IndexTTS 2.5 Merge Voice Emotions · T8star-Aix

This node exists because of search. Multi-role work in the IndexTTS 2.5 pack funnels through a node the author calls 角色音色 / 情感合并 - role voices / emotion merge. But English-speaking users searching for that thing kept typing "Merge Voice Emotions," so the pack ships this node as a search-friendly alias with the same output. Same function, different name, one less "why can't I find this" moment.

What it actually does is modest and worth understanding precisely: it gathers 1–16 already-built voice profiles into one role library that the multi-role generator can consume. Each voice profile carries a role name, its reference audio, and an optional emotion. This node stacks them. That's it.

What "merge" means here (and doesn't)

The name invites a misunderstanding the author goes out of the way to kill: merging voice emotions does not mean blending "sad 60% + angry 40%" into one new eight-dimension mood vector. That's a different operation you'd do inside a single emotion-control node by setting two dimensions. This node merges role configurations - it produces a cast, not a cocktail. If you want one character whose mood is a mix, don't look here.

It does one real piece of bookkeeping: duplicate role names are rejected before the queue runs, so you can't silently end up with two "旁白" tracks fighting over the same reference. That's a thoughtful failure mode - it fails at queue time, not after a wasted generation.

Inputs and outputs

The single input, voices, is an auto-growing list of T8_INDEXTTS25_VOICE connections (1–16). Each one comes from a 角色音色 voice-profile node, or from a 已保存音色 saved-voice bundle. The workflow shape is: reference audio → voice profile → merge → multi-role generate.

Outputs:

• role_library (T8_INDEXTTS25_ROLE_LIBRARY) - feeds the 多角色 / SRT 生成 node. This is the wire that matters.
• role_info - a plain-text list of roles and their emotions, handy for a text preview node or just sanity-checking the cast.

Installing and wiring

Standard pack install: ComfyUI Manager → search IndexTTS 2.5 · T8star-Aix, or git clone https://github.com/T8mars/comfyui-indextts25-t8.git into custom_nodes/, restart. Then: one Load Audio per character, each into its own 角色音色 node (with an optional emotion-control node feeding its default emotion), all into this merge node, and the library into the multi-role generator alongside a parsed script from 批量台词 / SRT.

Why you'd reach for it

For a single voice you don't need this at all - the plain generate node is enough. Reach for it the moment a script has two or more characters, which is exactly the scenario the pack's SRT and batch-script support is built around: different voices per role, each keeping its own default emotion, with per-line overrides handled downstream. It's plumbing, not magic, but it's the plumbing that makes a two-character dialogue a one-queue task instead of a per-line slog.