Dayward AIWeChat Mini Program
Adobe Firefly Creative Assets in 5 Days: From Brief to Delivery Review
Week 1 · D2🔒 LockedAbout 3 hours

Prompt Images and Review Style References

Translate the brief into comparable image prompts, use permitted style references, and select candidates for message, composition, and consistency.

Today's goals 0/3

Sign in to tick these off and save your progress.

Today's Goals

  1. Write prompts with subject, setting, composition, and constraints
  2. Distinguish a style reference from a composition reference
  3. Compare candidates with a scorecard rather than taste alone

Plain-Language Walkthrough

A prompt is a testable proposal

D1 gave HL-26 a bounded job: evoke neighbors making art together in deep teal, warm coral, cream, and paper-craft forms, without unverified event facts or unapproved people. D2 turns that brief into candidate images. It does not treat “beautiful” as a pass condition. A useful prompt names the subject, action, setting, composition, light or texture, reserved text region, and exclusions. The exclusions are about what the reviewer will reject, not a promise that a model can reliably avoid every failure. A prompt log records the exact wording and observable settings so a colleague can understand why one candidate was preferred.

Begin with the intended square tile. One possible prompt describes empty art tables with folded paper shapes, soft afternoon light, deep-teal surfaces, warm-coral accents, and a cream blank region on the right for human-set type. It asks for no readable lettering, logos, or identifiable faces. That is more inspectable than “make a vibrant festival poster.” The latter invites invented event words and uncontrolled visual motifs. It also fails to tell the team where final copy belongs. A second prompt might keep the same palette and subject but move the blank area to the top. Comparing those two versions can teach whether the space works for the planned layout.

Do not mistake a natural-language constraint for enforcement. If a generated image includes a sign, face, or implausible hand, record it as a failure and decline that candidate. If the result invents “Harbor Lantern 2026” inside the picture, do not call the lettering harmless because the event is fictional. The training lesson is that final facts must come from a verified facts sheet and an editable human layout. The model’s image is a candidate visual layer, not the campaign’s source of truth.

Sign in and link WeChat to read

Sign in, then link your WeChat mini program account on the account page. Linked accounts share one free article per day. Unlock additional articles with rewarded videos in the mini program.

Mini program QR code

Scan in WeChat to open Dayward AI. Open the same course and chapter, choose the rewarded video, then refresh this page after completion.

Interview questions

  • How do style and composition references differ?风格参照与构图参照有什么区别?
    Common in ChinaCommon overseasIntermediate#firefly#review

    🔒 Unlock this chapter to see the answer keys.

  • How do you decide whether a candidate fits the brief rather than merely looks attractive?如何判断候选图符合简报而不只是好看?
    Common in ChinaCommon overseasIntermediate#firefly#review

    🔒 Unlock this chapter to see the answer keys.

  • Why change only one or two prompt variables at a time?为什么一次只改一两个提示变量?
    Common in ChinaCommon overseasIntermediate#firefly#review

    🔒 Unlock this chapter to see the answer keys.

Comments