Variations JSON

Uses local LLM Requires ComfyUI Flow id: create.variations_json 6 steps

Variations JSON is Variations' structured sibling: instead of writing a [bracketed] template, you define your variables as plain JSON key-value pairs (subject, environment, style, mood, lighting — pre-filled with a starter example), and the LLM varies those fields directly, N times.

Reach for it when a bracketed template feels awkward for what you're describing, or you'd rather think in named fields than inline placeholders.

Guided walkthrough

Step type: Workflow picker
Choose engine

Pick the imported engine, or auto-selected if you only have one.
Step type: JSON input
Variables

A JSON editor pre-filled with an example: subject, environment, style, mood, lighting.
Step type: LLM transform — llm.studio_variations_json
Generate variations

Varies the JSON fields directly, N times (1–20, default 4), as one continuous run so the results don't repeat each other.
Step type: Parameter group
Parameters

Seed, steps, CFG, image count.
Step type: Action — generation.submit_queue
Generate Images

Queues all N generated prompts.
Step type: Result view
Result

Save, Upscale, Regenerate.

Examples

Variables: subject: "an elegant woman in a flowing champagne silk gown" · environment: "a grand marble staircase in a palatial ballroom" · lighting: "warm chandelier light" — Krea2 Roma · 9:16 · N=2

Variation 1: an elegant woman in a champagne silk gown on a marble staircase under chandelier light

Nomad Studio Variations JSON page showing both filled-in variations from the JSON variables

Variation 1 (left) · App showing both variations at 1920×1080 (right)

Same variables, second fill: deep sapphire velvet gown, cool moonlight mixed with chandelier glow

Variation 2: an elegant woman in a sapphire velvet gown lit by moonlight and chandelier glow

Variation 2 (left) · Same batch, same app view (right)

Variations JSON

Choose engine

Variables

Generate variations

Parameters

Generate Images

Result