This is not an article to read once and forget. It is a tool to keep open while you work. The checklist is organized into four phases, before you generate, while you generate, when refining, and before you ship, with a one-line justification per item so you understand why it matters. Run it on real projects and your hit rate climbs from the first session.
For the reasoning behind the underlying mechanics, pair this with our complete guide. Here we stay terse and actionable.
Phase 1: Before You Generate
Set yourself up before typing anything into the box.
- Define subject, style, setting, and mood on paper. Why: unspecified elements get filled with the model's bland defaults.
- Confirm the concept fits the model's strengths. Why: exact products, accurate text, and contact poses are known weak spots; reframe early if needed.
- Pick the right tool for the job. Why: photorealism, illustration, and text rendering are different strengths held by different tools.
- Choose an aspect ratio that suits the subject. Why: forcing wide frames on single subjects causes duplication.
- Pull your reusable style and negative-prompt fragments. Why: starting from proven recipes beats reinventing every time.
If any concept-fit item fails here, fix it now. Discovering a poor tool-task fit after twenty generations is the most common waste, and our common mistakes guide explains why.
Phase 2: While You Generate
The core production loop.
- Lead the prompt with the subject. Why: most models weight earlier tokens more heavily.
- Layer details: subject, setting, style, lighting, then sparse quality cues. Why: structured prompts steer more predictably than word dumps.
- Apply your standard negative prompt. Why: it removes a whole class of defects before they appear.
- Set steps around 25 to 40 and guidance scale around 6 to 9. Why: the practical sweet spot; higher is usually worse, not better.
- Generate a batch of four, not one. Why: a batch reveals how the model is interpreting your words.
- Pick a single anchor image, even if imperfect. Why: refinement needs a fixed starting point.
This loop is the heart of our step-by-step approach; the checklist is its condensed, repeatable form.
Phase 3: When Refining
Turn a promising anchor into a finished image.
- Lock the anchor's seed. Why: it makes subsequent changes controlled variations instead of new images.
- Change exactly one variable per generation. Why: isolating variables is the only way to learn what each lever does.
- Use inpainting for localized defects. Why: fixing a bad hand locally beats rerolling a good composition.
- Use outpainting or cropping for final framing. Why: it lets you reach unusual aspect ratios without duplication.
- Upscale only after the composition is locked. Why: separating composition from resolution yields cleaner results.
Discipline in this phase is what separates deliberate operators from people stuck rerolling, the central argument of our best practices guide.
Phase 4: Before You Ship
The final gate. Never skip it.
- Run a defect scan: hands, eyes, text, symmetry, repeated patterns. Why: these flaws are systematic and easy to miss when you like the image.
- Check the image at full size, not a thumbnail. Why: artifacts hide at small scale and surface after publishing.
- Verify any text in the image is correct or removed. Why: generated text is frequently gibberish even on strong models.
- Confirm the style matches the rest of the set or brand. Why: consistency is a competitive advantage, especially across a body of work.
- Save the full recipe: prompt, negative, seed, CFG, steps, sampler. Why: a reusable recipe library compounds over time.
A thirty-second defect scan before shipping prevents the embarrassing six-fingered-hand publish that erodes credibility.
How to Use This Checklist in Practice
Do not treat all twenty-one items as equally heavy every time. For high-stakes images, run the full list. For disposable, high-volume social content, run a lighter version, concept fit, negative prompt, batch, defect scan, and move fast. Matching rigor to stakes is itself a skill our real-world examples guide illustrates.
The phases that pay off most for almost everyone are concept fit in Phase 1, single-variable iteration in Phase 3, and the defect scan in Phase 4. If you adopt nothing else, adopt those three. They eliminate the largest sources of wasted time and shipped mistakes.
A Quick Pre-Project Decision Tree
Before you even open the tool, run three fast questions. They route you to the right approach and save the most time of anything on this page.
- Does this need exact text, a specific product, or a contact pose? If yes, plan for compositing, references, or reframing now, because pure text-to-image will fight you on all three.
- What are the stakes? High-stakes images get the full checklist; disposable content gets the light version. Decide before you start, not halfway through.
- Do I already have a recipe for this kind of image? If yes, start from it. Restarting from scratch when a proven recipe exists is pure waste.
Answering these three before generating prevents the two most expensive mistakes: forcing the wrong tool onto a job it cannot do, and over-investing effort in something disposable.
Common Checklist Failures and Fixes
Even people who keep a checklist tend to fail in predictable ways. Watch for these.
- Skipping the full-size view. Thumbnails hide artifacts. If you only ever judge images small, you will ship defects. Always zoom to full resolution before approving.
- Treating the anchor as final. The Render output is raw material. People fall in love with a near-miss and ship it instead of refining. Hold the bar at the refinement stage.
- Letting the negative prompt rot. A negative prompt copied once and never reviewed can carry contradictory or irrelevant terms. Prune it occasionally so it still targets the defects you actually see.
- Never updating recipes. A recipe that worked on an older tool version may drift. When a saved recipe underperforms, re-tune it rather than rerolling blindly.
Catching these keeps the checklist honest. A checklist you follow mechanically without judgment slowly stops protecting you.
Frequently Asked Questions
Which items matter most if I only adopt a few?
Three: confirm concept fit before starting, change one variable at a time while refining, and run a defect scan before shipping. These three eliminate the biggest sources of wasted effort, blind rerolling, poor tool-task fit, and embarrassing artifacts that slip through.
How do I run a fast defect scan?
View the image at full resolution and deliberately check the predictable failure points in order: hands and fingers, eyes, any text, facial symmetry, and repeated or duplicated patterns. It takes about thirty seconds and catches the systematic flaws that thumbnails hide.
Can I skip phases for casual work?
Yes. For high-volume, low-stakes content, run a lighter version, concept fit, negative prompt, a batch, and a defect scan. Reserve the full twenty-one-item run for high-stakes images where quality and consistency genuinely matter.
Why is saving the recipe on the checklist?
Because a library of proven recipes is the single biggest long-term productivity gain. Without it, every project restarts from zero. Recording prompt, negative prompt, seed, and parameters means your next similar image starts from a tested baseline.
Does this checklist work across different tools?
Yes. The items target the workflow, defining intent, structured prompting, controlled iteration, local fixes, and defect scanning, which transfer across nearly every diffusion-based generator. Only the specific buttons differ between tools.
Key Takeaways
- Define intent and confirm concept fit before generating anything
- Lead prompts with the subject, layer details, and always use a negative prompt
- Generate batches, pick an anchor, then lock the seed and change one variable at a time
- Fix defects locally with inpainting and upscale only after locking composition
- Run a full-size defect scan before shipping and save every working recipe
- Match the rigor of your run to the stakes of the job