DOGMA Restoration Brief v16

A one-input, one-output text node looks like a joke until you build a graph where the same sentence has to reach four places. In a restoration pipeline that sentence is the project brief - the thing that tells the model what era, what place, what photographic character it is allowed to assume - and DOGMA Restoration Brief v16 is a single place to hold it.

What it is

A multiline STRING widget with a STRING output of the same name, brief. The default text is Preserve the source image faithfully. That default is the giveaway: the author's intent is conservative restoration, not creative rewriting.

The mechanism is a passthrough. It normalizes the value to a string and emits it. That's all. Its usefulness is entirely in the wiring.

Why a passthrough earns a node

The pack's planners take a global_target string that gets baked into every per-category edit instruction - the slot previews literally read KLEIN: Reconstruct only the masked vehicles ... appropriate to {target}. That target string is the thing you fiddle with twenty times a session: "Milan, Italy, 1972–1978", then "Moscow, 1961", then "keep it neutral, don't date it at all".

Without a holder node, that text is typed into a widget on the planner. Change the planner in your graph - and this pack ships six generations of planner, V38 through V44 plus DOGMAScenePlanSlots - and you retype the brief. Worse, if the brief also has to reach a second consumer (a scene-instruction builder, a tile composer, a checkpoint prompt for the base pass), you now have two copies of a sentence that must stay in sync, and they never do.

So: one Brief node, fanned out. It also reads well. When you open a workflow someone shared, a node called "Restoration Brief" with your sentence in it explains the whole graph faster than any amount of tracing wires.

Inputs and outputs

brief in, brief out. No parameters, no model, no state, nothing computed. Since the planner's global_target is a widget with a default rather than a required input, the usual pattern is to right-click global_target and convert it to an input, then wire this node into it.

Keep the text short and factual. The preview output on the planner prints the whole composed instruction, so you can see how your sentence lands inside it. Two things are worth writing down: the setting (place + period) and the preservation contract (preserve count, positions, lighting, character). Everything this pipeline does downstream is "repair what's there", and a brief that asks for a redesign will fight every protection rule the planners encode.

Install

Manager → search DOGMA Nodes → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/axior/ComfyUI-DOGMA-Nodes

Zero dependency install - the pack's requirements.txt says "No external dependencies", no models are downloaded by this node. It ships in the DOGMA/Semantic Detailer category with the rest of the 1.0.4 semantic nodes.

DOGMA Nodes comes from axior, the studio side of a Milan production team that runs commercial work through heavy ComfyUI inpaint passes. A brief-holder node is a tell: it exists because somebody was running the same archival-restoration project repeatedly and got tired of retyping the setting.

Nothing to troubleshoot?

Correct - as long as you remember the widget-vs-input distinction. If you wire something into a text-consuming node and nothing happens, check whether that node's text field is an actual input socket or a widget that just looks connected. And since this is a passthrough, a typo in the brief propagates verbatim into every model call; there is no cleanup between here and the prompt. Keep it to one or two sentences, in the same language you write your prompts in.