The instruct parameter (natural-language delivery control) works correctly
via the API for Qwen3-TTS cloned profiles — I confirmed this by calling
POST /generate directly with "engine": "qwen" and an instruct string, and
it worked exactly as expected.
However, the generate box in the app only shows the instruct textarea/slider
when a Qwen CustomVoice profile is selected. It does not appear for cloned
profiles using the Qwen3-TTS engine, even though the backend fully supports
it for that engine too.
Steps to reproduce:
- Create a cloned voice profile using the Qwen3-TTS engine
- Select that profile in the generate box
- Notice there's no instruct/delivery-instructions field available
Expected: the instruct field should appear for any profile using the "qwen"
engine (both CustomVoice and cloned Qwen3-TTS), not just CustomVoice.
Workaround: calling POST /generate directly with "engine": "qwen" and an
instruct field works fine — this looks like a UI gating issue rather than
a backend limitation.
The
instructparameter (natural-language delivery control) works correctlyvia the API for Qwen3-TTS cloned profiles — I confirmed this by calling
POST /generate directly with
"engine": "qwen"and aninstructstring, andit worked exactly as expected.
However, the generate box in the app only shows the instruct textarea/slider
when a Qwen CustomVoice profile is selected. It does not appear for cloned
profiles using the Qwen3-TTS engine, even though the backend fully supports
it for that engine too.
Steps to reproduce:
Expected: the instruct field should appear for any profile using the "qwen"
engine (both CustomVoice and cloned Qwen3-TTS), not just CustomVoice.
Workaround: calling POST /generate directly with
"engine": "qwen"and aninstructfield works fine — this looks like a UI gating issue rather thana backend limitation.