Boyo H3 Trigger Injector
Bolt a different trigger word onto each scene of a MiniMax H3 prompt
- prompt
The name is more complicated than the job. BoyoH3TriggerInjector is a plain text-plumbing node from the Boyonodes pack: you feed it a multi-scene MiniMax H3 prompt - the kind where each scene block is separated by a --- line - and it splices a different prefix into the top of each of the first three blocks. That's it. No model loading, no VRAM, no API key. If you've ever hand-edited a long H3 prompt to jam "same character, gold jacket" above every scene block so your subject doesn't drift, this is the node that stops you doing that by hand.
Where it fits. MiniMax H3 (the 33B open-weights omni model that landed in August 2026) wants long, structured prompts, and the community's "promptor" tools happily generate them for you. For anything longer than a single shot you end up chaining: several ~10-second beats, one prompt block each, joined by ---, where every new shot conditions on the last frame of the previous one. Identity drift across those beats is the recurring pain - which is why good promptor workflows pin a subject anchor at the top of every block. The H3 Trigger Injector is the manual, surgical version of that habit: keep the big prompt text as-is and vary only the per-scene prefixes.
What it actually does
The source is short enough to read in one sitting. It splits your prompt on every ---, strips whitespace off each block, then walks the blocks in order. Scene 1's trigger text goes above the first block, scene 2's above the second, scene 3's above the third, each joined by a newline. Everything gets re-joined with \n---\n. A few behaviors worth knowing:
- Blank trigger boxes are skipped - no content inserted for that scene. Leave all three empty and the node is a clean passthrough; the docstring says it's safe to wire in permanently and fill in only when needed.
- It only touches the first three blocks. A fourth or fifth scene block passes through untouched, so don't expect triggers on a six-shot script.
- Output is normalized. Input blocks get stripped and rejoined with a canonical
\n---\n, so cosmetic blank lines around separators get tidied. Usually a feature, occasionally a surprise if a downstream node is fussy about exact formatting. - It splits on any
---, not just one on its own line. Don't use---as decoration inside a block's description; it'll chop the block in two. - One more quirk: if you want the same persistent token on every scene (the usual "keep the character" case), you paste it into all three boxes - each box is strictly per-scene.
The inputs that matter
prompt(required, multiline) - the full----separated multi-scene H3 prompt string.scene_1_triggers/scene_2_triggers/scene_3_triggers(optional, multiline) - trigger words or any prefix text for each scene. The tooltip is the spec here: "Leave blank to skip."
The single output is prompt (a STRING), which wires into whatever was already consuming your H3 prompt - the text-encode/conditioning stage or the script box of a multi-shot H3 sampler. Swap the injector in between your prompt source and that consumer, and you've got per-scene prefixing without touching the source text.
Installing it
It ships in the Boyonodes pack, so either install through ComfyUI Manager (search "Boyonodes") or:
cd ComfyUI/custom_nodes
git clone https://github.com/DragonDiffusionbyBoyo/Boyonodes
Restart ComfyUI and the node appears under H3/Promptor. This specific node is pure Python string handling - zero extra dependencies. The catch is that the pack's requirements.txt is a heavy kitchen sink (transformers, librosa, gradio, funasr, bitsandbytes…) aimed at its audio/TTS half, and Manager will happily try to install all of it. If all you want is this node, a bare clone without pip install runs fine.
Gotchas and honest notes
The pack's own README doesn't mention this node at all - it's a recent addition and its real documentation is the tooltips and the module docstring. If you ever forget what a field does, hover it. And one piece of context that isn't about the node itself: if you're in the US, EU, UK, or South Korea, the MiniMax H3 weights are license-locked out of your territory, even though this text utility is obviously fine anywhere. Build the workflow, know what you're legally allowed to run.
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | Multi-scene H3 prompt string separated by --- | |
| scene_1_triggersopt | STRING | Trigger words or prefix text for Scene 1. Leave blank to skip. | |
| scene_2_triggersopt | STRING | Trigger words or prefix text for Scene 2. Leave blank to skip. | |
| scene_3_triggersopt | STRING | Trigger words or prefix text for Scene 3. Leave blank to skip. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| prompt | STRING | — |