Module 6 • Lesson 1🪄 Generative for Compositing

Generative fill is not a "make art" button — for a compositor it is a fast assistant that extends backgrounds, erases distractions, and grows the connective tissue between elements. You direct it, judge its guesses, and finish by hand. Used that way, it removes hours of cloning.

📚 What You'll Learn

By the end of this lesson, you will be able to:

Frame generative fill as a compositing tool, not a replacement for craft
Extend a background with Generative Expand to fit a layout
Remove distractions and fill gaps with Generative Fill + a prompt
Generate connective elements (shadow ground, foliage, debris) to sell a composite
Judge, mask, and hand-finish generative output — and respect its limits and licensing

⏱️ Estimated Time: 55 minutes

🎯 Project: Take a tight, awkward composite and use generative fill to give it room, clean it up, and blend its seams — then finish by hand.

In This Lesson

👤 The Goal: Fill the Gaps, Not the Vision

A composite cropped too tight, twice. On the right, the subject is jammed against the frame with a distracting sign behind them and no room to breathe. On the left, the canvas is extended and the sign painted out, so there is space to compose and the eye rests on the subject. Drag the handle:

Extended & clean Cramped

A diagram of the principle. Generative fill did the grunt work — new sky and ground, sign removed — and you spend your time on composition instead of cloning.

🧠 Mental Model: A Fast, Fallible Assistant

Treat generative fill like a talented but literal intern. You make a selection, type a short prompt (or leave it empty to just match surroundings), and it returns a few variations on their own generative layer — always non-destructive. It is superb at plausible texture (sky, grass, pavement, walls) and removals, and unreliable at specifics (hands, text, exact brands, precise counts).

So the workflow is never "generate and ship." It is generate → pick the best of three → mask and blend → hand-fix the tells. The AI buys you a fast, decent starting point; your craft from Modules 1–4 turns it into something that holds up.

Figure 1: The loop is always generate → choose → refine. The model gives you raw material on a non-destructive layer; you are still the one who finishes it.

🛠️ The Generative Workflow

The contextual task bar surfaces Generative Fill whenever you have a selection, and Generative Expand appears when you enlarge the canvas with the Crop tool. Both write to a new generative layer and offer a Variations panel; regenerate as many times as you like, and each attempt is kept so you can compare.

Contextual Task Bar

Figure 2: Type a short prompt, generate, and audition the variations. Keep the best on its generative layer and add a mask — never flatten until you have blended the seams.

✅ Short prompts, small selections

Generative fill does its best work on modest areas with simple prompts ("grass", "clear blue sky", or empty to match). Ask it to invent a whole complex scene and it flails. Break big asks into small, well-selected passes and results jump in quality.

⚖️ Limits, Ethics & Licensing

Generative output is probabilistic, so it drifts on anything that needs to be exactly right — faces, hands, readable text, logos, product details, repeat patterns. It can also introduce artifacts that only show at 100%. Always zoom in and audit.

Treat it as raw material. Expect to mask, clone, and dodge & burn the result into place. It starts the job; it does not finish it.
Match resolution. Generated regions may be softer or lower-detail than your source. Add matching grain/sharpening so they blend.
Know the licensing. Adobe's model is trained for commercial-safe use, but confirm the terms for your account and deliverable, especially for client work.
Be honest. For journalism or documentary work, generative changes may be off-limits; disclose edits when integrity matters.

⚠️ It won't fix a weak composite

Generative fill can't rescue mismatched perspective, light, or scale from Module 4 — it fills texture, it doesn't reason about your scene. Get the fundamentals right first; use generation to accelerate the tedious parts, not to paper over structural mistakes.

🧪 Guided Build: Harmonize a Composite

Open a cramped or messy composite (yours from Module 4, or any busy photo).

Step 1: Give it room · 8 min

Crop tool, drag the canvas edges outward, and choose Generative Expand (empty prompt to match). Generate, pick the cleanest variation.
Check the new edges at 100% for repeats or seams.

Step 2: Remove distractions · 10 min

Lasso a distracting object with a little margin. Run Generative Fill with an empty prompt to have it vanish into the surroundings.
If a variation leaves a smudge, regenerate or clean it with the Healing Brush.

Step 3: Grow connective tissue · 10 min

Select ground beneath a pasted subject and prompt something like "gravel with small shadows" to help it sit; or generate foreground foliage to overlap and integrate an edge.
Mask each generative layer so it only contributes where you want.

Step 4: Finish by hand · 8 min

Add matching grain and a touch of sharpening to the generated regions. Dodge & burn to marry their light to the scene.
Apply your Module 4 unifying grade over everything so the generated and real pixels share one color. 🏆

✅ Project Completion Checklist

☐ Canvas extended with Generative Expand, edges audited
☐ A distraction removed cleanly with Generative Fill
☐ Connective element generated and masked in
☐ Generated regions grain/sharpen-matched and hand-blended
☐ One unifying grade over the whole result

🧗 Now You: Solo Variation

🌟 Your challenge

Take a phone snapshot and turn it into a wide banner purely with Generative Expand — then fix every seam by hand until it is undetectable.
Remove three increasingly hard distractions (a sign, a person, a shadow across the subject) and note where the AI succeeds and where you have to take over.
Generate the same fill five times and study how the variations differ. Learning to curate is the real skill here.

Going further: combine generation with your masking (Module 1): generate a busy background, then composite your cleanly-cut subject on top so the subject stays pixel-perfect and only the surroundings are AI.

🍳 Recipe Card: Generative Assist

Generate, curate, finish

Generative Expand (Crop out + empty prompt) for room
Generative Fill + empty prompt to remove distractions
Short prompts, small selections for connective texture
Keep each result on its generative layer; mask, don't flatten
Grain/sharpen-match, hand-blend, then one unifying grade

Mantra: the model gives you a head start, never the finish line.

📓 Learning Journal

Add to your journal after this lesson:

Key concepts you learned
Techniques that clicked for you
Questions or confusion points to revisit
Ideas you want to try
Your progress and feelings about learning this

✍️ This lesson's prompt: Where does generative fill genuinely save you time, and where does trusting it cost you more time cleaning up than doing it by hand? Draw your own line between the two.

🔭 Where This Leads

You have a smart assistant for one-off gaps. Next you teach Photoshop to repeat your steps flawlessly across many files: Lesson 6.2 — Advanced Actions & Conditionals records, branches, and batches your workflow so volume work runs itself.

Is it "cheating" to use generative fill professionally?

No more than the Healing Brush was. It is a tool; the judgment of what to keep, blend, and reject is still yours. Clients pay for the finished result and your taste, not for which brush filled a patch of sky.

The generated area looks softer than the rest. Why?

Generated pixels often carry less fine detail than your capture. Match them with a little sharpening and the same grain as the source; that single step is what makes a generative fill invisible.