Generation is only half the toolkit

Typing a prompt and hoping is the beginner mode. The real power comes from editing — starting from an existing image (or one you just generated) and steering it precisely. These techniques turn "roll the dice again" into "fix exactly this."

The core editing moves

• Image-to-image (img2img) — feed in a starting image plus a prompt; the model keeps the overall composition but reinterprets it. Great for turning a rough sketch or photo into a finished style.
• Inpainting — mask a region and regenerate only that part: remove an object, fix a mangled hand, swap a background, change a shirt's color. The rest of the image stays untouched.
• Outpainting — extend an image beyond its borders, letting the model invent what lies outside the frame.
• Upscaling — increase resolution and add detail after you're happy with the composition (tools like Magnific or Topaz). Compose small and fast, then upscale the keeper.

Controlling structure

When you need the pose or layout to match a reference exactly, ControlNet-style tools condition generation on an input — an edge map, a depth map, a stick-figure pose. You provide the skeleton; the model paints the character onto it. This is how you get an image that follows a precise composition instead of the model's whim.

The iterative mindset

Professionals rarely one-shot an image. A typical loop: generate a rough composition, inpaint to fix flaws, img2img to refine the style, then upscale. Each step is cheap and reversible.

Don't keep rerolling the whole prompt to fix one bad hand. Mask it and regenerate just that patch — editing tools let you change exactly what's wrong and keep everything that's right.

Try this: Generate an image you almost like, then use your tool's inpainting or eraser feature to mask one flawed area and regenerate only that spot. The difference between rerolling everything and surgically fixing one patch is the whole leap from hobbyist to capable.