Generator basics

Start a generation job and understand modes, queue state, quotas, outputs, and safe retry behavior.

Last updated: 30 June 2026 · Template: getting-started · Audience: signed-in, supporter · Difficulty: intermediate · Reading time: 9 min

Before you start

A signed-in account with Generator access.
A prompt idea and a model/workflow that is currently enabled.

Workspace map

The Generator workspace is a task surface, not a gallery. It combines prompt inputs, generation controls, model selection, queue state, output review, history, and actions for download or publishing. Treat each job as a reproducible experiment: prompt, model, seed, dimensions, and advanced controls all matter.

Arc en Ciel generator workspace with prompt and output panels. — **Generator workspace** - The Generator workspace combines prompt input, controls, queue state, outputs, history, and publishing actions.

Arc en Ciel generator history panel. — **History is part of the basic workflow** - Treat history as the place to compare jobs, recover outputs, and verify settings before downloading or publishing generated media.

Core terms

txt2img: Generate from a text prompt and selected model settings.
img2img: Use an existing image as a starting point and guide how much it changes.
Inpaint: Mask part of an image and regenerate that area.
Queue: The line of pending or running jobs. A submitted job may wait before it starts.
Seed: A number that helps reproduce or vary results when other settings are similar.

Run your first job

Open Generate and wait for the policy/state controls to load.
Choose the mode that matches your input: txt2img for text-only, img2img or inpaint when starting from an image.
Pick a model or workflow that matches your intended style and base model expectations.
Write a positive prompt and optional negative prompt with clear subject, style, lighting, and constraints.
Review dimensions, steps, sampler, seed, and quota cost before submitting.
Submit once, then watch queue/history instead of clicking repeatedly.
Open the output before publishing so you can inspect quality, rating, and metadata.

Before pressing Generate

Prompt is specific enough - The model cannot infer missing subject, medium, angle, lighting, or style constraints reliably.
Model and LoRAs match - A LoRA made for one base model family may behave poorly on another.
Dimensions are intentional - Large or unusual sizes can cost more and fail more often.
You understand quota cost - Do not spend quota debugging a typo repeatedly.

Common first-run problems

Generate controls are unavailable

Likely cause: Generator may be disabled, your account may lack access, or state has not loaded.

Fix: Refresh once, check account role/support state, then check Status/Changelog if it persists.

Job waits in queue

Likely cause: Workers may be busy or concurrency may be limited.

Fix: Wait for queue state; avoid duplicate submissions unless the first job clearly failed.

Result ignores a concept

Likely cause: Prompt too vague, wrong model, missing trigger word, or incompatible LoRA.

Fix: Change one variable at a time and compare history.

Open Generator