Text Template

Templating is how you stop rewriting prompts by hand, and VLMTextTemplate is the pack's safe version of it. You write a template with {named} placeholders, supply values from a JSON object, and it renders the final string - with the twist that missing values are handled explicitly instead of silently. No more "why does my prompt have a literal {label} in it?" debugging sessions.

The template is rendered from two sources: a variables_json object for arbitrary named placeholders, and four dedicated live text sockets - text1 through text4 - for the values you want wired in from other nodes rather than typed into JSON. It's the difference between "prompt as data" and "prompt as wiring," and you usually want a mix.

How it works

Your template is a plain multiline string: "Describe {object} in the {context}.". Placeholders are named - no Python attribute/index expressions, no eval - which is a deliberate safety decision: the template engine never evaluates code, never follows references, and never touches files or the network. What you see is what gets substituted.

Values come from:

• variables_json - a JSON object whose keys are placeholder names. Default example: {"instruction":"Describe the input accurately."}
• text1…text4 - convenient live sockets mapped to {text1}…{text4} placeholders. Wire a node's output here and reference it in the template without touching JSON.

Then the important part: missing_values policy, one of three:

• Keep placeholder - leave {name} in the output (good for partial templates).
• Empty string - render missing as empty.
• Error (default) - fail the node loudly so you find out immediately that a variable never got supplied.

Error is the right default for automation. A template that renders with a hole in it is a prompt you'll discover is broken two queues later; a template that fails instantly is a bug you fix in thirty seconds.

Outputs

• text - the rendered result.
• variables_json - the values actually used, echoed back for inspection.
• missing_keys_json - the keys that were referenced but missing, which makes the Error case diagnosable instead of opaque.

Install

Part of ComfyUI VLM Nodes (gokayfem/ComfyUI_VLM_nodes). ComfyUI Manager → search "VLM Nodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/gokayfem/ComfyUI_VLM_nodes
python -m pip install -r ComfyUI/custom_nodes/ComfyUI_VLM_nodes/requirements.txt

Run with ComfyUI's Python; the repo won't touch torch. Pure string rendering, no models, no downloads.

Gotchas

The template field is direct-name substitution - {a.b} or {a[0]} style expressions are not supported, and that's on purpose. If you need nested JSON lookups, pre-process with VLMJSONExtract and feed the result in as a named variable or a text socket. And keep an eye on the missing_values default: if you're building a template where a variable is sometimes absent by design, switch the policy to Keep placeholder or Empty string rather than letting it error on you every other run.