Nomad Studio/ Documentation/ Create/ Variations

Create · Series & batch

Variations

Variations takes one prompt template with [bracketed placeholders] and has the LLM fill in every bracket with a different, specific, visually concrete value — N times (1 to 20, default 4) — producing N genuinely distinct prompts from one structure, generated as one batch.

You don't have to write the template by hand: paste a finished prompt (or drop an image with embedded ComfyUI/A1111/CivitAI metadata) and a dedicated step extracts a bracketed template from it automatically.

Guided walkthrough

1. Step type: Workflow picker

   Choose engine

   Pick the imported engine, or auto-selected if you only have one.

2. Step type: Prompt input (optional)

   Input prompt

   Optional. Paste any finished prompt, or drop an image whose embedded generation metadata gets extracted automatically, as raw material for the next step.

3. Step type: LLM transform — llm.studio_templateize

   Templateize

   Converts the input prompt into a bracketed template automatically — identifying which parts should become [placeholders].

4. Step type: Prompt input (pre-filled)

   Prompt template

   Required. Prefilled from Templateize if you used it, or write your own [bracket] template from scratch.

5. Step type: LLM transform — llm.studio_variations

   Generate variations

   Fills in every bracket with a specific value, N times (1–20, default 4). Runs as one continuous conversation across all N so the model doesn't repeat itself between variations.

   Show the exact system prompt

   You are a creative AI image prompt specialist. The user provides a prompt template with [bracketed placeholders]. Fill in every [bracket] with a specific, vivid, and creative value to produce one complete image-generation prompt.
   
   Rules:
   - Preserve the entire template structure and all text outside the brackets exactly as given.
   - Replace EVERY [bracketed placeholder] with a specific and visually concrete value.
   - Write in the comma-separated descriptive style used by Stable Diffusion and Flux image generation models.
   - Do NOT include any explanation, markdown, or wrapper text.
   Output ONLY the single filled-in prompt string — no JSON array, no quotes, no labels.

6. Step type: Parameter group

   Parameters

   Seed, steps, CFG, image count.

7. Step type: Action — generation.submit_queue

   Generate Images

   Queues all N filled-in prompts.

8. Step type: Result view

   Result

   Save, Upscale, Regenerate — no Send to Refine on this one.

Examples