T8 Candidate Selector(候选选择)
Pick a prompt direction without burning another API call
- selected_candidate
The whole point of generating multiple candidate prompts is that you get to choose - and T8 Candidate Selector(候选选择) is the choosing node. It's about as simple as a ComfyUI node gets: it takes the JSON that T8 Creative Candidate Lab produced, you set a number, and out comes one prompt as a plain STRING. No LLM call, no API key, nothing leaves your machine.
It fills a genuinely annoying gap. The Candidate Lab burns one paid request to return 2–4 structurally different directions, and if you want to actually try direction #3, someone has to extract it from the JSON. You could do that by hand - or you could stick this node in the graph and let it be the thing that picks. The nice part is you can change your mind without re-running the Lab: the candidates are still sitting there in the JSON, so flipping candidate_index from 1 to 3 is instant and free.
How it works
Mechanically it's trivial, which is why the source is so short. The node parses candidates_json, walks the candidates array, grabs the prompt field from the entry you selected, and passes it through.
candidates_json- wire this from the Candidate Lab'scandidates_jsonoutput. The Lab saves the full payload here, including the soft scores and comparison notes, so you can peek at the report before you commit.candidate_index- 1 to 4, default 1. This is the one input that matters.
Output is a single selected_candidate STRING, which you then feed into the main prompt of your H3 or Seedance enhancer, or into T8 Storyboard Pack's concept input to storyboard the direction you picked. That second path is the smart one: Lab → Selector → Storyboard, so the whole pipeline chases the same concept instead of you copy-pasting.
Getting it installed
It ships inside the comfyui-minimax-h3-prompt-enhancer-T8 pack (the T8 Creative Suite alongside the H3 / Seedance / Music 3 prompt enhancers). ComfyUI Manager - search "MiniMax H3 / Seedance 2.0 / Music 3 Prompt Enhancer (T8)" - or the terminal route:
cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/comfyui-minimax-h3-prompt-enhancer-T8
Restart ComfyUI and hard-refresh the browser with Ctrl+F5 after installing. No extra dependencies for this one; it's pure Python with no network access.
The traps
Both failure modes are self-inflicted and clearly reported. If the JSON isn't valid (or isn't a Candidate Lab payload), you get a "候选 JSON 无效" error. If candidate_index points past the number of candidates that actually came back - say you asked for 2 but set index 3 - you get an out-of-range error. That's it. There's no caching or state; the node reads whatever JSON is on the wire, so if the Lab upstream re-runs with a different seed, the indices shift accordingly. One habit worth keeping: glance at the comparison report from the Lab before you commit to a pick, because the soft scores are honest heuristics, not gospel - they're meant to assist choice, never to overrule it.
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| candidates_json | STRING | — | |
| candidate_index | INT | 11–4 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| selected_candidate | STRING | — |