McAfee-Secured Website

Exam Code: AI-103

Exam Name: Developing AI Apps and Agents on Azure

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: Azure AI Apps and Agents Developer Associate

100% Microsoft AI-103 Guaranteed Success With Testking Training Materials

Get Real AI-103 Exam Questions & Expert Verified Answers as Experienced on the Real Exam!

187 Questions & Answers with Testing Engine

"Developing AI Apps and Agents on Azure Exam", also known as AI-103 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date AI-103 Exam Engine. Your AI-103 training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

Microsoft AI-103 Practice Test 39

Question 380

A catalog team needs many color-background variants without altering the original product silhouette. Which variation policy is appropriate?

  1. Mask background edits and compare the original product silhouette.
  2. Generate each variant solely from text and accept any product shape.
  3. Apply unmasked global reshaping to maximize novelty.
  4. Validate only output file count regardless of visual content.

Explanation

Correct Answer

A:

Limiting transformations protects fixed product geometry while allowing intended background diversity.

Incorrect Answers

B:

Unanchored generation can change the advertised merchandise.

C:

Full-image changes conflict with the silhouette requirement.

D:

Quantity alone does not establish preserved product identity.

Question 381

A web designer requires a new icon with transparency, while another task must replace the background of an existing image. Which distinction matters?

  1. Use transparency-capable generation for a new asset and masked editing for replacing an existing background.
  2. Apply only a background-removal step to a nonexistent source icon.
  3. Use full text-to-image generation without the product reference for the background replacement.
  4. Assume a white RGB backdrop equals a transparent alpha channel.

Explanation

Correct Answer

A:

Creating an alpha background differs from transforming pixels in a supplied source image.

Incorrect Answers

B:

There is no existing image to separate from its background.

C:

Ignoring the reference risks altering content intended to remain.

D:

White pixels are opaque unless actual transparency is encoded.

Question 382

A seasonal campaign needs ten ads featuring the same approved mascot. Independent prompts gradually change its facial features. What should production use?

  1. Rotate prompt seeds without supplying consistent approved visual references for the mascot.
  2. Verify only the slogan spelling and ignore mascot features.
  3. Use consistent approved mascot references across all generated image variants.
  4. Remove the mascot from all references and expect textual descriptions to suffice for exact identity.

Explanation

Correct Answer

C:

Reference anchoring and comparative QA help preserve a recognizable subject through multiple outputs.

Incorrect Answers

A:

Changing the seed does not constrain character identity, causing the features to drift between variants.

B:

Correct text does not establish visual identity consistency.

D:

Text often lacks enough geometric detail for fidelity.

Question 383

A product image is compositionally correct but its lighting should match an approved watercolor campaign style. The objects must not be repositioned. Which adjustment is preferable?

  1. Rewrite the prompt to add and remove major objects as the way to change painting style.
  2. Use approved image style presets while preserving required composition.
  3. Discard the source and generate a different product category.
  4. Use OCR to replace all visible objects with extracted words.

Explanation

Correct Answer

B:

Style controls can alter appearance without necessarily changing semantic scene requirements.

Incorrect Answers

A:

Semantic content edits risk altering the approved composition.

C:

Changing subject matter does not satisfy a style-only request.

D:

Text extraction is not a visual styling workflow.

Question 384

An image generator refuses a request under its content policy. The application should not silently retry with a weaker filter. What behavior is appropriate?

  1. Honor generation content-filter refusal and offer a compliant revision.
  2. Change the prompt covertly to evade safeguards while keeping the same unsafe objective.
  3. Claim an image was generated successfully even though the service rejected it.
  4. Disable every safety threshold after the first refusal.

Explanation

Correct Answer

A:

A content-filter outcome must not be bypassed by automatic weaker-policy retries.

Incorrect Answers

B:

Circumvention undermines the configured content policy.

C:

False success misleads users and downstream consumers.

D:

A single rejection does not justify removing production protections.

Question 385

A medical advertiser wants to modify a patient's clinical image for promotional use. The result could misrepresent outcomes. What approval should precede release?

  1. Accept model self-confidence as proof the edit is medically faithful.
  2. Publish immediately if the image looks professional and clear.
  3. Request human review for medical-advertising image edits.
  4. Remove the original image so reviewers cannot compare alterations.

Explanation

Correct Answer

C:

Sensitive medical imagery requires oversight beyond automated aesthetic quality checks.

Incorrect Answers

A:

Generated confidence is not clinical verification.

B:

Visual polish cannot establish medical accuracy or consent.

D:

Losing source evidence makes manipulation harder to detect.

Question 386

A masked editing operation changes everything except the intended logo patch. What likely issue should the developer check?

  1. Increase export resolution as the first fix for wrong-region changes.
  2. Regenerate the complete image from the textual prompt rather than correcting the mask polarity.
  3. Inspect mask polarity/inversion and the region actually selected for modification.
  4. Change the text caption so users will not notice the patch error.

Explanation

Correct Answer

C:

An inverted mask can select the protected area rather than the desired editable region.

Incorrect Answers

A:

More pixels do not correct mask selection.

B:

Full regeneration is not the needed correction for a localized inverted mask.

D:

Caption wording cannot correct pixels changed outside the intended mask.

Question 387

A generated poster contains text that looks almost like an approved legal disclaimer, but one word is misspelled. How should the release pipeline handle it?

  1. Accept near-matches because the model's general intent was correct.
  2. Check only the prompt text and assume the rendered pixels reproduce it exactly.
  3. Downsample the image and treat lower resolution as proof its rendered text matches the approved copy.
  4. Check that generated text in an image matches the approved wording.

Explanation

Correct Answer

D:

Image generation can distort letters; mandatory notices require exact text QA.

Incorrect Answers

A:

Legal or brand copy may require exact wording, not semantic approximation.

B:

The output may differ from the textual instruction.

C:

Resolution changes cannot validate lettering or spelling in generated imagery.

Question 388

A video edit introduces flicker in a person's coat between consecutive frames. The first and last frames look fine. Which QA is needed?

  1. Compare video frames for temporal consistency after editing.
  2. Validate only the thumbnail frame before release.
  3. Measure audio volume as evidence of stable colors.
  4. Compare video container metadata but not rendered intermediate frames for visual continuity.

Explanation

Correct Answer

A:

Frame-by-frame or sampled sequence review exposes transient artifacts hidden at endpoints.

Incorrect Answers

B:

A thumbnail cannot show flicker during motion.

C:

Audio levels have no causal relationship to visual continuity.

D:

File metadata will not expose temporal artifacts visible between generated frames.

Question 389

A designer chooses an image-generation model that is not deployed in the organization's allowed Azure region. What should be verified before provisioning?

  1. Assume any general chat model in the region has image generation enabled.
  2. Select a supported generation deployment in the organization's region.
  3. Change the name of the selected unsupported model to a regional-sounding label.
  4. Route to a forbidden geography while tagging files with the approved country.

Explanation

Correct Answer

B:

Regional product availability and residency requirements determine whether inference can be used.

Incorrect Answers

A:

Capabilities differ by model family and region.

C:

A name cannot provision missing regional support.

D:

Metadata does not change processing location.

Question 390

A retailer supplies a photo of a specific chair and asks for a new background while preserving its upholstery and legs. What input design helps?

  1. Choose a multimodal prompt with a reference image for visual consistency.
  2. Use an audio-only input instead of the photograph.
  3. Describe the chair only as 'furniture' with no reference asset.
  4. Generate multiple approximate product variants and choose by aesthetics without comparing the approved reference.

Explanation

Correct Answer

A:

Multimodal reference grounding constrains generation to the actual source appearance.

Incorrect Answers

B:

Speech evidence does not encode the requested visual details.

C:

A generic prompt loses the unique upholstery and silhouette.

D:

Visual resemblance alone cannot guarantee preservation of the particular supplied product.

Developing AI Apps and Agents on Azure (AI-103) Practice Tests