Nomad Studio/ Documentation/ Create/ Chat

Create · Conversational

Chat

Chat is a looping flow, not a one-shot one. You write or paste a starting prompt, generate, then keep evolving the same scene turn by turn with short instructions — "move to the right", "it starts raining", "make it night". Each turn the LLM rewrites the prompt to apply just your instruction, preserving everything else, and the whole session is remembered (a running chat history and a scene index) so the result view can show it as a timeline.

Reach for it when you want to iterate on one scene rather than generate unrelated one-offs — the LLM always has the previous prompt as its starting point, so "the same but at night" actually stays the same except for the light.

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

   Your starting prompt

   Your opening prompt. Confirming it stores it as the session's world context, which stays available to every later turn.

3. Step type: LLM transform — llm.story_continue

   Update scene prompt

   On the first turn, takes your starting prompt as-is (only cleaning up formatting if it's already complete). On every later turn, takes your change instruction plus the full chat history and applies just that change — same Level/Flair controls as Assist.

   Show the exact system prompt

   You are an image prompt generator for AI image generation. You receive a prompt or an instruction to change the current scene, along with the history of previous prompts.
   Rules:
   - The prompt must be fully self-contained and detailed enough to generate the image on its own.
   - On the first turn (no chat history), if the input is already a complete image prompt, return it with minimal or no changes — only clean up obvious formatting issues.
   - On subsequent turns, apply the user's instruction to the last prompt in the history. Preserve characters, setting, and style unless the instruction says otherwise. Only change what is asked.
   - Write in the comma-separated descriptive style used by Stable Diffusion and Flux image models.
   - Output ONLY the image prompt text. No JSON, no labels, no explanation.

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

   Scene prompt

   The AI's rewritten prompt for this turn, fully editable before generating.

5. Step type: Parameter group

   Parameters

   Seed, steps, CFG, image count.

6. Step type: Action — generation.submit_queue

   Generate Image

   Sends the scene prompt to the queue. On success, the prompt is appended to the session's chat history and the scene counter increments — this is what makes the next turn aware of it.

7. Step type: Result view

   Result

   Shows the current scene number and a horizontal timeline of the whole session so far. Five actions:

   • Save, Upscale — as usual.
   • Regenerate — same prompt, new roll.
   • Continue story — opens the next step to keep evolving the scene.
   • End story — closes the session.
8. Step type: Prompt input (loop trigger)

   Change the scene

   Appears only after clicking "Continue story". Describe what should change — everything else stays the same. Submitting it loops back to the "Update scene prompt" step above, and the cycle repeats for as many turns as you like.

Examples