
+        if str(device).startswith("cpu"):
+            return torch.float32
+        if normalized in valid:
+            return valid[normalized]
+        if str(device).startswith("cuda") and torch.cuda.is_available():
+            major, _minor = torch.cuda.get_device_capability(torch.device(device))
+            return torch.bfloat16 if major >= 8 else torch.float16
+        if str(device).startswith("xpu"):
+            return torch.bfloat16
+        return torch.float16
+
+    def _ensure_model_loaded(self):
+        """Load the processor, model, and bundled codec strictly from local files."""
+        if (
+            self._processor is not None
+            and self._model is not None
+            and self._codec is not None
+        ):
+            return self
+
+        if not os.path.isdir(self.model_dir):
+            raise FileNotFoundError(
+                f"Audio8 TTS local model directory does not exist: {self.model_dir}"
+            )
+
+        try:
+            from transformers import AutoModel, AutoProcessor
+        except ImportError as exc:
+            raise ImportError(
+                "Audio8 TTS requires the suite's shared Transformers 4 runtime"
+            ) from exc
+
+        print(
+            f"📦 Loading Audio8 TTS from {self.model_dir} "
+            f"on {self.device} ({self.dtype})"
+        )
+        with warnings.catch_warnings():
+            warnings.filterwarnings(
+                "ignore",
+                message=(
+                    r"`torch\.nn\.utils\.weight_norm` is deprecated in favor of "
+                    r"`torch\.nn\.utils\.parametrizations\.weight_norm`\."
+                ),
+                category=FutureWarning,
+            )
+            self._processor = AutoProcessor.from_pretrained(
+                self.model_dir,
+                trust_remote_code=True,
+                local_files_only=True,
+            )
+            self._model = AutoModel.from_pretrained(
+                self.model_dir,
+                trust_remote_code=True,
+                local_files_only=True,
+                dtype=self.dtype,
+            )
+        self._model = self._model.eval().to(self.device)
+
+        # The official model stores the codec outside nn.Module registration.
+        # Hold and move it explicitly so model offload/reload cannot strand it.
+        self._codec = self._model.load_codec(
+            device=self.device,
+            dtype=self.dtype,
+        )
+        self._codec.eval()
+
+        sample_rate = int(
+            getattr(self._model.config, "codec_sample_rate", self.SAMPLE_RATE)
+        )
+        if sample_rate != self.SAMPLE_RATE:
+            raise RuntimeError(
+                f"Audio8 TTS checkpoint reports unexpected sample rate "
+                f"{sample_rate}; expected {self.SAMPLE_RATE}"
+            )
+        print("✅ Audio8 TTS model and codec loaded")
+        return self
+
+    # Compatibility alias for callers using the shorter established name.
+    _ensure_loaded = _ensure_model_loaded
+
+    def _clear_kv_caches(self) -> None:
+        """Drop every static slow/fast attention cache before device movement."""
+        if self._model is None:
+            return
+        for collection_name in ("layers", "fast_layers"):
+            for layer in getattr(self._model, collection_name, ()) or ():
+                attention = getattr(layer, "attention", None)
+                if attention is not None and hasattr(attention, "kv_cache"):
+                    attention.kv_cache = None
+
+    def to(self, device):
+        """Move all weights for ComfyUI Clear VRAM and reload handling."""
+        target = resolve_torch_device(
+            str(device) if not isinstance(device, str) else device
+        )
+        self._clear_kv_caches()
+        self.device = target
+        if self._model is not None:
+            self._model = self._model.to(target).eval()
+        if self._codec is not None:
+            codec_dtype = torch.float32 if str(target).startswith("cpu") else self.dtype
+            self._codec = self._codec.to(
+                device=target,
+                dtype=codec_dtype,
+            ).eval()
+        return self
+
+    def parameters(self):
+        """Expose all weight tensors to ComfyUI memory accounting."""
+        if self._model is not None:
+            yield from self._model.parameters()
+        if self._codec is not None:
+            yield from self._codec.parameters()
+
+    @staticmethod
+    def _mono_audio(waveform: Any) -> torch.Tensor:
+        audio = torch.as_tensor(waveform, dtype=torch.float32).detach().cpu()
+        if audio.ndim == 3:
+            if audio.shape[0] != 1:
+                raise ValueError(
+                    "Audio8 TTS reference audio must contain one batch item"
+                )
+            audio = audio[0]
+        if audio.ndim == 2:
+            audio = audio.mean(dim=0)
+        if audio.ndim != 1 or audio.numel() == 0:
+            raise ValueError(
+                "Audio8 TTS reference audio must be non-empty mono or "
+                "channels-first audio"
+            )
+        return audio.contiguous()
+
+    def _normalize_reference_audio(
+        self,
+        reference_audio: Any,
+    ) -> Tuple[Any, Optional[int]]:
+        if isinstance(reference_audio, (str, os.PathLike)):
+            return os.fspath(reference_audio), None
+        if isinstance(reference_audio, dict):
+            waveform = reference_audio.get(
+                "waveform",
+                reference_audio.get("array"),
+            )
+            sample_rate = reference_audio.get(
+                "sample_rate",
+                reference_audio.get("sampling_rate"),
+            )
+            if waveform is None or sample_rate is None:
+                raise ValueError(
+                    "Audio8 TTS in-memory reference audio requires waveform "
+                    "and sample_rate"
+                )
+            return {
+                "array": self._mono_audio(waveform),
+                "sampling_rate": int(sample_rate),
+            }, int(sample_rate)
+        if isinstance(reference_audio, (tuple, list)) and len(reference_audio) == 2:
+            waveform, sample_rate = reference_audio
+            if not isinstance(sample_rate, (int, float)):
+                raise ValueError(
+                    "Audio8 TTS reference tuple must be (waveform, sample_rate)"
+                )
+            return {
+                "array": self._mono_audio(waveform),
+                "sampling_rate": int(sample_rate),
+            }, int(sample_rate)
+        raise TypeError(
+            f"Unsupported Audio8 TTS reference audio type: {type(reference_audio)}"
+        )
+
+    def _prepare_inputs(
+        self,
+        text: str,
+        reference_audio: Any,
+        reference_text: Optional[str],
+    ) -> Dict[str, torch.Tensor]:
+        processor_kwargs: Dict[str, Any] = {
+            "text": [text],
+            "return_tensors": "pt",
+        }
+        if reference_audio is not None:
+            normalized_audio, sample_rate = self._normalize_reference_audio(
+                reference_audio
+            )
+            processor_kwargs.update(
+                reference_audio=[normalized_audio],
+                reference_text=[reference_text],
+            )
+            if sample_rate is not None:
+                processor_kwargs["sampling_rate"] = [sample_rate]
+
+        inputs = self._processor(**processor_kwargs)
+        return {name: value.to(self.device) for name, value in inputs.items()}
+
+    def _make_generator(self, seed: int):
+        seed = int(seed)
+        random.seed(seed)
+        torch.manual_seed(seed)
+        if torch.cuda.is_available():
+            torch.cuda.manual_seed_all(seed)
+        try:
+            return torch.Generator(device=torch.device(self.device)).manual_seed(seed)
+        except Exception:
+            return None
+
+    def _generate_once(
+        self,
+        inputs: Dict[str, torch.Tensor],
+        *,
+        max_new_tokens: int,
+        temperature: float,
+        top_p: float,
+        top_k: int,
+        do_sample: bool,
+        generator,
+    ):
+        criteria, tracker = build_audio8_stopping_criteria(max_new_tokens)
+        try:
+            output = self._model.generate(
+                **inputs,
+                max_new_tokens=max_new_tokens,
+                temperature=temperature,
+                top_p=top_p,
+                top_k=top_k,
+                do_sample=do_sample,
+                generator=generator,
+                stopping_criteria=criteria,
+                return_dict_in_generate=True,
+            )
+        except BaseException:
+            tracker.abort()
+            raise
+        tracker.close()
+        return output
+
+    def generate(
+        self,
+        text: str,
+        reference_audio: Any = None,
+        reference_text: Optional[str] = None,
+        *,
+        max_new_tokens: int = 1024,
+        retry_max_new_tokens: int = 2000,
+        temperature: float = 0.8,
+        top_p: float = 0.95,
+        top_k: int = 50,
+        do_sample: bool = True,
+        seed: int = 42,
+    ) -> torch.Tensor:
+        """Generate one utterance as a CPU float tensor ``[channels, samples]``."""
+        text = str(text or "").strip()
+        if not text:
+            return torch.zeros(1, 0, dtype=torch.float32)
+
+        reference_text = str(reference_text or "").strip()
+        if reference_audio is not None and not reference_text:
+            raise ValueError(
+                "Audio8 TTS voice cloning requires the exact transcript of "
+                "the reference audio"
+            )
+        if reference_audio is None and reference_text:
+            raise ValueError(
+                "Audio8 TTS reference_text requires matching reference audio"
+            )
+
+        max_new_tokens = int(max_new_tokens)
+        retry_max_new_tokens = int(retry_max_new_tokens)
+        temperature = float(temperature)
+        top_p = float(top_p)
+        top_k = int(top_k)
+        if max_new_tokens < 1:
+            raise ValueError("Audio8 TTS max_new_tokens must be positive")
+        if retry_max_new_tokens < max_new_tokens:
+            raise ValueError(
+                "Audio8 TTS retry_max_new_tokens must be >= max_new_tokens"
+            )
+        if temperature <= 0:
+            raise ValueError("Audio8 TTS temperature must be positive")
+        if not 0 < top_p <= 1:
+            raise ValueError("Audio8 TTS top_p must be in (0, 1]")
+        if top_k < 1:
+            raise ValueError("Audio8 TTS top_k must be at least 1")
+
+        self._ensure_model_loaded()
+        inputs = self._prepare_inputs(text, reference_audio, reference_text)
+        generator = self._make_generator(seed)
+
+        output = self._generate_once(
+            inputs,
+            max_new_tokens=max_new_tokens,
+            temperature=temperature,
+            top_p=top_p,
+            top_k=top_k,
+            do_sample=bool(do_sample),
+            generator=generator,
+        )
+        finished = bool(output.finished[0].item())
+
+        if not finished and retry_max_new_tokens > max_new_tokens:
+            print(
+                "⚠️ Audio8 TTS reached max_new_tokens without EOS; "
+                f"retrying with {retry_max_new_tokens}"
+            )
+            output = self._generate_once(
+                inputs,
+                max_new_tokens=retry_max_new_tokens,
+                temperature=temperature,
+                top_p=top_p,
+                top_k=top_k,
+                do_sample=bool(do_sample),
+                generator=generator,
+            )
+            finished = bool(output.finished[0].item())
+
+        if not finished:
+            print(
+                "⚠️ Audio8 TTS generation ended without EOS; returning the "
+                "valid decoded frames"
+            )
+
+        waveforms, waveform_lengths = self._model.decode_audio(output.codes)
+        waveform_length = int(waveform_lengths[0].item())
+        waveform = waveforms[0, :waveform_length].detach().float().cpu()
+        return waveform.unsqueeze(0).contiguous()

engines/audio8_tts/downloader.py

@@ -0,0 +1,180 @@
+"""Organized model download and discovery for official Audio8 TTS."""
+
+from __future__ import annotations
+
+import os
+from typing import List, Optional
+
+import folder_paths
+
+from utils.downloads.unified_downloader import unified_downloader
+from utils.models.extra_paths import (
+    get_all_tts_model_paths,
+    get_preferred_download_path,
+)
+
+
+class Audio8TTSDownloader:
+    """Resolve Audio8 checkpoints without using the Hugging Face model cache."""
+
+    MODEL_NAME = "Audio8-TTS-Preview-0.6b"
+    REPO_ID = "Audio8/Audio8-TTS-Preview-0.6b"
+    REVISION = "1b17c91db5f4dccb6914aa4aa5cb0e56661a6c17"
+
+    # Complete runtime snapshot at the pinned revision. Non-runtime model-card
+    # assets are deliberately excluded.
+    REQUIRED_FILES = [
+        "codec.pth",
+        "config.json",
+        "configuration_arktts.py",
+        "generation_config.json",
+        "model.safetensors",
+        "modeling_arktts.py",
+        "modeling_arktts_codec.py",
+        "preprocessor_config.json",
+        "processing_arktts.py",
+        "processor_config.json",
+        "special_tokens_map.json",
+        "tokenizer.json",
+        "tokenizer_config.json",
+    ]
+
+    def __init__(self, base_path: Optional[str] = None):
+        if base_path is None:
+            try:
+                self.base_path = get_preferred_download_path(
+                    model_type="TTS",
+                    engine_name="audio8_tts",
+                )
+            except Exception:
+                self.base_path = os.path.join(
+                    folder_paths.models_dir,
+                    "TTS",
+                    "audio8_tts",
+                )
+        else:
+            self.base_path = os.path.abspath(os.fspath(base_path))
+        os.makedirs(self.base_path, exist_ok=True)
+
+    def get_available_models(self) -> List[str]:
+        """Return the canonical model plus complete local installations."""
+        models = [self.MODEL_NAME]
+        canonical_path = os.path.normcase(
+            os.path.abspath(os.path.join(self.base_path, self.MODEL_NAME))
+        )
+        for base_path in get_all_tts_model_paths("TTS"):
+            for folder_name in ("audio8_tts", "Audio8_TTS", ""):
+                root = (
+                    os.path.join(base_path, folder_name) if folder_name else base_path
+                )
+                if not os.path.isdir(root):
+                    continue
+                for item in sorted(os.listdir(root)):
+                    candidate = os.path.join(root, item)
+                    if os.path.normcase(os.path.abspath(candidate)) == canonical_path:
+                        continue
+                    local_name = f"local:{item}"
+                    if local_name not in models and self._is_model_complete(candidate):
+                        models.append(local_name)
+        return models
+
+    def resolve_model_path(
+        self,
+        model_identifier: str = MODEL_NAME,
+    ) -> str:
+        """Resolve an absolute path, ``local:`` name, or canonical model name."""
+        model_identifier = str(model_identifier or self.MODEL_NAME).strip()
+
+        if os.path.isabs(model_identifier) or os.path.isdir(model_identifier):
+            candidate = os.path.abspath(model_identifier)
+            if self._is_model_complete(candidate, verbose=True):
+                return candidate
+            raise FileNotFoundError(
+                f"Audio8 TTS model path is missing required files: {candidate}"
+            )
+
+        if model_identifier.startswith("local:"):
+            local_name = model_identifier[6:].strip()
+            if not local_name:
+                raise ValueError("Audio8 TTS local model name must not be empty")
+            for base_path in get_all_tts_model_paths("TTS"):
+                for folder_name in ("audio8_tts", "Audio8_TTS", ""):
+                    candidate = (
+                        os.path.join(base_path, folder_name, local_name)
+                        if folder_name
+                        else os.path.join(base_path, local_name)
+                    )
+                    if self._is_model_complete(candidate):
+                        print(f"📁 Using local Audio8 TTS model: {candidate}")
+                        return candidate
+            raise FileNotFoundError(
+                f"Local Audio8 TTS model not found or incomplete: {local_name}"
+            )
+
+        if model_identifier != self.MODEL_NAME:
+            raise ValueError(f"Unknown Audio8 TTS model: {model_identifier}")
+        return self.get_model_path()
+
+    def get_model_path(self, model_name: str = MODEL_NAME) -> str:
+        """Return the organized canonical model path, downloading if needed."""
+        if model_name != self.MODEL_NAME:
+            raise ValueError(f"Unknown Audio8 TTS model: {model_name}")
+        model_dir = os.path.join(self.base_path, self.MODEL_NAME)
+        if not self._is_model_complete(model_dir):
+            return self.download_model(model_dir)
+        return model_dir
+
+    def download_model(
+        self,
+        model_dir: Optional[str] = None,
+        force: bool = False,
+    ) -> str:
+        """Download the pinned official snapshot into the organized model folder."""
+        model_dir = model_dir or os.path.join(self.base_path, self.MODEL_NAME)
+
+        print("\n" + "=" * 60)
+        print("📦 Audio8 TTS Model Download")
+        print("=" * 60)
+        print(f"Repository: {self.REPO_ID}")
+        print(f"Revision: {self.REVISION}")
+        print(f"Target: {model_dir}")
+        print("=" * 60 + "\n")
+
+        unified_downloader.download_huggingface_snapshot(
+            repo_id=self.REPO_ID,
+            target_dir=model_dir,
+            revision=self.REVISION,
+            allow_patterns=self.REQUIRED_FILES,
+            required_files=self.REQUIRED_FILES,
+            force_download=bool(force),
+            description=self.MODEL_NAME,
+        )
+        if not self._is_model_complete(model_dir, verbose=True):
+            raise RuntimeError(
+                f"Downloaded Audio8 TTS model is incomplete: {model_dir}"
+            )
+        print(f"✅ Audio8 TTS model ready: {model_dir}")
+        return model_dir
+
+    def _is_model_complete(
+        self,
+        model_dir: str,
+        *,
+        verbose: bool = False,
+    ) -> bool:
+        if not os.path.isdir(model_dir):
+            return False
+        missing = [
+            rel_path
+            for rel_path in self.REQUIRED_FILES
+            if not os.path.isfile(os.path.join(model_dir, rel_path))
+            or os.path.getsize(os.path.join(model_dir, rel_path)) <= 0
+        ]
+        if missing and verbose:
+            print(
+                f"❌ Audio8 TTS model incomplete. Missing or empty "
+                f"{len(missing)} file(s):"
+            )
+            for rel_path in missing:
+                print(f"   - {rel_path}")
+        return not missing

engines/audio8_tts/progress.py

@@ -0,0 +1,116 @@
+"""Native generation progress and interruption support for Audio8 TTS."""
+
+from __future__ import annotations
+
+import time
+from typing import Any, Optional
+
+import torch
+from transformers import StoppingCriteria, StoppingCriteriaList
+
+
+def _make_comfy_progress_bar(total_steps: int) -> Optional[Any]:
+    try:
+        from comfy.utils import ProgressBar
+
+        return ProgressBar(total_steps)
+    except Exception:
+        return None
+
+
+class Audio8TTSStoppingCriteria(StoppingCriteria):
+    """Track native codec-frame generation and honor ComfyUI interruption."""
+
+    def __init__(self, max_steps: int, progress_bar: Optional[Any] = None):
+        self.max_steps = max(1, int(max_steps))
+        self.progress_bar = progress_bar
+        self.steps = 0
+        self.start_time = time.monotonic()
+        self.last_print_time = self.start_time
+        self.last_print_step = 0
+        self.rendered = False
+
+    @staticmethod
+    def _check_interrupted() -> None:
+        try:
+            import comfy.model_management as model_management
+        except Exception:
+            return
+
+        checker = getattr(
+            model_management,
+            "throw_exception_if_processing_interrupted",
+            None,
+        )
+        if callable(checker):
+            checker()
+        elif getattr(model_management, "interrupt_processing", False):
+            raise InterruptedError("Audio8 TTS generation interrupted by user")
+
+    def __call__(self, input_ids, scores, **kwargs):
+        del scores, kwargs
+        self._check_interrupted()
+        self.steps = min(self.steps + 1, self.max_steps)
+        if self.progress_bar is not None:
+            try:
+                self.progress_bar.update(1)
+            except Exception:
+                pass
+
+        now = time.monotonic()
+        if (
+            self.steps == 1
+            or self.steps >= self.max_steps
+            or now - self.last_print_time >= 0.5
+        ):
+            delta_time = max(now - self.last_print_time, 1e-6)
+            delta_steps = self.steps - self.last_print_step
+            rate = delta_steps / delta_time
+            elapsed = now - self.start_time
+            width = 12
+            filled = min(
+                width,
+                int(width * self.steps / self.max_steps),
+            )
+            bar = "█" * filled + "░" * (width - filled)
+            print(
+                f"\r   Audio8: [{bar}] {self.steps}/{self.max_steps} | "
+                f"{rate:.1f} frames/s | {elapsed:.1f}s      ",
+                end="",
+                flush=True,
+            )
+            self.last_print_time = now
+            self.last_print_step = self.steps
+            self.rendered = True
+
+        # The official model handles EOS itself. Interruption raises above.
+        batch_size = int(input_ids.shape[0]) if input_ids.ndim else 1
+        return torch.zeros(
+            batch_size,
+            dtype=torch.bool,
+            device=input_ids.device,
+        )
+
+    def close(self) -> None:
+        elapsed = max(time.monotonic() - self.start_time, 1e-6)
+        average = self.steps / elapsed
+        if self.rendered:
+            print(
+                f"\r   Audio8 complete: {self.steps} frames in "
+                f"{elapsed:.1f}s ({average:.1f} frames/s)" + " " * 20
+            )
+
+    def abort(self) -> None:
+        if self.rendered:
+            print()
+
+
+def build_audio8_stopping_criteria(
+    max_steps: int,
+) -> tuple[StoppingCriteriaList, Audio8TTSStoppingCriteria]:
+    """Create the native stopping-criteria list and its progress tracker."""
+    tracker = Audio8TTSStoppingCriteria(
+        max_steps=max_steps,
+        progress_bar=_make_comfy_progress_bar(max(1, int(max_steps))),
+    )
+    return StoppingCriteriaList([tracker]), tracker

nodes.py

@@ -179,6 +179,14 @@ except Exception as e:
     print(f"❌ Dots TTS Engine failed: {e}")
     DOTS_TTS_ENGINE_AVAILABLE = False
 
+try:
+    audio8_tts_engine_module = load_node_module("audio8_tts_engine_node", "engines/audio8_tts_engine_node.py")
+    Audio8TTSEngineNode = audio8_tts_engine_module.Audio8TTSEngineNode
+    AUDIO8_TTS_ENGINE_AVAILABLE = True
+except Exception as e:
+    print(f"❌ Audio8 TTS Engine failed: {e}")
+    AUDIO8_TTS_ENGINE_AVAILABLE = False
+
 try:
     dramabox_engine_module = load_node_module("dramabox_engine_node", "engines/dramabox_engine_node.py")
     DramaBoxEngineNode = dramabox_engine_module.DramaBoxEngineNode
@@ -670,6 +678,10 @@ if DOTS_TTS_ENGINE_AVAILABLE:
     NODE_CLASS_MAPPINGS["DotsTTSEngineNode"] = DotsTTSEngineNode
     NODE_DISPLAY_NAME_MAPPINGS["DotsTTSEngineNode"] = "⚙️ Dots TTS Engine"
 
+if AUDIO8_TTS_ENGINE_AVAILABLE:
+    NODE_CLASS_MAPPINGS["Audio8TTSEngineNode"] = Audio8TTSEngineNode
+    NODE_DISPLAY_NAME_MAPPINGS["Audio8TTSEngineNode"] = "⚙️ Audio8 TTS Engine"
+
 if DRAMABOX_ENGINE_AVAILABLE:
     NODE_CLASS_MAPPINGS["DramaBoxEngineNode"] = DramaBoxEngineNode
     NODE_DISPLAY_NAME_MAPPINGS["DramaBoxEngineNode"] = "⚙️ DramaBox Engine"

nodes/audio8_tts/__init__.py

@@ -0,0 +1 @@
+"""Audio8 TTS text and SRT processors."""

nodes/audio8_tts/audio8_tts_processor.py

@@ -0,0 +1,403 @@
+"""Suite orchestration for Audio8 TTS text generation."""
+
+import os
+import re
+import sys
+from typing import Any, Dict, List, Tuple, Union
+
+import comfy.model_management as model_management
+import torch
+
+current_dir = os.path.dirname(__file__)
+nodes_dir = os.path.dirname(current_dir)
+project_root = os.path.dirname(nodes_dir)
+if project_root not in sys.path:
+    sys.path.insert(0, project_root)
+
+from utils.audio.chunk_timing import ChunkTimingHelper
+from utils.audio.edit_post_processor import (
+    process_segments as apply_edit_post_processing,
+)
+from utils.text.character_parser import character_parser
+from utils.text.pause_processor import PauseTagProcessor
+from utils.text.segment_parameters import ParameterValidator, apply_segment_parameters
+from utils.text.step_audio_editx_special_tags import get_edit_tags_for_segment
+from utils.voice.character_logging import (
+    format_resolved_character_block,
+    resolved_character_label,
+)
+from utils.voice.discovery import (
+    get_available_characters,
+    get_character_mapping,
+    voice_discovery,
+)
+from utils.voice.reference import effective_voice_audio
+
+
+class Audio8TTSProcessor:
+    """Handle suite-level tags, voices, chunking, and Audio8 generation."""
+
+    SAMPLE_RATE = 44100
+
+    def __init__(self, adapter, engine_config: Dict[str, Any]):
+        self.adapter = adapter
+        self.config = engine_config.copy() if engine_config else {}
+
+    def update_config(self, new_config: Dict[str, Any]):
+        self.config = new_config.copy() if new_config else {}
+        self.adapter.update_config(self.config)
+
+    @staticmethod
+    def _check_interrupt(context: str = "") -> None:
+        if not model_management.interrupt_processing:
+            return
+        suffix = f" {context}" if context else ""
+        raise InterruptedError(f"Audio8 TTS generation interrupted{suffix}")
+
+    def _setup_character_parser(self, text: str) -> None:
+        character_tags = re.findall(r"\[([^\]]+)\]", text or "")
+        characters_from_tags = [
+            tag.split("|")[0].strip()
+            for tag in character_tags
+            if not tag.lower().startswith("pause:")
+        ]
+
+        all_available = set(get_available_characters() or [])
+        for alias, target in voice_discovery.get_character_aliases().items():
+            all_available.add(alias.lower())
+            all_available.add(target.lower())
+        all_available.update(
+            character.lower() for character in characters_from_tags if character
+        )
+        all_available.add("narrator")
+
+        character_parser.set_available_characters(list(all_available))
+        character_parser.reset_session_cache()
+
+    @staticmethod
+    def _as_voice_reference(value: Any) -> Dict[str, Any]:
+        if isinstance(value, dict):
+            return value.copy()
+        if value is None:
+            return {}
+        return {"audio": value}
+
+    def _resolve_voice(
+        self,
+        character: str,
+        voice_mapping: Dict[str, Any],
+        discovered_mapping: Dict[str, Tuple[Any, Any]],
+    ) -> Dict[str, Any]:
+        if character in voice_mapping:
+            mapped_voice = self._as_voice_reference(voice_mapping[character])
+            if effective_voice_audio(mapped_voice) is not None or mapped_voice.get(
+                "reference_text"
+            ):
+                return mapped_voice
+
+        if character != "narrator":
+            audio_path, reference_text = discovered_mapping.get(character, (None, None))
+            if audio_path:
+                return {
+                    "audio_path": audio_path,
+                    "reference_text": reference_text or "",
+                }
+            print(
+                f"⚠️ Audio8 TTS: No voice found for '{character}'; "
+                "using narrator/no-reference fallback"
+            )
+
+        return self._as_voice_reference(voice_mapping.get("narrator"))
+
+    @staticmethod
+    def _voice_log_note(voice_ref: Dict[str, Any]) -> str:
+        reference_text = ""
+        if isinstance(voice_ref, dict):
+            reference_text = str(voice_ref.get("reference_text") or "").strip()
+        has_audio = (
+            isinstance(voice_ref, dict) and effective_voice_audio(voice_ref) is not None
+        )
+
+        if not has_audio and not reference_text:
+            return " [no reference voice]"
+        if has_audio and reference_text:
+            return f" [reference transcript: {len(reference_text)} chars]"
+        if has_audio:
+            return " [reference audio has no transcript; adapter will validate]"
+        return " [reference transcript has no audio; adapter will validate]"
+
+    @staticmethod
+    def _format_parameter_log(
+        filtered_params: Dict[str, Any],
+        current_config: Dict[str, Any],
+        current_seed: int,
+    ) -> str:
+        if not filtered_params:
+            return ""
+
+        values = []
+        for key in ("seed", "temperature", "top_p", "top_k", "max_new_tokens"):
+            if key not in filtered_params:
+                continue
+            value = current_seed if key == "seed" else current_config.get(key)
+            values.append(f"{key}={value}")
+        return ", ".join(values)
+
+    @staticmethod
+    def _validate_segment_config(
+        config: Dict[str, Any],
+        filtered_params: Dict[str, Any],
+    ) -> Dict[str, Any]:
+        """Apply Audio8-specific bounds and keep retry budgets consistent."""
+        if "top_p" in filtered_params and float(config["top_p"]) <= 0:
+            raise ValueError("Audio8 segment top_p must be greater than 0")
+
+        if "max_new_tokens" in filtered_params:
+            max_new_tokens = int(config["max_new_tokens"])
+            if max_new_tokens > 2048:
+                raise ValueError("Audio8 segment max_new_tokens must be at most 2048")
+            config["retry_max_new_tokens"] = max(
+                max_new_tokens,
+                int(config.get("retry_max_new_tokens", 2000)),
+            )
+        return config
+
+    def _log_generation(
+        self,
+        character: str,
+        text: str,
+        voice_ref: Dict[str, Any],
+        config: Dict[str, Any],
+        chunk_count: int,
+        parameter_log: str,
+        show_text_logging: bool,
+    ) -> None:
+        display_name = resolved_character_label(character, voice_ref)
+        print(
+            f"🎭 Audio8 TTS - Generating for '{display_name}'"
+            f"{self._voice_log_note(voice_ref)}"
+        )
+        print(
+            "   Settings: "
+            f"mode={'Sampling' if config.get('do_sample', True) else 'Greedy'}, "
+            f"temperature={config.get('temperature', 0.8)}, "
+            f"top_p={config.get('top_p', 0.95)}, "
+            f"top_k={config.get('top_k', 50)}, "
+            f"max_new_tokens={config.get('max_new_tokens', 1024)}"
+        )
+        if parameter_log:
+            print(f"🎛️ Audio8 TTS segment params: {parameter_log}")
+        if show_text_logging:
+            print(format_resolved_character_block(character, text, voice_ref))
+        if chunk_count > 1:
+            print(
+                f"📝 Audio8 TTS: Chunking '{display_name}' into "
+                f"{chunk_count} suite chunks"
+            )
+
+    @staticmethod
+    def _normalize_audio(audio: Any) -> torch.Tensor:
+        if not isinstance(audio, torch.Tensor):
+            audio = torch.as_tensor(audio, dtype=torch.float32)
+        audio = audio.detach().to(device="cpu", dtype=torch.float32)
+        if audio.dim() == 1:
+            audio = audio.unsqueeze(0)
+        elif audio.dim() == 3 and audio.shape[0] == 1:
+            audio = audio.squeeze(0)
+        if audio.dim() != 2:
+            raise ValueError(
+                "Audio8 adapter returned an invalid waveform shape; "
+                "expected [channels, samples]"
+            )
+        return audio
+
+    def process_text(
+        self,
+        text: str,
+        voice_mapping: Dict[str, Any],
+        seed: int,
+        enable_chunking: bool = True,
+        max_chars_per_chunk: int = 400,
+        chunk_combination_method: str = "auto",
+        silence_between_chunks_ms: int = 100,
+        enable_audio_cache: bool = True,
+        apply_edit_postprocessing: bool = True,
+        show_text_logging: bool = True,
+        **_unused,
+    ) -> List[Dict[str, Any]]:
+        del chunk_combination_method, silence_between_chunks_ms
+
+        self._check_interrupt()
+        self._setup_character_parser(text)
+        base_config = self.config.copy()
+        segment_objects = character_parser.parse_text_segments(text)
+        if not segment_objects:
+            segment_objects = character_parser.parse_text_segments(
+                "narrator " + (text or "")
+            )
+
+        characters = list(
+            {segment.character for segment in segment_objects if segment.character}
+        )
+        discovered_mapping = get_character_mapping(characters, engine_type="audio_only")
+        segment_records: List[Dict[str, Any]] = []
+
+        try:
+            for segment_index, segment in enumerate(segment_objects):
+                character = segment.character or "narrator"
+                self._check_interrupt(
+                    f"before segment {segment_index + 1}/{len(segment_objects)} "
+                    f"for '{character}'"
+                )
+                segment_text = (segment.text or "").strip()
+                if not segment_text:
+                    continue
+
+                segment_params = segment.parameters or {}
+                filtered_params = ParameterValidator.filter_parameters_for_engine(
+                    segment_params, "audio8_tts"
+                )
+                current_config = (
+                    apply_segment_parameters(base_config, filtered_params, "audio8_tts")
+                    if filtered_params
+                    else base_config.copy()
+                )
+                current_config = self._validate_segment_config(
+                    current_config, filtered_params
+                )
+                current_seed = int(current_config.get("seed", seed))
+                parameter_log = self._format_parameter_log(
+                    filtered_params, current_config, current_seed
+                )
+                self.adapter.update_config(current_config)
+
+                voice_ref = self._resolve_voice(
+                    character, voice_mapping, discovered_mapping
+                )
+                seed_offset = 0
+
+                def generate_chunks(text_content: str, edit_tags: list) -> None:
+                    nonlocal seed_offset
+                    clean_content = (text_content or "").strip()
+                    if not clean_content:
+                        return
+
+                    if enable_chunking:
+                        from utils.text.chunking import ImprovedChatterBoxChunker
+
+                        max_chars = ImprovedChatterBoxChunker.validate_chunking_params(
+                            max_chars_per_chunk
+                        )
+                        chunks = ImprovedChatterBoxChunker.split_into_chunks(
+                            clean_content, max_chars=max_chars
+                        )
+                    else:
+                        chunks = [clean_content]
+                    chunks = [chunk.strip() for chunk in chunks if chunk.strip()]
+
+                    self._log_generation(
+                        character=character,
+                        text=clean_content,
+                        voice_ref=voice_ref,
+                        config=current_config,
+                        chunk_count=len(chunks),
+                        parameter_log=parameter_log,
+                        show_text_logging=show_text_logging,
+                    )
+
+                    for chunk_index, chunk in enumerate(chunks):
+                        self._check_interrupt(
+                            f"before chunk {chunk_index + 1}/{len(chunks)} "
+                            f"for '{character}'"
+                        )
+                        chunk_seed = current_seed + seed_offset
+                        seed_offset += 1
+                        audio = self.adapter.generate_single(
+                            text=chunk,
+                            voice_ref=voice_ref,
+                            seed=chunk_seed,
+                            enable_audio_cache=enable_audio_cache,
+                            character_name=character,
+                        )
+                        segment_records.append(
+                            {
+                                "waveform": self._normalize_audio(audio),
+                                "sample_rate": self.SAMPLE_RATE,
+                                "text": chunk,
+                                "edit_tags": (edit_tags if chunk_index == 0 else []),
+                            }
+                        )
+
+                if PauseTagProcessor.has_pause_tags(segment_text):
+                    pause_segments, _ = PauseTagProcessor.parse_pause_tags(segment_text)
+                    for fragment_type, fragment_content in pause_segments:
+                        self._check_interrupt(
+                            f"while processing segment {segment_index + 1}"
+                        )
+                        if fragment_type == "text":
+                            clean_text, edit_tags = get_edit_tags_for_segment(
+                                fragment_content
+                            )
+                            generate_chunks(clean_text, edit_tags)
+                        elif fragment_type == "pause":
+                            silence = PauseTagProcessor.create_silence_segment(
+                                fragment_content,
+                                self.SAMPLE_RATE,
+                                torch.device("cpu"),
+                                torch.float32,
+                            )
+                            if silence.dim() == 1:
+                                silence = silence.unsqueeze(0)
+                            segment_records.append(
+                                {
+                                    "waveform": silence.cpu(),
+                                    "sample_rate": self.SAMPLE_RATE,
+                                    "text": f"[pause:{fragment_content}s]",
+                                    "edit_tags": [],
+                                }
+                            )
+                else:
+                    clean_text, edit_tags = get_edit_tags_for_segment(segment_text)
+                    generate_chunks(clean_text, edit_tags)
+        finally:
+            self.adapter.update_config(base_config)
+
+        if (
+            apply_edit_postprocessing
+            and segment_records
+            and any(record.get("edit_tags") for record in segment_records)
+        ):
+            self._check_interrupt("before edit post-processing")
+            segment_records = apply_edit_post_processing(
+                segment_records, engine_config=base_config
+            )
+            for record in segment_records:
+                record["waveform"] = self._normalize_audio(record["waveform"])
+
+        return segment_records
+
+    def combine_audio_segments(
+        self,
+        segments: List[Dict[str, Any]],
+        method: str = "auto",
+        silence_ms: int = 100,
+        original_text: str = "",
+        return_info: bool = False,
+    ) -> Union[torch.Tensor, Tuple[torch.Tensor, Dict[str, Any]]]:
+        self._check_interrupt("before audio assembly")
+        if not segments:
+            empty = torch.zeros(0, dtype=torch.float32)
+            return (empty, {}) if return_info else empty
+
+        text_chunks = [segment.get("text", "") for segment in segments]
+        combined_audio, chunk_info = ChunkTimingHelper.combine_audio_with_timing(
+            audio_segments=[segment["waveform"] for segment in segments],
+            combination_method=method,
+            silence_ms=silence_ms,
+            crossfade_duration=0.1,
+            sample_rate=self.SAMPLE_RATE,
+            text_length=len(" ".join(text_chunks)),
+            original_text=original_text,
+            text_chunks=text_chunks,
+        )
+        return (combined_audio, chunk_info) if return_info else combined_audio

nodes/audio8_tts/audio8_tts_srt_processor.py

@@ -0,0 +1,222 @@
+"""SRT orchestration for Audio8 TTS."""
+
+import importlib.util
+import os
+
+import comfy.model_management as model_management
+import torch
+
+from engines.adapters.audio8_tts_adapter import Audio8TTSEngineAdapter
+from utils.system.import_manager import import_manager
+from utils.timing.assembly import AudioAssemblyEngine
+from utils.timing.engine import TimingEngine
+from utils.timing.overlap_detection import SRTOverlapHandler
+from utils.timing.reporting import SRTReportGenerator
+
+
+class Audio8TTSSRTProcessor:
+    """Generate Audio8 speech per subtitle and apply shared timing modes."""
+
+    SAMPLE_RATE = 44100
+
+    def __init__(self, node_instance, config):
+        self.node_instance = node_instance
+        self.config = dict(config or {})
+        self.adapter = Audio8TTSEngineAdapter(self.config)
+        self._processor = None
+
+        success, modules, message = import_manager.import_srt_modules()
+        if not success:
+            raise ImportError(f"Audio8 TTS SRT unavailable: {message}")
+        self.SRTParser = modules["SRTParser"]
+
+    @property
+    def processor(self):
+        if self._processor is None:
+            processor_path = os.path.join(
+                os.path.dirname(__file__), "audio8_tts_processor.py"
+            )
+            spec = importlib.util.spec_from_file_location(
+                "audio8_tts_processor_module", processor_path
+            )
+            module = importlib.util.module_from_spec(spec)
+            spec.loader.exec_module(module)
+            self._processor = module.Audio8TTSProcessor(self.adapter, self.config)
+        return self._processor
+
+    def update_config(self, config):
+        self.config = dict(config or {})
+        self.adapter.update_config(self.config)
+        if self._processor is not None:
+            self._processor.update_config(self.config)
+
+    @staticmethod
+    def _check_interrupt(index=None, total=None):
+        if not model_management.interrupt_processing:
+            return
+        location = f" at subtitle {index + 1}/{total}" if index is not None else ""
+        raise InterruptedError(f"Audio8 TTS SRT generation interrupted{location}")
+
+    def process_srt_content(
+        self,
+        srt_content,
+        voice_mapping,
+        seed,
+        timing_mode,
+        timing_params,
+        enable_audio_cache=True,
+    ):
+        self._check_interrupt()
+        subtitles = self.SRTParser().parse_srt_content(srt_content, allow_overlaps=True)
+        has_overlaps = SRTOverlapHandler.detect_overlaps(subtitles)
+        active_mode, switched = SRTOverlapHandler.handle_smart_natural_fallback(
+            timing_mode, has_overlaps, "Audio8 TTS SRT"
+        )
+
+        audio_segments = []
+        adjustments = []
+        for index, subtitle in enumerate(subtitles):
+            self._check_interrupt(index, len(subtitles))
+            text = (subtitle.text or "").strip()
+            if text:
+                records = self.processor.process_text(
+                    text=text,
+                    voice_mapping=voice_mapping,
+                    seed=seed + index,
+                    enable_chunking=False,
+                    enable_audio_cache=enable_audio_cache,
+                    apply_edit_postprocessing=True,
+                )
+                audio, _ = self.processor.combine_audio_segments(
+                    records,
+                    method="auto",
+                    silence_ms=0,
+                    original_text=text,
+                    return_info=True,
+                )
+                if audio.dim() == 1:
+                    audio = audio.unsqueeze(0)
+            else:
+                audio = torch.zeros(1, int(subtitle.duration * self.SAMPLE_RATE))
+
+            audio = audio.detach().cpu().float()
+            audio_segments.append(audio)
+            natural_duration = audio.shape[-1] / self.SAMPLE_RATE
+            target_duration = subtitle.duration
+            stretch_factor = (
+                target_duration / natural_duration if natural_duration else 1.0
+            )
+            adjustments.append(
+                {
+                    "index": index,
+                    "segment_index": index,
+                    "sequence": subtitle.sequence,
+                    "natural_duration": natural_duration,
+                    "target_start": subtitle.start_time,
+                    "target_end": subtitle.end_time,
+                    "target_duration": target_duration,
+                    "start_time": subtitle.start_time,
+                    "end_time": subtitle.end_time,
+                    "stretch_factor": stretch_factor,
+                    "needs_stretching": (abs(stretch_factor - 1.0) > 0.05),
+                    "stretch_type": (
+                        "compress"
+                        if stretch_factor < 1.0
+                        else "expand"
+                        if stretch_factor > 1.0
+                        else "none"
+                    ),
+                    "adjustment": natural_duration - target_duration,
+                    "adjusted_start": subtitle.start_time,
+                    "adjusted_end": subtitle.end_time,
+                    "adjusted_duration": natural_duration,
+                }
+            )
+
+        self._check_interrupt()
+        final_audio, replacement, stretch_method = self._assemble(
+            audio_segments, subtitles, active_mode, timing_params or {}
+        )
+        self._check_interrupt()
+        if replacement is not None:
+            adjustments = replacement
+
+        reporter = SRTReportGenerator()
+        timing_report = reporter.generate_timing_report(
+            subtitles,
+            adjustments,
+            active_mode,
+            has_overlaps,
+            switched,
+            timing_mode if switched else None,
+            stretch_method,
+        )
+        adjusted_srt = reporter.generate_adjusted_srt_string(
+            subtitles, adjustments, active_mode
+        )
+
+        final_audio = final_audio.detach().cpu().float()
+        if final_audio.dim() == 1:
+            final_audio = final_audio.unsqueeze(0).unsqueeze(0)
+        elif final_audio.dim() == 2:
+            final_audio = final_audio.unsqueeze(0)
+        duration = final_audio.shape[-1] / self.SAMPLE_RATE
+        generation_info = (
+            f"Generated {duration:.1f}s Audio8 TTS SRT audio from "
+            f"{len(subtitles)} subtitles using {active_mode} mode"
+        )
+        return (
+            {"waveform": final_audio, "sample_rate": self.SAMPLE_RATE},
+            generation_info,
+            timing_report,
+            adjusted_srt,
+        )
+
+    def _assemble(self, audio_segments, subtitles, mode, params):
+        if mode == "stretch_to_fit":
+            from engines.chatterbox.audio_timing import TimedAudioAssembler
+
+            assembler = TimedAudioAssembler(self.SAMPLE_RATE)
+            audio, method = assembler.assemble_timed_audio(
+                audio_segments,
+                [(subtitle.start_time, subtitle.end_time) for subtitle in subtitles],
+                fade_duration=params.get("fade_for_StretchToFit", 0.01),
+            )
+            return audio, None, method
+
+        assembler = AudioAssemblyEngine(self.SAMPLE_RATE)
+        if mode == "pad_with_silence":
+            audio = assembler.assemble_with_overlaps(
+                audio_segments, subtitles, torch.device("cpu")
+            )
+            return audio, None, None
+
+        timing = TimingEngine(self.SAMPLE_RATE)
+        if mode == "concatenate":
+            replacements = timing.calculate_concatenation_adjustments(
+                audio_segments, subtitles
+            )
+            audio = assembler.assemble_concatenation(
+                audio_segments,
+                params.get("fade_for_StretchToFit", 0.01),
+            )
+            return audio, replacements, None
+
+        if mode != "smart_natural":
+            raise ValueError(f"Unsupported Audio8 SRT timing mode: {mode}")
+        replacements, processed = timing.calculate_smart_timing_adjustments(
+            audio_segments,
+            subtitles,
+            params.get("timing_tolerance", 2.0),
+            params.get("max_stretch_ratio", 1.0),
+            params.get("min_stretch_ratio", 0.5),
+            torch.device("cpu"),
+        )
+        audio = assembler.assemble_smart_natural(
+            audio_segments,
+            processed,
+            replacements,
+            subtitles,
+            torch.device("cpu"),
+        )
+        return audio, replacements, None

nodes/engines/audio8_tts_engine_node.py

@@ -0,0 +1,230 @@
+"""Audio8 TTS Preview engine configuration node."""
+
+import importlib.util
+import os
+import sys
+
+current_dir = os.path.dirname(__file__)
+nodes_dir = os.path.dirname(current_dir)
+project_root = os.path.dirname(nodes_dir)
+if project_root not in sys.path:
+    sys.path.insert(0, project_root)
+
+base_node_path = os.path.join(nodes_dir, "base", "base_node.py")
+base_spec = importlib.util.spec_from_file_location("base_node_module", base_node_path)
+base_module = importlib.util.module_from_spec(base_spec)
+sys.modules["base_node_module"] = base_module
+base_spec.loader.exec_module(base_module)
+BaseTTSNode = base_module.BaseTTSNode
+
+
+class Audio8TTSEngineNode(BaseTTSNode):
+    """Configure the official Audio8 TTS Preview checkpoint."""
+
+    DEFAULT_MODEL = "Audio8-TTS-Preview-0.6b"
+
+    @classmethod
+    def NAME(cls):
+        return "⚙️ Audio8 TTS Engine"
+
+    @classmethod
+    def _get_model_options(cls):
+        try:
+            from engines.audio8_tts.downloader import Audio8TTSDownloader
+
+            options = Audio8TTSDownloader().get_available_models()
+            return options or [cls.DEFAULT_MODEL]
+        except Exception:
+            return [cls.DEFAULT_MODEL]
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "model_variant": (
+                    cls._get_model_options(),
+                    {
+                        "default": cls.DEFAULT_MODEL,
+                        "tooltip": (
+                            "Official 0.6B Audio8 TTS Preview checkpoint. "
+                            "The canonical option downloads from "
+                            "Audio8/Audio8-TTS-Preview-0.6b; local: options are "
+                            "detected model folders.\n\n"
+                            "Preview scope: multilingual speech and zero-shot "
+                            "voice cloning, with 11 recommended languages. The "
+                            "model infers language from text and has no language "
+                            "selection control.\n\n"
+                            "Voice cloning requires reference audio and its exact "
+                            "matching transcript. Generation without a reference "
+                            "is supported. Audio8 has no instruction-conditioned "
+                            "voice-design mode. The suite runs it in the existing "
+                            "shared Transformers 4 runtime for correct cloning. "
+                            "Code and weights: Apache License 2.0."
+                        ),
+                    },
+                ),
+                "device": (
+                    ["auto", "cuda", "cpu"],
+                    {
+                        "default": "auto",
+                        "tooltip": (
+                            "Execution device. CUDA is recommended. CPU inference "
+                            "is supported but slow and runs in float32."
+                        ),
+                    },
+                ),
+                "temperature": (
+                    "FLOAT",
+                    {
+                        "default": 0.8,
+                        "min": 0.1,
+                        "max": 2.0,
+                        "step": 0.05,
+                        "tooltip": (
+                            "Official sampling temperature. Used only in Sampling "
+                            "mode; lower values are more conservative."
+                        ),
+                    },
+                ),
+                "top_p": (
+                    "FLOAT",
+                    {
+                        "default": 0.95,
+                        "min": 0.01,
+                        "max": 1.0,
+                        "step": 0.01,
+                        "tooltip": (
+                            "Official nucleus-sampling cutoff. Used only in "
+                            "Sampling mode."
+                        ),
+                    },
+                ),
+                "top_k": (
+                    "INT",
+                    {
+                        "default": 50,
+                        "min": 1,
+                        "max": 100,
+                        "step": 1,
+                        "tooltip": (
+                            "Official top-k sampling limit. Used only in Sampling mode."
+                        ),
+                    },
+                ),
+                "max_new_tokens": (
+                    "INT",
+                    {
+                        "default": 1024,
+                        "min": 64,
+                        "max": 2048,
+                        "step": 64,
+                        "tooltip": (
+                            "Maximum acoustic frames for the first generation "
+                            "attempt. Audio8 emits about 21.5 frames per second. "
+                            "Larger budgets take more time and context memory."
+                        ),
+                    },
+                ),
+            },
+            "optional": {
+                "retry_max_new_tokens": (
+                    "INT",
+                    {
+                        "default": 2048,
+                        "min": 64,
+                        "max": 2048,
+                        "step": 8,
+                        "tooltip": (
+                            "Second-attempt frame budget when the first attempt "
+                            "does not emit EOS. Must be at least max_new_tokens. "
+                            "Set it equal to max_new_tokens to disable the larger "
+                            "retry."
+                        ),
+                    },
+                ),
+                "dtype": (
+                    ["auto", "bfloat16", "float16", "float32"],
+                    {
+                        "default": "auto",
+                        "tooltip": (
+                            "Model precision. Auto prefers bfloat16 on Ampere-or-"
+                            "newer CUDA GPUs, uses float16 on older CUDA GPUs, and "
+                            "float32 on CPU. CPU coerces reduced-precision choices "
+                            "to float32."
+                        ),
+                    },
+                ),
+                "sampling_mode": (
+                    ["Sampling", "Greedy"],
+                    {
+                        "default": "Sampling",
+                        "tooltip": (
+                            "Official decoding mode. Sampling uses temperature, "
+                            "top_p, and top_k. Greedy is deterministic for a fixed "
+                            "prompt and ignores those sampling controls."
+                        ),
+                    },
+                ),
+            },
+        }
+
+    RETURN_TYPES = ("TTS_ENGINE",)
+    RETURN_NAMES = ("TTS_engine",)
+    FUNCTION = "create_engine_config"
+    CATEGORY = "TTS Audio Suite/⚙️ Engines"
+
+    def create_engine_config(
+        self,
+        model_variant: str,
+        device: str,
+        temperature: float,
+        top_p: float,
+        top_k: int,
+        max_new_tokens: int,
+        retry_max_new_tokens: int = 2048,
+        dtype: str = "auto",
+        sampling_mode: str = "Sampling",
+    ) -> tuple:
+        if retry_max_new_tokens < max_new_tokens:
+            raise ValueError(
+                "Audio8 retry_max_new_tokens must be at least max_new_tokens"
+            )
+
+        do_sample = sampling_mode == "Sampling"
+        config = {
+            "engine_type": "audio8_tts",
+            "model_variant": model_variant,
+            "device": device,
+            "dtype": dtype,
+            "max_new_tokens": int(max_new_tokens),
+            "retry_max_new_tokens": int(retry_max_new_tokens),
+            "temperature": float(temperature),
+            "top_p": float(top_p),
+            "top_k": int(top_k),
+            "do_sample": do_sample,
+        }
+
+        print(f"⚙️ Audio8 TTS Preview: {model_variant} on {device} ({dtype})")
+        print(
+            "   Settings: "
+            f"mode={sampling_mode}, temperature={temperature}, top_p={top_p}, "
+            f"top_k={top_k}, max_new_tokens={max_new_tokens}, "
+            f"retry_max_new_tokens={retry_max_new_tokens}"
+        )
+        print(
+            "   Voice cloning requires reference audio plus its exact matching "
+            "transcript; no-reference generation is supported."
+        )
+        print(
+            "   Runtime: shared Transformers 4 profile (required for correct cloning)"
+        )
+        print("   Preview checkpoint | Apache License 2.0")
+
+        return (
+            {
+                "engine_type": "audio8_tts",
+                "config": config,
+                "adapter_class": "Audio8TTSEngineAdapter",
+                "capabilities": ["tts"],
+            },
+        )

nodes/shared/character_voices_node.py

@@ -69,10 +69,11 @@ PRIORITY SYSTEM - When both .txt and .reference.txt exist:
                     "default": "",
                     "tooltip": """Create reference text on-the-fly for connected audio input.
 
-ENGINE REQUIREMENTS:
-• F5-TTS: REQUIRES reference text (must match spoken audio exactly)
-• Higgs Audio 2: Optional but uses reference text if provided
-• ChatterBox/VibeVoice/IndexTTS: Don't use reference text
+ENGINE REQUIREMENTS:
+• Audio8 TTS: REQUIRES the exact spoken transcript for voice cloning
+• F5-TTS: REQUIRES reference text (must match spoken audio exactly)
+• Higgs Audio 2: Optional but uses reference text if provided
+• ChatterBox/VibeVoice/IndexTTS: Don't use reference text
 
 Selecting a library voice loads its transcription here automatically. Edits are temporary workflow overrides and never modify the source .txt file."""
                 }),

nodes/step_audio_editx_special/step_audio_editx_audio_editor_node.py

@@ -208,13 +208,20 @@ class StepAudioEditXAudioEditorNode:
             self._cached_settings = None
 
         # Check if engine was deleted (MUST be after settings change check, before early return)
-        engine_was_deleted = (self._engine is not None and
-                             hasattr(self._engine, '_tts_engine') and
-                             self._engine._tts_engine is None)
-
-        if engine_was_deleted:
-            print("⚠️ Step Audio EditX engine was deleted, reloading...")
-            self._engine = None
+        engine_was_deleted = self._engine is not None and (
+            (
+                hasattr(self._engine, '_tts_engine')
+                and self._engine._tts_engine is None
+            )
+            or (
+                hasattr(self._engine, '_initialized')
+                and not self._engine._initialized
+            )
+        )
+
+        if engine_was_deleted:
+            print("⚠️ Step Audio EditX engine was unloaded, reloading...")
+            self._engine = None
         elif self._engine is not None:
             # Using cached engine - ensure it's on the correct device
             from utils.device import resolve_torch_device

nodes/unified/tts_srt_node.py

@@ -84,7 +84,7 @@ Hello! This is unified SRT TTS with character switching.
                 }),
                 "narrator_voice": (reference_files, {
                     "default": "none",
-                    "tooltip": "Fallback narrator voice from voice folders. Used when opt_narrator is not connected. Select 'none' for engines that support direct TTS without voice cloning, such as MOSS."
+                    "tooltip": "Fallback narrator voice from voice folders. Used when opt_narrator is not connected. Select 'none' for engines that support direct TTS without voice cloning, such as MOSS or Audio8."
                 }),
                 "seed": ("INT", {
                     "default": 1, "min": 0, "max": 2**32 - 1,
@@ -223,6 +223,12 @@ Hello! This is unified SRT TTS with character switching.
                 stable_params['optimize'] = config.get('optimize', False)
                 stable_params['max_generate_length'] = config.get('max_generate_length', 500)
 
+            if engine_type == "audio8_tts":
+                stable_params['model_variant'] = config.get(
+                    'model_variant', 'Audio8-TTS-Preview-0.6b'
+                )
+                stable_params['dtype'] = config.get('dtype', 'auto')
+
             if engine_type == "dramabox":
                 stable_params['model_name'] = config.get('model_name', 'DramaBox')
                 stable_params['precision'] = config.get('precision', 'auto')
@@ -578,6 +584,40 @@ Hello! This is unified SRT TTS with character switching.
                 }
                 return engine_instance
 
+            elif engine_type == "audio8_tts":
+                processor_path = os.path.join(
+                    nodes_dir, "audio8_tts", "audio8_tts_srt_processor.py"
+                )
+                processor_spec = importlib.util.spec_from_file_location(
+                    "audio8_tts_srt_processor_module", processor_path
+                )
+                processor_module = importlib.util.module_from_spec(processor_spec)
+                processor_spec.loader.exec_module(processor_module)
+                Audio8TTSSRTProcessor = processor_module.Audio8TTSSRTProcessor
+
+                class Audio8TTSSRTWrapper:
+                    def __init__(self, cfg):
+                        self.config = cfg.copy()
+                        self.processor = Audio8TTSSRTProcessor(self, self.config)
+
+                    def update_config(self, new_config):
+                        self.config = new_config.copy()
+                        self.processor.update_config(new_config)
+
+                    def check_interrupt(self):
+                        if model_management.interrupt_processing:
+                            raise InterruptedError(
+                                "Audio8 TTS SRT processing interrupted by user"
+                            )
+
+                engine_instance = Audio8TTSSRTWrapper(config)
+                import time
+                self._cached_engine_instances[cache_key] = {
+                    'instance': engine_instance,
+                    'timestamp': time.time()
+                }
+                return engine_instance
+
             elif engine_type == "dramabox":
                 processor_path = os.path.join(
                     nodes_dir, "dramabox", "dramabox_srt_processor.py"
@@ -987,7 +1027,13 @@ Hello! This is unified SRT TTS with character switching.
                 raise ValueError(f"Unknown engine type: {engine_type}")
                 
         except Exception as e:
-            if isinstance(e, InterruptedError):
+            if isinstance(
+                e,
+                (
+                    InterruptedError,
+                    model_management.InterruptProcessingException,
+                ),
+            ):
                 raise
             print(f"❌ Failed to create engine SRT node instance: {e}")
             return None
@@ -1053,7 +1099,7 @@ Hello! This is unified SRT TTS with character switching.
                     print(f"📺 TTS SRT: Using direct audio input ({character_name})")
                     print(
                         "⚠️ TTS SRT: Direct audio input has no reference text - "
-                        "F5-TTS and OmniVoice cloning will fail"
+                        "Audio8 TTS, F5-TTS, and OmniVoice cloning will fail"
                     )
                     return None, audio_tensor, reference_text, character_name
             
@@ -1080,7 +1126,13 @@ Hello! This is unified SRT TTS with character switching.
             return None, None, "", "narrator"
             
         except Exception as e:
-            if isinstance(e, InterruptedError):
+            if isinstance(
+                e,
+                (
+                    InterruptedError,
+                    model_management.InterruptProcessingException,
+                ),
+            ):
                 raise
             print(f"❌ Voice reference error: {e}")
             return None, None, "", "narrator"
@@ -1175,6 +1227,13 @@ Hello! This is unified SRT TTS with character switching.
                     "OmniVoice voice cloning requires reference text. "
                     "Do not connect raw audio directly. Use Character Voices node or a narrator voice with a matching .reference.txt file."
                 )
+            if engine_type == "audio8_tts" and (audio_tensor is not None or audio_path) and not reference_text.strip():
+                raise ValueError(
+                    "Audio8 TTS voice cloning requires reference text. "
+                    "Do not connect raw audio directly. Use Character Voices with "
+                    "the exact transcript, or a narrator voice with a matching "
+                    ".reference.txt file."
+                )
             
             # Create proper engine SRT node instance to preserve ALL functionality
             engine_instance = self._create_proper_engine_node_instance(TTS_engine)
@@ -1375,6 +1434,30 @@ Hello! This is unified SRT TTS with character switching.
                     enable_audio_cache=enable_audio_cache
                 )
 
+            elif engine_type == "audio8_tts":
+                timing_params = {
+                    'fade_for_StretchToFit': fade_for_StretchToFit,
+                    'max_stretch_ratio': max_stretch_ratio,
+                    'min_stretch_ratio': min_stretch_ratio,
+                    'timing_tolerance': timing_tolerance,
+                }
+                voice_mapping = {}
+                if audio_tensor is not None or audio_path:
+                    voice_mapping['narrator'] = {
+                        'audio': audio_tensor,
+                        'audio_path': audio_path,
+                        'reference_text': reference_text or '',
+                        'character_name': character_name or 'narrator',
+                    }
+                result = engine_instance.processor.process_srt_content(
+                    srt_content=srt_content,
+                    voice_mapping=voice_mapping,
+                    seed=seed,
+                    timing_mode=timing_mode,
+                    timing_params=timing_params,
+                    enable_audio_cache=enable_audio_cache,
+                )
+
             elif engine_type == "dramabox":
                 timing_params = {
                     'fade_for_StretchToFit': fade_for_StretchToFit,
@@ -1688,9 +1771,16 @@ Hello! This is unified SRT TTS with character switching.
                 "is a voice-design model and cannot be used with TTS SRT" in msg
                 or "Pause tags are not compatible with force_speaker_kv" in msg
                 or "MOSS-TTSD Native Multi-Speaker Dialogue does not support this SRT input" in msg
+                or "Audio8 TTS voice cloning requires reference text" in msg
             ):
                 raise
-            if isinstance(e, InterruptedError):
+            if isinstance(
+                e,
+                (
+                    InterruptedError,
+                    model_management.InterruptProcessingException,
+                ),
+            ):
                 raise
             error_msg = f"❌ TTS SRT generation failed: {e}"
             print(error_msg)

nodes/unified/tts_text_node.py

@@ -89,7 +89,7 @@ Back to the main narrator voice for the conclusion.""",
                 }),
                 "narrator_voice": (reference_files, {
                     "default": "none",
-                    "tooltip": "Fallback narrator voice from voice folders. Used when opt_narrator is not connected. Select 'none' for engines that support direct TTS without voice cloning, such as MOSS."
+                    "tooltip": "Fallback narrator voice from voice folders. Used when opt_narrator is not connected. Select 'none' for engines that support direct TTS without voice cloning, such as MOSS or Audio8."
                 }),
                 "seed": ("INT", {
                     "default": 1, "min": 0, "max": 2**32 - 1,
@@ -218,6 +218,12 @@ Back to the main narrator voice for the conclusion.""",
                 stable_params['optimize'] = config.get('optimize', False)
                 stable_params['max_generate_length'] = config.get('max_generate_length', 500)
 
+            if engine_type == "audio8_tts":
+                stable_params['model_variant'] = config.get(
+                    'model_variant', 'Audio8-TTS-Preview-0.6b'
+                )
+                stable_params['dtype'] = config.get('dtype', 'auto')
+
             if engine_type == "dramabox":
                 stable_params['model_name'] = config.get('model_name', 'DramaBox')
                 stable_params['precision'] = config.get('precision', 'auto')
@@ -445,7 +451,13 @@ Back to the main narrator voice for the conclusion.""",
                                         voice_mapping[char] = {"waveform": waveform, "sample_rate": sample_rate}
                                         print(f"🎭 VibeVoice: Using character-specific voice for '{char}'")
                                     except Exception as e:
-                                        if isinstance(e, InterruptedError):
+                                        if isinstance(
+                                            e,
+                                            (
+                                                InterruptedError,
+                                                model_management.InterruptProcessingException,
+                                            ),
+                                        ):
                                             raise
                                         print(f"⚠️ Failed to load character audio for '{char}': {e}")
                                         voice_mapping[char] = char_audio  # Fallback to main voice
@@ -580,6 +592,38 @@ Back to the main narrator voice for the conclusion.""",
 
                 return engine_instance
 
+            elif engine_type == "audio8_tts":
+                from engines.adapters.audio8_tts_adapter import Audio8TTSEngineAdapter
+
+                processor_path = os.path.join(
+                    nodes_dir, "audio8_tts", "audio8_tts_processor.py"
+                )
+                processor_spec = importlib.util.spec_from_file_location(
+                    "audio8_tts_processor_module", processor_path
+                )
+                processor_module = importlib.util.module_from_spec(processor_spec)
+                processor_spec.loader.exec_module(processor_module)
+                Audio8TTSProcessor = processor_module.Audio8TTSProcessor
+
+                class Audio8TTSWrapper:
+                    def __init__(self, cfg):
+                        self.config = cfg.copy()
+                        self.adapter = Audio8TTSEngineAdapter(self.config)
+                        self.processor = Audio8TTSProcessor(self.adapter, self.config)
+
+                    def update_config(self, new_config):
+                        self.config = new_config.copy()
+                        self.adapter.update_config(new_config)
+                        self.processor.update_config(new_config)
+
+                engine_instance = Audio8TTSWrapper(config)
+                import time
+                self._cached_engine_instances[cache_key] = {
+                    'instance': engine_instance,
+                    'timestamp': time.time()
+                }
+                return engine_instance
+
             elif engine_type == "dramabox":
                 from engines.adapters.dramabox_adapter import DramaBoxEngineAdapter
                 processor_path = os.path.join(nodes_dir, "dramabox", "dramabox_processor.py")
@@ -837,7 +881,13 @@ Back to the main narrator voice for the conclusion.""",
                 raise ValueError(f"Unknown engine type: {engine_type}")
                 
         except Exception as e:
-            if isinstance(e, InterruptedError):
+            if isinstance(
+                e,
+                (
+                    InterruptedError,
+                    model_management.InterruptProcessingException,
+                ),
+            ):
                 raise
             if "MOSS LoRA/base model mismatch" in str(e):
                 raise
@@ -906,7 +956,8 @@ Back to the main narrator voice for the conclusion.""",
                     print(f"🎤 TTS Text: Using direct audio input ({character_name})")
                     print(
                         "⚠️ TTS Text: Direct audio input has no reference text - "
-                        "F5-TTS and OmniVoice cloning will fail, Qwen3-TTS will use x_vector_only mode (lower quality)"
+                        "Audio8 TTS, F5-TTS, and OmniVoice cloning will fail; "
+                        "Qwen3-TTS will use x_vector_only mode (lower quality)"
                     )
                     return None, audio_tensor, reference_text, character_name
             
@@ -952,7 +1003,13 @@ Back to the main narrator voice for the conclusion.""",
             return None, None, "", "narrator"
             
         except Exception as e:
-            if isinstance(e, InterruptedError):
+            if isinstance(
+                e,
+                (
+                    InterruptedError,
+                    model_management.InterruptProcessingException,
+                ),
+            ):
                 raise
             print(f"❌ Voice reference error: {e}")
             return None, None, "", "narrator"
@@ -1059,6 +1116,13 @@ Back to the main narrator voice for the conclusion.""",
                     "OmniVoice voice cloning requires reference text. "
                     "Do not connect raw audio directly. Use Character Voices node or a narrator voice with a matching .reference.txt file."
                 )
+            if engine_type == "audio8_tts" and (audio_tensor is not None or audio_path) and not reference_text.strip():
+                raise ValueError(
+                    "Audio8 TTS voice cloning requires reference text. "
+                    "Do not connect raw audio directly. Use Character Voices with "
+                    "the exact transcript, or a narrator voice with a matching "
+                    ".reference.txt file."
+                )
             
             # Create proper engine node instance to preserve ALL functionality
             engine_instance = self._create_proper_engine_node_instance(TTS_engine)
@@ -1520,6 +1584,59 @@ Back to the main narrator voice for the conclusion.""",
                 formatted_audio = AudioProcessingUtils.format_for_comfyui(combined_audio, 48000)
                 result = (formatted_audio, generation_info)
 
+            elif engine_type == "audio8_tts":
+                import re
+                from utils.audio.chunk_timing import ChunkTimingHelper
+
+                voice_mapping = {}
+                if audio_tensor is not None or audio_path:
+                    voice_mapping['narrator'] = {
+                        'audio': audio_tensor,
+                        'audio_path': audio_path,
+                        'reference_text': reference_text or '',
+                        'character_name': character_name or 'narrator',
+                    }
+
+                segment_records = engine_instance.processor.process_text(
+                    text=text,
+                    voice_mapping=voice_mapping,
+                    seed=seed,
+                    enable_chunking=enable_chunking,
+                    max_chars_per_chunk=max_chars_per_chunk,
+                    chunk_combination_method=chunk_combination_method,
+                    silence_between_chunks_ms=silence_between_chunks_ms,
+                    enable_audio_cache=enable_audio_cache,
+                )
+                combined_audio, chunk_info = engine_instance.processor.combine_audio_segments(
+                    segments=segment_records,
+                    method=chunk_combination_method,
+                    silence_ms=silence_between_chunks_ms,
+                    original_text=text,
+                    return_info=True,
+                )
+
+                total_duration = (
+                    combined_audio.shape[-1] / 44100.0
+                    if combined_audio.numel()
+                    else 0.0
+                )
+                clean_text = re.sub(r'\[.*?\]', '', text)
+                base_info = (
+                    f"Generated {total_duration:.1f}s audio from {len(clean_text)} characters "
+                    f"(Audio8 TTS, narrator: {char_display})"
+                )
+                base_info += (
+                    "\n🎭 Reference-free speech, zero-shot cloning, character switching, "
+                    "and pause tags enabled"
+                )
+                generation_info = ChunkTimingHelper.enhance_generation_info(
+                    f"✅ {base_info}", chunk_info
+                )
+                result = (
+                    AudioProcessingUtils.format_for_comfyui(combined_audio, 44100),
+                    generation_info,
+                )
+
             elif engine_type == "dramabox":
                 import re
                 from utils.audio.chunk_timing import ChunkTimingHelper
@@ -1900,7 +2017,13 @@ Back to the main narrator voice for the conclusion.""",
                                 }
                                 print(f"🎭 Qwen3-TTS: Using character-specific voice for '{character}' (ICL mode)")
                             except Exception as e:
-                                if isinstance(e, InterruptedError):
+                                if isinstance(
+                                    e,
+                                    (
+                                        InterruptedError,
+                                        model_management.InterruptProcessingException,
+                                    ),
+                                ):
                                     raise
                                 print(f"⚠️ Failed to load character audio for '{character}': {e}")
                                 # Fallback to narrator voice if available
@@ -1945,7 +2068,13 @@ Back to the main narrator voice for the conclusion.""",
                                 print(f"⚠️⚠️ Qwen3-TTS: Character '{character}' has audio but NO reference text")
                                 print(f"⚠️⚠️ Using x_vector_only mode (speaker embedding only) - LOWER QUALITY than ICL mode")
                             except Exception as e:
-                                if isinstance(e, InterruptedError):
+                                if isinstance(
+                                    e,
+                                    (
+                                        InterruptedError,
+                                        model_management.InterruptProcessingException,
+                                    ),
+                                ):
                                     raise
                                 print(f"⚠️ Failed to load character audio for '{character}': {e}")
                                 # Fallback to narrator voice if available
@@ -2082,9 +2211,17 @@ Back to the main narrator voice for the conclusion.""",
                 raise
             if "MOSS LoRA/base model mismatch" in str(e):
                 raise
+            if "Audio8 TTS voice cloning requires reference text" in str(e):
+                raise
             if engine_type == "index_tts":
                 raise
-            if isinstance(e, InterruptedError):
+            if isinstance(
+                e,
+                (
+                    InterruptedError,
+                    model_management.InterruptProcessingException,
+                ),
+            ):
                 raise
             error_msg = f"❌ TTS Text generation failed: {e}"
             print(error_msg)

tests/unit/test_runtime_launcher.py

@@ -1,4 +1,7 @@
+import json
+import os
 from pathlib import Path
+import subprocess
 import sys
 
 import pytest
@@ -41,3 +44,40 @@ def test_build_env_preserves_valid_pythonhashseed(monkeypatch, value):
     env = IsolatedRuntimeLauncher().build_env(profile)
 
     assert env["PYTHONHASHSEED"] == value
+
+
+@pytest.mark.unit
+def test_audio8_worker_prioritizes_project_root_over_shadowing_utils(tmp_path):
+    shadow_dir = tmp_path / "step_audio_editx_impl"
+    shadow_dir.mkdir()
+    (shadow_dir / "utils.py").write_text(
+        'raise RuntimeError("shadow utils imported")\n',
+        encoding="utf-8",
+    )
+
+    worker_path = REPO_ROOT / "utils" / "runtimes" / "workers" / "audio8_tts_worker.py"
+    env = os.environ.copy()
+    env["PYTHONPATH"] = os.pathsep.join((str(shadow_dir), str(REPO_ROOT)))
+    completed = subprocess.run(
+        [sys.executable, str(worker_path)],
+        input=json.dumps(
+            {
+                "action": "ping",
+                "request_id": "audio8-path-priority-test",
+            }
+        )
+        + "\n",
+        capture_output=True,
+        text=True,
+        encoding="utf-8",
+        errors="replace",
+        env=env,
+        cwd=str(REPO_ROOT),
+        timeout=30,
+        check=False,
+    )
+
+    assert completed.returncode == 0, completed.stderr
+    response = json.loads(completed.stdout.strip())
+    assert response["ok"] is True
+    assert response["result"] == {"pong": True}

utils/audio/cache.py

@@ -577,6 +577,30 @@ class DotsTTSCacheKeyGenerator(CacheKeyGenerator):
         return hashlib.md5(cache_string.encode()).hexdigest()
 
 
+class Audio8TTSCacheKeyGenerator(CacheKeyGenerator):
+    """Cache key generator for official Audio8 TTS inference."""
+
+    def generate_cache_key(self, **params) -> str:
+        cache_data = {
+            'text': params.get('text', ''),
+            'audio_component': params.get('audio_component', ''),
+            'reference_text': params.get('reference_text', ''),
+            'model_variant': params.get('model_variant', 'Audio8-TTS-Preview-0.6b'),
+            'max_new_tokens': int(params.get('max_new_tokens', 1024)),
+            'retry_max_new_tokens': int(params.get('retry_max_new_tokens', 2000)),
+            'temperature': round(float(params.get('temperature', 0.8)), 4),
+            'top_p': round(float(params.get('top_p', 0.95)), 4),
+            'top_k': int(params.get('top_k', 50)),
+            'do_sample': bool(params.get('do_sample', True)),
+            'dtype': params.get('dtype', 'auto'),
+            'device': params.get('device', 'auto'),
+            'seed': int(params.get('seed', 42)),
+            'character': params.get('character', 'narrator'),
+            'engine': 'audio8_tts',
+        }
+        return hashlib.md5(str(sorted(cache_data.items())).encode()).hexdigest()
+
+
 class DramaBoxCacheKeyGenerator(CacheKeyGenerator):
     """Cache key generator for official DramaBox inference."""
 
@@ -710,6 +734,7 @@ class AudioCache:
             'cosyvoice': CosyVoiceCacheKeyGenerator(),
             'qwen3_tts': Qwen3TTSCacheKeyGenerator(),
             'dots_tts': DotsTTSCacheKeyGenerator(),
+            'audio8_tts': Audio8TTSCacheKeyGenerator(),
             'dramabox': DramaBoxCacheKeyGenerator(),
             'fish_audio_s2': FishAudioS2CacheKeyGenerator(),
             'omnivoice': OmniVoiceCacheKeyGenerator(),
@@ -792,7 +817,9 @@ class AudioCache:
             sample_rate = 24000
         elif engine_type in ('index_tts', 'cosyvoice'):
             sample_rate = 22050
-        else:
+        elif engine_type == 'audio8_tts':
+            sample_rate = 44100
+        else:
             sample_rate = 44100
         return num_samples / sample_rate
     

utils/models/engine_registry.py

@@ -157,6 +157,15 @@ ENGINE_REGISTRY: Dict[str, EngineCapabilities] = {
         fallback_languages=["English"],
     ),
 
+    "audio8_tts": EngineCapabilities(
+        supports_voice_conversion=False,
+        multilingual_model_switching=False,
+        can_corrupt_on_reload=False,
+        fallback_languages=["English"],
+        supports_runtime_isolation=True,
+        default_runtime_profile="vibevoice_transformers4_shared",
+    ),
+
     "dramabox": EngineCapabilities(
         supports_voice_conversion=False,
         multilingual_model_switching=False,

utils/models/unified_model_interface.py

@@ -11,9 +11,14 @@ from typing import Any, Dict, Optional, Callable, Union
 from pathlib import Path
 
 from utils.models.comfyui_model_wrapper import tts_model_manager, ModelInfo
-from utils.models.factory_config import ModelLoadConfig, runtime_uses_isolation
+from utils.models.factory_config import (
+    ModelLoadConfig,
+    RUNTIME_MODE_SHARED,
+    runtime_uses_isolation,
+)
 from utils.models.engine_registry import get_default_runtime_profile
 from utils.runtimes import (
+    build_audio8_tts_isolated_proxy,
     build_fish_audio_s2_proxy,
     build_higgs_audio_isolated_proxy,
     build_qwen3_asr_isolated_proxy,
@@ -124,6 +129,10 @@ class UnifiedModelInterface:
         # Check PyTorch consistency on first model load
         self._check_pytorch_consistency()
 
+        if config.engine_name == "audio8_tts":
+            config.runtime_mode = RUNTIME_MODE_SHARED
+            config.runtime_profile = "vibevoice_transformers4_shared"
+
         if runtime_uses_isolation(config.runtime_mode):
             return self._load_isolated_model(config, force_reload=force_reload)
 
@@ -219,6 +228,13 @@ class UnifiedModelInterface:
 
         cached = self._isolated_model_cache.get(cache_key)
         if cached is not None:
+            ensure_registration = getattr(
+                cached,
+                "_ensure_comfy_model_registration",
+                None,
+            )
+            if callable(ensure_registration):
+                ensure_registration()
             return cached
 
         profile_name = config.runtime_profile or get_default_runtime_profile(config.engine_name or "")
@@ -244,6 +260,11 @@ class UnifiedModelInterface:
             self._isolated_model_cache[cache_key] = proxy
             return proxy
 
+        if config.engine_name == "audio8_tts" and config.model_type == "tts":
+            proxy = build_audio8_tts_isolated_proxy(config)
+            self._isolated_model_cache[cache_key] = proxy
+            return proxy
+
         if config.engine_name == "qwen3_asr" and config.model_type in ("asr", "aligner"):
             proxy = build_qwen3_asr_isolated_proxy(config)
             self._isolated_model_cache[cache_key] = proxy
@@ -1751,6 +1772,19 @@ def register_dots_tts_factory():
     unified_model_interface.register_model_factory("dots_tts", "tts", dots_tts_factory)
 
 
+def register_audio8_tts_factory():
+    """Register Audio8 TTS through its required Transformers 4 runtime."""
+    def audio8_tts_factory(config: ModelLoadConfig):
+        del config
+        raise RuntimeError(
+            "Audio8 TTS reached its embedded factory unexpectedly; all "
+            "Audio8 loads must be routed through the shared Transformers 4 "
+            "runtime"
+        )
+
+    unified_model_interface.register_model_factory("audio8_tts", "tts", audio8_tts_factory)
+
+
 def register_dramabox_factory():
     """Register official DramaBox in the main Transformers 5 environment."""
     def dramabox_factory(config: ModelLoadConfig):
@@ -2018,6 +2052,7 @@ def initialize_all_factories():
     register_qwen3_tts_factory()
     register_fish_audio_s2_factory()
     register_dots_tts_factory()
+    register_audio8_tts_factory()
     register_dramabox_factory()
     register_omnivoice_factory()
     register_qwen3_asr_factory()

utils/runtimes/__init__.py

@@ -6,6 +6,7 @@ separate Python environments without changing the existing ComfyUI-side UI.
 """
 
 from .bootstrap import ensure_runtime, resolve_runtime_dir, resolve_runtime_python
+from .audio8_tts_proxy import Audio8TTSIsolatedProxy, build_audio8_tts_isolated_proxy
 from .higgs_audio_proxy import HiggsAudioIsolatedProxy, build_higgs_audio_isolated_proxy
 from .fish_audio_s2_proxy import FishAudioS2Proxy, build_fish_audio_s2_proxy
 from .launcher import IsolatedRuntimeLauncher
@@ -20,6 +21,8 @@ __all__ = [
     "ensure_runtime",
     "resolve_runtime_dir",
     "resolve_runtime_python",
+    "Audio8TTSIsolatedProxy",
+    "build_audio8_tts_isolated_proxy",
     "HiggsAudioIsolatedProxy",
     "IsolatedRuntimeLauncher",
     "RuntimeProfile",

utils/runtimes/audio8_tts_proxy.py

@@ -0,0 +1,546 @@
+"""Audio8 TTS proxy for the shared official Transformers 4 runtime."""
+
+from __future__ import annotations
+
+import os
+import sys
+import tempfile
+import threading
+import uuid
+import weakref
+from pathlib import Path
+from typing import Any, Dict, Optional
+
+import torch
+
+from utils.models.factory_config import ModelLoadConfig
+from .bootstrap import PROJECT_ROOT, ensure_runtime
+from .launcher import IsolatedRuntimeLauncher
+from .profiles import RuntimeProfile, get_runtime_profile
+from .protocol import RuntimeJobRequest
+from .session import JsonLineWorkerSession
+
+
+_INTERRUPT_EVENTS: set[threading.Event] = set()
+_INTERRUPT_EVENTS_LOCK = threading.Lock()
+_INTERRUPT_DISPATCH_LOCK = threading.Lock()
+
+
+def _install_comfy_interrupt_dispatcher() -> None:
+    """Forward Comfy's server interrupt edge to active Audio8 requests."""
+    server_module = sys.modules.get("server")
+    comfy_nodes = getattr(server_module, "nodes", None)
+    if comfy_nodes is None:
+        comfy_nodes = sys.modules.get("nodes")
+    if comfy_nodes is None:
+        return
+
+    with _INTERRUPT_DISPATCH_LOCK:
+        current_interrupt = getattr(comfy_nodes, "interrupt_processing", None)
+        if not callable(current_interrupt):
+            return
+        if getattr(current_interrupt, "_tts_audio8_dispatcher", False):
+            if (
+                getattr(current_interrupt, "_tts_audio8_events", None)
+                is _INTERRUPT_EVENTS
+            ):
+                return
+            current_interrupt = getattr(
+                current_interrupt,
+                "_tts_audio8_original_interrupt",
+                current_interrupt,
+            )
+
+        def dispatch_interrupt(value=True):
+            # TTS Audio Suite patch: preserve Comfy's interrupt edge for
+            # blocking Audio8 subprocess requests without replacing the
+            # process-wide callback separately for every generation.
+            result = current_interrupt(value)
+            if value:
+                with _INTERRUPT_EVENTS_LOCK:
+                    active_events = tuple(_INTERRUPT_EVENTS)
+                for active_event in active_events:
+                    active_event.set()
+            return result
+
+        dispatch_interrupt._tts_audio8_dispatcher = True
+        dispatch_interrupt._tts_audio8_original_interrupt = current_interrupt
+        dispatch_interrupt._tts_audio8_events = _INTERRUPT_EVENTS
+        comfy_nodes.interrupt_processing = dispatch_interrupt
+
+
+def _register_interrupt_event(interrupt_event: threading.Event) -> None:
+    _install_comfy_interrupt_dispatcher()
+    with _INTERRUPT_EVENTS_LOCK:
+        _INTERRUPT_EVENTS.add(interrupt_event)
+
+
+def _unregister_interrupt_event(interrupt_event: threading.Event) -> None:
+    with _INTERRUPT_EVENTS_LOCK:
+        _INTERRUPT_EVENTS.discard(interrupt_event)
+
+
+class Audio8TTSIsolatedProxy:
+    """Expose Audio8's engine API while inference runs under Transformers 4."""
+
+    SAMPLE_RATE = 44100
+
+    def __init__(self, config: ModelLoadConfig, profile: RuntimeProfile):
+        self.config = config
+        self.profile = profile
+        self.model_name = config.model_name or "Audio8-TTS-Preview-0.6b"
+        self.model_path = config.model_path
+        self.device = str(config.device)
+        self.current_device = self.device
+        self.dtype_name = str((config.additional_params or {}).get("dtype", "auto"))
+        self.dtype = self._resolve_dtype(self.dtype_name)
+        self.load_device = self.current_loaded_device()
+        self.offload_device = torch.device("cpu")
+        self.currently_used = True
+        self.model = self
+        self.engine = self
+        self.parent = None
+        self.model_options = {}
+        self.model_keys = set()
+        estimate_gib = (
+            3.5
+            if self.device.startswith("cpu") or self.dtype == torch.float32
+            else 1.75
+        )
+        self._estimated_memory_size = int(estimate_gib * 1024**3)
+        self._comfy_loaded_model = None
+        self._comfy_model_management = None
+        self._initialized = False
+
+        launcher = IsolatedRuntimeLauncher(runtime_root=str(PROJECT_ROOT))
+        python_path = ensure_runtime(profile)
+        worker_script = PROJECT_ROOT / "utils/runtimes/workers/audio8_tts_worker.py"
+        self._session = JsonLineWorkerSession(
+            python_path=str(python_path),
+            worker_script=str(worker_script),
+            env=launcher.build_env(profile),
+        )
+        try:
+            self._initialize_remote_engine()
+        except Exception:
+            self._session.close()
+            raise
+        self._register_with_comfy_model_management()
+
+    def _resolve_dtype(self, dtype_name: str) -> torch.dtype:
+        normalized = str(dtype_name or "auto").lower()
+        if not self.device.startswith("cuda"):
+            return torch.float32
+        if normalized == "float32":
+            return torch.float32
+        if normalized == "float16":
+            return torch.float16
+        if normalized == "bfloat16":
+            return torch.bfloat16
+        if torch.cuda.is_available():
+            major, _minor = torch.cuda.get_device_capability(torch.device(self.device))
+            return torch.bfloat16 if major >= 8 else torch.float16
+        return torch.float16
+
+    def _request(self, action: str, payload: Dict[str, Any]):
+        response = self._session.request(
+            RuntimeJobRequest(
+                engine_name="audio8_tts",
+                action=action,
+                model_name=self.model_name,
+                device=self.device,
+                runtime_profile=self.profile.name,
+                payload=payload,
+                request_id=str(uuid.uuid4()),
+            )
+        )
+        if not response.ok:
+            details = response.error or (
+                f"Audio8 TTS isolated action '{action}' failed"
+            )
+            if response.logs:
+                details = f"{details}\n" + "\n".join(response.logs)
+            raise RuntimeError(details)
+        return response
+
+    def _request_generation(self, payload: Dict[str, Any]):
+        """Run generation while the main process monitors Comfy interruption."""
+        try:
+            import comfy.model_management as model_management
+        except Exception:
+            return self._request("generate", payload)
+
+        checker = getattr(
+            model_management,
+            "throw_exception_if_processing_interrupted",
+            None,
+        )
+        if not callable(checker):
+            return self._request("generate", payload)
+
+        interrupt_event = threading.Event()
+        _register_interrupt_event(interrupt_event)
+        result: Dict[str, Any] = {}
+        request_thread = None
+        request_started = False
+
+        def run_request() -> None:
+            try:
+                result["response"] = self._request("generate", payload)
+            except BaseException as exc:
+                result["error"] = exc
+
+        try:
+            checker()
+            request_thread = threading.Thread(
+                target=run_request,
+                name="audio8-tts-worker-request",
+                daemon=True,
+            )
+            request_thread.start()
+            request_started = True
+
+            while request_thread.is_alive():
+                request_thread.join(timeout=0.05)
+                if interrupt_event.is_set():
+                    raise model_management.InterruptProcessingException()
+                checker()
+        except BaseException as exc:
+            if request_started:
+                self._initialized = False
+                try:
+                    self._session.terminate_now()
+                except Exception:
+                    pass
+                request_thread.join(timeout=5)
+            if isinstance(
+                exc,
+                model_management.InterruptProcessingException,
+            ):
+                interrupt_setter = getattr(
+                    model_management,
+                    "interrupt_current_processing",
+                    None,
+                )
+                if callable(interrupt_setter):
+                    interrupt_setter(False)
+            raise
+        finally:
+            _unregister_interrupt_event(interrupt_event)
+
+        if "error" in result:
+            raise result["error"]
+        return result["response"]
+
+    def _initialize_remote_engine(self) -> None:
+        self._request(
+            "initialize",
+            {
+                "model_path": self.model_path,
+                "dtype": self.dtype_name,
+            },
+        )
+        self._initialized = True
+        print(f"✅ Audio8 TTS isolated runtime ready ({self.profile.name})")
+
+    def _ensure_remote_engine(self) -> None:
+        process = getattr(self._session, "_process", None)
+        if not self._initialized or process is None or process.poll() is not None:
+            self._initialize_remote_engine()
+
+    @staticmethod
+    def _serialize_reference_audio(
+        reference_audio: Any,
+        bundle_dir: Path,
+    ) -> Optional[Dict[str, Any]]:
+        if reference_audio is None:
+            return None
+        if isinstance(reference_audio, (str, os.PathLike)):
+            return {
+                "kind": "audio_path",
+                "audio_path": os.fspath(reference_audio),
+            }
+
+        waveform = None
+        sample_rate = None
+        if isinstance(reference_audio, dict):
+            waveform = reference_audio.get(
+                "waveform",
+                reference_audio.get("array"),
+            )
+            sample_rate = reference_audio.get(
+                "sample_rate",
+                reference_audio.get("sampling_rate"),
+            )
+        elif isinstance(reference_audio, (tuple, list)) and len(reference_audio) == 2:
+            waveform, sample_rate = reference_audio
+
+        if waveform is None or sample_rate is None:
+            raise TypeError(
+                "Audio8 TTS isolated reference audio must be a path, "
+                "audio dict, or (waveform, sample_rate) pair"
+            )
+
+        tensor_path = bundle_dir / "reference_audio.pt"
+        waveform_tensor = torch.as_tensor(waveform).detach().float().cpu()
+        torch.save(
+            {
+                "waveform": waveform_tensor,
+                "sample_rate": int(sample_rate),
+            },
+            tensor_path,
+        )
+        return {
+            "kind": "tensor_path",
+            "tensor_path": str(tensor_path),
+        }
+
+    def generate(
+        self,
+        text: str,
+        reference_audio: Any = None,
+        reference_text: Optional[str] = None,
+        *,
+        max_new_tokens: int = 1024,
+        retry_max_new_tokens: int = 2000,
+        temperature: float = 0.8,
+        top_p: float = 0.95,
+        top_k: int = 50,
+        do_sample: bool = True,
+        seed: int = 42,
+    ) -> torch.Tensor:
+        self._ensure_remote_engine()
+        with tempfile.TemporaryDirectory(prefix="tts_audio8_iso_") as temp_dir:
+            bundle_dir = Path(temp_dir)
+            output_path = bundle_dir / "result.pt"
+            serialized_reference = self._serialize_reference_audio(
+                reference_audio,
+                bundle_dir,
+            )
+            self._request_generation(
+                {
+                    "text": text,
+                    "reference_audio": serialized_reference,
+                    "reference_text": reference_text,
+                    "max_new_tokens": int(max_new_tokens),
+                    "retry_max_new_tokens": int(retry_max_new_tokens),
+                    "temperature": float(temperature),
+                    "top_p": float(top_p),
+                    "top_k": int(top_k),
+                    "do_sample": bool(do_sample),
+                    "seed": int(seed),
+                    "output_path": str(output_path),
+                }
+            )
+            if not output_path.exists():
+                raise RuntimeError(
+                    "Audio8 TTS isolated worker returned no output payload"
+                )
+            result = torch.load(output_path, map_location="cpu")
+            audio = torch.as_tensor(result["audio"]).detach().float().cpu()
+            sample_rate = int(result["sample_rate"])
+            if sample_rate != self.SAMPLE_RATE:
+                raise RuntimeError(
+                    f"Audio8 TTS worker returned {sample_rate} Hz; "
+                    f"expected {self.SAMPLE_RATE} Hz"
+                )
+            if audio.ndim == 1:
+                audio = audio.unsqueeze(0)
+            if audio.ndim != 2 or audio.shape[0] != 1:
+                raise RuntimeError(
+                    "Audio8 TTS worker returned invalid audio shape "
+                    f"{tuple(audio.shape)}"
+                )
+            return audio.contiguous()
+
+    def _register_with_comfy_model_management(self) -> None:
+        self._ensure_comfy_model_registration()
+
+    def _ensure_comfy_model_registration(self) -> None:
+        try:
+            import comfy.model_management as model_management
+        except Exception:
+            return
+        if not hasattr(model_management, "LoadedModel") or not hasattr(
+            model_management,
+            "current_loaded_models",
+        ):
+            return
+        for loaded_model in model_management.current_loaded_models:
+            wrapper = getattr(loaded_model, "_tts_wrapper_ref", None)
+            real_model = getattr(loaded_model, "real_model", None)
+            resolved_real_model = (
+                real_model()
+                if isinstance(real_model, weakref.ReferenceType)
+                else real_model
+            )
+            if wrapper is self or resolved_real_model is self:
+                self._comfy_loaded_model = loaded_model
+                self._comfy_model_management = model_management
+                return
+        try:
+            loaded_model = model_management.LoadedModel(self)
+            loaded_model.real_model = weakref.ref(self)
+            finalizer = getattr(
+                model_management,
+                "cleanup_models",
+                lambda: None,
+            )
+            loaded_model.model_finalizer = weakref.finalize(self, finalizer)
+            loaded_model._tts_wrapper_ref = self
+            model_management.current_loaded_models.insert(0, loaded_model)
+            self._comfy_loaded_model = loaded_model
+            self._comfy_model_management = model_management
+        except Exception as exc:
+            print(
+                "⚠️ Failed to register isolated Audio8 TTS runtime with "
+                f"ComfyUI model management: {exc}"
+            )
+
+    def _unregister_from_comfy_model_management(self) -> None:
+        model_management = self._comfy_model_management
+        loaded_model = self._comfy_loaded_model
+        if model_management is None or loaded_model is None:
+            return
+        try:
+            if loaded_model in model_management.current_loaded_models:
+                model_management.current_loaded_models.remove(loaded_model)
+        except Exception as exc:
+            print(
+                "⚠️ Failed to remove isolated Audio8 TTS runtime from "
+                f"ComfyUI tracking: {exc}"
+            )
+        finally:
+            self._comfy_loaded_model = None
+            self._comfy_model_management = None
+
+    def to(self, device):
+        self.device = str(device)
+        self.current_device = self.device
+        self.load_device = self.current_loaded_device()
+        return self
+
+    def eval(self):
+        return self
+
+    def loaded_size(self) -> int:
+        if self._initialized and self.device.startswith("cuda"):
+            return self._estimated_memory_size
+        return 0
+
+    def model_size(self) -> int:
+        return self._estimated_memory_size
+
+    def model_memory(self) -> int:
+        return self._estimated_memory_size
+
+    def get_ram_usage(self) -> int:
+        return self._estimated_memory_size
+
+    def model_offloaded_memory(self) -> int:
+        return (
+            0
+            if self._initialized and self.device.startswith("cuda")
+            else self._estimated_memory_size
+        )
+
+    def model_mmap_residency(self, free: bool = False) -> tuple[int, int]:
+        del free
+        return 0, self._estimated_memory_size
+
+    def pinned_memory_size(self) -> int:
+        return 0
+
+    def lowvram_patch_counter(self) -> int:
+        return 0
+
+    def model_dtype(self):
+        return self.dtype
+
+    def model_patches_models(self):
+        return ()
+
+    def model_patches_to(self, target) -> None:
+        if isinstance(target, torch.dtype):
+            self.dtype = target
+        elif isinstance(target, torch.device):
+            self.device = str(target)
+
+    def is_dynamic(self) -> bool:
+        return False
+
+    def current_loaded_device(self) -> torch.device:
+        if self.device.startswith("cuda") and torch.cuda.is_available():
+            return torch.device(self.device)
+        return torch.device("cpu")
+
+    def partially_load(
+        self,
+        device,
+        extra_memory,
+        force_patch_weights=False,
+    ) -> int:
+        del extra_memory, force_patch_weights
+        self.to(device)
+        return 0
+
+    def partially_unload_ram(self, ram_to_unload: int) -> int:
+        return self.partially_unload("cpu", ram_to_unload)
+
+    def partially_unload(self, device, memory_to_free) -> int:
+        del device, memory_to_free
+        freed = self.loaded_size()
+        self.cleanup(unregister=False)
+        return freed
+
+    def model_unload(
+        self,
+        memory_to_free=None,
+        unpatch_weights=True,
+    ) -> bool:
+        del memory_to_free, unpatch_weights
+        self.cleanup(unregister=False)
+        return True
+
+    def detach(self, unpatch_weights=True) -> None:
+        del unpatch_weights
+        self.cleanup(unregister=False)
+
+    def is_clone(self, other: Any) -> bool:
+        return other is self
+
+    def cleanup(self, unregister: bool = True) -> None:
+        if unregister:
+            self._unregister_from_comfy_model_management()
+        else:
+            self._comfy_loaded_model = None
+            self._comfy_model_management = None
+        self._initialized = False
+        if getattr(self, "_session", None) is not None:
+            self._session.close()
+
+    close = cleanup
+
+    def __del__(self):
+        try:
+            self.cleanup()
+        except Exception:
+            pass
+
+
+def build_audio8_tts_isolated_proxy(
+    config: ModelLoadConfig,
+) -> Audio8TTSIsolatedProxy:
+    profile_name = config.runtime_profile or "vibevoice_transformers4_shared"
+    profile = get_runtime_profile(profile_name)
+    if profile is None:
+        raise RuntimeError(
+            f"Unknown isolated runtime profile '{profile_name}' for Audio8 TTS"
+        )
+    if profile.name != "vibevoice_transformers4_shared":
+        raise RuntimeError(
+            "Audio8 TTS requires the shared Transformers 4 runtime profile "
+            "'vibevoice_transformers4_shared'"
+        )
+    config.runtime_profile = profile.name
+    return Audio8TTSIsolatedProxy(config=config, profile=profile)

utils/runtimes/profiles.py

@@ -28,7 +28,7 @@ _VIBEVOICE_T4_PACKAGES = [
     "numpy>=1.26.4,<2.3.0",
     "soundfile>=0.12.0",
     "omegaconf>=2.3.0",
-    "transformers>=4.51.3,<=4.57.3",
+    "transformers>=4.57.0,<=4.57.3",
     "kernels>=0.6.1,<=0.9",
     "accelerate",
     "requests",
@@ -63,9 +63,12 @@ _QWEN3_T4_PACKAGES = [
 RUNTIME_PROFILES: Dict[str, RuntimeProfile] = {
     "vibevoice_transformers4_shared": RuntimeProfile(
         name="vibevoice_transformers4_shared",
-        engine_names=["vibevoice", "step_audio_editx"],
+        engine_names=["vibevoice", "step_audio_editx", "audio8_tts"],
         python_path_hint="runtimes/shared_legacy_t4/Scripts/python.exe",
-        description="Shared legacy Transformers 4 runtime for VibeVoice/Kugel, Step Audio EditX, and similar engines.",
+        description=(
+            "Shared legacy Transformers 4 runtime for VibeVoice/Kugel, "
+            "Step Audio EditX, Audio8 TTS, and similar engines."
+        ),
         inherit_base_site_packages=True,
         pip_packages=list(_VIBEVOICE_T4_PACKAGES),
         pip_packages_no_deps=[

utils/runtimes/session.py

@@ -1,9 +1,9 @@
-from __future__ import annotations
-
 """
 Persistent JSON-line worker session for isolated runtimes.
 """
 
+from __future__ import annotations
+
 import json
 import subprocess
 import threading
@@ -118,6 +118,24 @@ class JsonLineWorkerSession:
 
         self._process = None
 
+    def terminate_now(self) -> None:
+        """Stop a worker without waiting for the request lock."""
+        process = self._process
+        if process is None or process.poll() is not None:
+            return
+        try:
+            process.terminate()
+        except OSError:
+            return
+        try:
+            process.wait(timeout=2)
+        except subprocess.TimeoutExpired:
+            try:
+                process.kill()
+                process.wait(timeout=2)
+            except OSError:
+                pass
+
     def __del__(self):
         try:
             self.close()

utils/runtimes/workers/audio8_tts_worker.py

@@ -0,0 +1,191 @@
+"""Persistent Audio8 TTS worker for the shared Transformers 4 runtime."""
+
+from __future__ import annotations
+
+import json
+import sys
+import traceback
+from pathlib import Path
+from typing import Any, Dict, Optional
+
+import torch
+
+PROJECT_ROOT = Path(__file__).resolve().parents[3]
+project_root_str = str(PROJECT_ROOT)
+sys.path = [
+    entry for entry in sys.path if str(Path(entry).resolve()) != project_root_str
+]
+sys.path.insert(0, project_root_str)
+
+from utils.runtimes.protocol import RuntimeJobResponse  # noqa: E402
+
+
+def _emit(protocol_out, response: RuntimeJobResponse) -> None:
+    protocol_out.write(json.dumps(response.to_dict(), ensure_ascii=True) + "\n")
+    protocol_out.flush()
+
+
+def _load_reference_audio(
+    payload: Optional[Dict[str, Any]],
+) -> Any:
+    if payload is None:
+        return None
+    kind = payload.get("kind")
+    if kind == "audio_path":
+        audio_path = payload.get("audio_path")
+        if not audio_path:
+            raise RuntimeError("Audio8 TTS audio_path payload is missing its file path")
+        return audio_path
+    if kind == "tensor_path":
+        tensor_path = payload.get("tensor_path")
+        if not tensor_path:
+            raise RuntimeError(
+                "Audio8 TTS tensor_path payload is missing its file path"
+            )
+        tensor_payload = torch.load(tensor_path, map_location="cpu")
+        return {
+            "array": tensor_payload["waveform"],
+            "sampling_rate": int(tensor_payload["sample_rate"]),
+        }
+    raise RuntimeError(f"Unsupported Audio8 TTS reference payload kind: {kind}")
+
+
+def main() -> int:
+    protocol_out = sys.stdout
+    sys.stdout = sys.stderr
+    engine = None
+
+    for line in sys.stdin:
+        stripped = line.strip()
+        if not stripped:
+            continue
+
+        request = None
+        try:
+            request = json.loads(stripped)
+            action = request.get("action")
+            payload = request.get("payload") or {}
+            request_id = request.get("request_id")
+
+            if action == "shutdown":
+                _emit(
+                    protocol_out,
+                    RuntimeJobResponse(
+                        ok=True,
+                        result={"shutdown": True},
+                        request_id=request_id,
+                    ),
+                )
+                break
+
+            if action == "ping":
+                _emit(
+                    protocol_out,
+                    RuntimeJobResponse(
+                        ok=True,
+                        result={"pong": True},
+                        request_id=request_id,
+                    ),
+                )
+                continue
+
+            if action == "initialize":
+                import transformers
+                from packaging.version import Version
+
+                transformers_version = Version(transformers.__version__)
+                if not (Version("4.57.0") <= transformers_version < Version("5.0.0")):
+                    raise RuntimeError(
+                        "Audio8 TTS requires Transformers >=4.57.0,<5; "
+                        f"worker loaded Transformers {transformers.__version__}"
+                    )
+
+                from engines.audio8_tts.audio8_tts_engine import (
+                    Audio8TTSEngine,
+                )
+
+                engine = Audio8TTSEngine(
+                    model_name=(request.get("model_name") or "Audio8-TTS-Preview-0.6b"),
+                    model_dir=payload.get("model_path"),
+                    device=request.get("device") or "auto",
+                    dtype=payload.get("dtype", "auto"),
+                )
+                engine._ensure_model_loaded()
+                _emit(
+                    protocol_out,
+                    RuntimeJobResponse(
+                        ok=True,
+                        result={
+                            "model_name": engine.model_name,
+                            "sample_rate": engine.SAMPLE_RATE,
+                            "transformers_version": (transformers.__version__),
+                        },
+                        request_id=request_id,
+                    ),
+                )
+                continue
+
+            if engine is None:
+                raise RuntimeError(
+                    "Audio8 TTS worker received generation before initialization"
+                )
+            if action != "generate":
+                raise RuntimeError(f"Unsupported Audio8 TTS worker action '{action}'")
+
+            reference_audio = _load_reference_audio(payload.get("reference_audio"))
+            audio = engine.generate(
+                text=payload["text"],
+                reference_audio=reference_audio,
+                reference_text=payload.get("reference_text"),
+                max_new_tokens=int(payload.get("max_new_tokens", 1024)),
+                retry_max_new_tokens=int(payload.get("retry_max_new_tokens", 2000)),
+                temperature=float(payload.get("temperature", 0.8)),
+                top_p=float(payload.get("top_p", 0.95)),
+                top_k=int(payload.get("top_k", 50)),
+                do_sample=bool(payload.get("do_sample", True)),
+                seed=int(payload.get("seed", 42)),
+            )
+            audio = torch.as_tensor(audio).detach().float().cpu()
+            if audio.ndim == 1:
+                audio = audio.unsqueeze(0)
+            if audio.ndim != 2 or audio.shape[0] != 1:
+                raise RuntimeError(
+                    f"Audio8 TTS generated invalid audio shape {tuple(audio.shape)}"
+                )
+
+            output_path = Path(payload["output_path"])
+            torch.save(
+                {
+                    "audio": audio.contiguous(),
+                    "sample_rate": engine.SAMPLE_RATE,
+                },
+                output_path,
+            )
+            _emit(
+                protocol_out,
+                RuntimeJobResponse(
+                    ok=True,
+                    result={
+                        "output_path": str(output_path),
+                        "sample_rate": engine.SAMPLE_RATE,
+                    },
+                    request_id=request_id,
+                ),
+            )
+        except Exception as exc:
+            request_id = (
+                request.get("request_id") if isinstance(request, dict) else None
+            )
+            _emit(
+                protocol_out,
+                RuntimeJobResponse(
+                    ok=False,
+                    error=f"{exc}\n{traceback.format_exc()}",
+                    request_id=request_id,
+                ),
+            )
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

utils/text/segment_parameters.py

@@ -115,12 +115,13 @@ PARAMETER_ENGINES = {
     'seed': {
         'chatterbox', 'chatterbox_official_23lang', 'f5tts', 'higgs_audio',
         'higgs_audio_v3', 'vibevoice', 'index_tts', 'step_audio_editx', 'cosyvoice', 'qwen3_tts',
-        'dots_tts', 'fish_audio_s2', 'omnivoice',
+        'dots_tts', 'audio8_tts', 'fish_audio_s2', 'omnivoice',
         'echo_tts', 'moss_tts', 'moss_soundeffect_v2', 'dramabox'
     },
     'temperature': {
         'chatterbox', 'chatterbox_official_23lang', 'f5tts', 'higgs_audio',
-        'higgs_audio_v3', 'vibevoice', 'index_tts', 'step_audio_editx', 'qwen3_tts', 'moss_tts', 'fish_audio_s2'
+        'higgs_audio_v3', 'vibevoice', 'index_tts', 'step_audio_editx', 'qwen3_tts',
+        'moss_tts', 'fish_audio_s2', 'audio8_tts'
     },
     'cfg': {
         'f5tts', 'vibevoice', 'index_tts', 'chatterbox', 'chatterbox_official_23lang',
@@ -214,10 +215,12 @@ PARAMETER_ENGINES = {
         'f5tts', 'cosyvoice', 'omnivoice'
     },
     'top_p': {
-        'higgs_audio', 'higgs_audio_v3', 'vibevoice', 'index_tts', 'qwen3_tts', 'moss_tts', 'fish_audio_s2'
+        'higgs_audio', 'higgs_audio_v3', 'vibevoice', 'index_tts', 'qwen3_tts',
+        'moss_tts', 'fish_audio_s2', 'audio8_tts'
     },
     'top_k': {
-        'higgs_audio', 'higgs_audio_v3', 'index_tts', 'qwen3_tts', 'moss_tts'
+        'higgs_audio', 'higgs_audio_v3', 'index_tts', 'qwen3_tts', 'moss_tts',
+        'audio8_tts'
     },
     'audio_temperature': {
         'moss_tts'
@@ -238,7 +241,7 @@ PARAMETER_ENGINES = {
         'moss_tts'
     },
     'max_new_tokens': {
-        'higgs_audio_v3', 'moss_tts', 'fish_audio_s2'
+        'higgs_audio_v3', 'moss_tts', 'fish_audio_s2', 'audio8_tts'
     },
     'max_generate_length': {
         'dots_tts'