[v0.3.0] Instruction/fine-tune feature removed from generation UI

[hariOneb]

Hi! In v0.2.x there was a sliders button in the generation bar
that opened "Fine tune instructions & effects" with:

• An instruction text field (e.g. "very happy and excited")
• Load preset dropdown
• Add effect dropdown

This feature is completely gone in v0.3.0. The generation bar
now only shows a text input, effects dropdown, and generate button.

Was this intentional or accidentally removed during the UI refactor?
This was one of the most useful features for controlling voice style.

Please bring it back! 🙏

[jamiepine]

Hey! This was intentional — though I totally understand the confusion. I actually thought it was working too for a while.

After getting reports that the instruction field had no effect on the output, I dug into it and found that the Qwen3-TTS model we ship (the Base variant) silently drops the instruct parameter. It accepts it without error, but it has zero effect on generation. So the UI was giving users a text field that did nothing.

There is a Qwen3-TTS variant that supports instructions — Qwen3-TTS-12Hz-1.7B-CustomVoice — but that model uses predefined speaker IDs instead of reference audio. It's fundamentally incompatible with voice cloning, which is the core of Voicebox. You can't use a voice sample and delivery instructions with Qwen.

Here's the plan going forward for both features:

Voice Design (CustomVoice) — We're going to add support for the CustomVoice model as a "voice design" feature, where you can create profiles using a text description of the voice (tone, style, accent) instead of an audio sample. These would be a different type of profile — design-prompt-based rather than sample-based.

Delivery Instructions — We're working on adding CosyVoice2-0.5B, which has a proper inference_instruct2() API that supports text instructions for emotions, speed, volume, and dialects — and it works alongside voice cloning. This is the model that will bring the instruct UI back for real.

Sorry for the confusion with the silent removal. Should have called this out in the release notes.

[helpmebro911]

Will we be able to fully use and customize our voice?