JLC CaptionForge Template Options

If you've ever hand-captioned a LoRA set, you know the drill: "describe the hair, the clothing material, the pose, the lighting, the framing…" - and you know exactly how many captions you've seen that nailed the face and forgot the rest. JLC CaptionForge Template Options is the pack's answer to that: one sidecar node with 31 boolean toggles that inject consistent prompt modifiers into your Joy, Qwen, and Ollama caption witnesses.

The point is consistency at scale. Instead of pasting the same "describe the accessories, describe the materials, avoid meta lead-ins" paragraph into three caption nodes, you flip checkboxes once here and every witness shares the same instruction envelope. The author's framing is blunt about it: this is the only source for template modifiers and the name input across the caption nodes. One place to edit, one behavior everywhere.

How it works

It's a config carrier, not a captioner. Each checkbox maps to a specific instruction that gets folded into the caption node's prompt when caption_template_mode is active on that node. The tooltips are the author's own wording, so they're reliable - e.g. option 02 is "describe only visible details, do not infer backstory, unseen anatomy, or hidden clothing," and option 05 kills the "this image shows…" meta lead-ins. The name_input__replacement_for_name field swaps into any template option that references {name}, which is how you get a character name threaded through a batch without retyping it.

The toggle set reads like the community's caption checklist distilled into checkboxes: facial features, eye color vs. makeup, hair, skin tone and texture, body shape, clothing pieces, materials and textures, precise colors, accessories, pose, hands, framing, viewing angle, background, lighting, style/medium, watermarks, visible text - plus the rating options (SFW/suggestive/NSFW, neutral sensual details, vulgar wording allowed, no polite euphemisms) that matter for the uncensored end of this pack.

Inputs and outputs

Inputs are just the name replacement field and the 31 booleans. All of them default to off, which is the gotcha: wire this up and do nothing else, and you'll get the same generic captions as before. You have to actually switch on the details you want.

Outputs are two: template_options (the CAPTIONFORGE_EXTRA_OPTIONS object you connect into a caption node's template_options pin) and template_options_json (the same thing as a string, if you want to inspect or log it).

Install

It's part of the pack, so one install serves everything:

git clone https://github.com/Damkohler/CaptionForge.git ComfyUI/custom_nodes/CaptionForge

Restart ComfyUI. No models, no extra deps - it's pure configuration plumbing.

Common issues

The two ways people get burned: forgetting to connect the template_options output into a caption node (nothing propagates), and forgetting that toggles are off by default. Also worth knowing: if you set custom_prompt_mode on a caption node, it takes precedence over the template path, and your carefully-built options get skipped. It's a thin node on purpose - if a feature is missing, it's because it's meant to be prompt text, not a new knob.