Workflows

Text to image

Use txt2img for a new composition. Set the model, prompt, negative prompt, resolution, steps, CFG scale, sampler, and optional seed. Generation is queue-backed, so the interface remains available while the job runs.

Image to image

Switch the primary mode to img2img, then upload an image or use a generated result as the base canvas. Add a prompt and adjust strength: lower values preserve more of the source; higher values allow a larger visual change.

Refine a design

Select a result, choose a design edit model in the Refine panel, and write an instruction such as “make the lighting warmer” or “replace the background with a quiet studio wall.” Refine creates a new version, preserving the original in the version strip.

Output gallery

The gallery is the unified stream for local generations and agent-created images. Select a result to make it the primary canvas image, save it to disk, or add it to the project asset library.

Parameters at a glance

Control	Use it for
Steps	More denoising passes; slower, sometimes more detail
CFG scale	How strongly the image follows the prompt
Seed	Reproducibility; use a fixed value to iterate
Sampler	The denoising strategy and speed/quality tradeoff
Strength	How far img2img moves from the source image