Dia 1.6B 0626
Dia 1.6B 0626 is the dated release used by Nari's Transformers examples. It retains the script-to-dialogue workflow and accepts reference audio for voice conditioning. The repository was created on 26 June 2025.
The processor handles text and audio together
The voice-conditioning example supplies audio and a transcript to the processor. It obtains the audio-prompt length before decoding the newly generated waveform. This keeps the reference prefix separate from the requested output.Dia-1.6B-0626 model card
The generation limit uses audio tokens
Nari's example sets max_new_tokens to 256 and describes this as roughly two seconds of audio. Treat this as an audio-generation setting, not a language-model context-window specification. The card uses 44.1 kHz reference audio in its example.Dia-1.6B-0626 model card
A new repository does not prove every claimed improvement
Practical guidance. The dated repository and integration examples establish this release's identity. They do not provide a complete comparison of its weights or speech quality against the April checkpoint. Compare both with the same script and reference recording when changing an existing workflow.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | nari-labs/Dia-1.6B-0626 |
| Published weight size | 1,611,160,576 parameters, matching the size count of the original release. Equal counts do not establish identical weights.Dia June revision repository metadataOriginal Dia repository metadata |
| Language and license | English; Apache 2.0, with the model card's stated research-use and misuse conditions.Dia-1.6B-0626 model card |
| Library interface | The documented Transformers classes are AutoProcessor and DiaForConditionalGeneration.Dia-1.6B-0626 model card |
Related models
Nari's original English dialogue speech model, with speaker tags, audio prompting and downloadable 1.6B weights.
Launch date not confirmedDownloadableThe larger Dia2 streaming dialogue model, with a wider decoder, audio-prefix conditioning and word-timing output.
Launch date not confirmedDownloadable