AI-103 Implement computer vision solutions: 169 practice questions

12 of the 169 Implement computer vision solutions questions in the Certsqill AI-103 bank, shown in full below. Each one carries an explanation for every option, not just the correct one — the wrong answers are where the marks go.

Preparing for AI-103? Take the free 5-min readiness check →

1. Use a supported image-editing model with a mask covering: Which implementation should the developer choose?

1. Use a supported image-editing model with the full photograph as reference, requesting a regenerated storefront.

   This is a real editing workflow, but regenerating the storefront changes a broader area than the requirement permits.
2. Use a supported image-generation model with the photograph as reference, prompting it to recreate the entire storefront.

   Reference-image generation can guide output, but recreating the entire storefront does not target only the damaged sign.
3. Use a supported image-editing model with a mask covering only the damaged sign, then review the returned image. ✓

   A mask defines the selected editing region, while review validates the generated result and preserves responsible media handling.
4. Use a supported image-generation model with an image prompt, requesting a new storefront that resembles the photograph.

   Generating a similar new image neither selects the damaged region nor reliably preserves the existing scene details.

Use mask-based editing when the business requires a localized change to an existing image.

2. Select an image-generation model explicitly supporting: Which action satisfies the requirement?

1. Run image safety filtering on the prompt and output, then resubmit the same reference image and mask.

   Safety filtering is an appropriate media-pipeline control, but it does not change the selected model's generation-only capability into image editing.
2. Select an image-generation model explicitly supporting mask-based editing, then submit the prompt, reference image, and mask. ✓

   A model supporting the requested editing modality is the missing prerequisite. The mask can then identify the region to replace, subject to model-specific constraints.
3. Keep the selected model and refine the prompt to describe the desired replacement while retaining the supplied reference image and mask.

   Prompt refinement can improve instructions, but it cannot add image-editing capability to a generation-only model. The required modality remains unsupported.
4. Apply a watermark and branding policy to the generated output, while retaining the supplied prompt, reference image, and mask.

   Watermarking and branding address governance and provenance signaling; they do not enable mask-based editing or selected-region replacement.

Use a model that explicitly supports image editing and mask-based workflows; prompts, masks, and safety controls cannot provide missing model capability.

3. Add image-aware unsafe-content filtering to generated: Which targeted correction best addresses this finding?

1. Add a text-only prompt filter before sending requests to the image-generation model.

   Prompt filtering can address some unsafe or injection-related text, but it does not reliably inspect unsafe content produced in image pixels.
2. Add image-aware unsafe-content filtering to generated outputs before displaying them to users. ✓

   This directly addresses unsafe images that pass despite clean prompts by applying safety controls to the actual generated-media pipeline.
3. Require reference images and mask-based editing for every generation request before returning results.

   Reference images and masks control editing workflows, but they do not constitute unsafe-image filtering for generated outputs.
4. Add a visible watermark to every generated image before publishing it to users.

   Watermarks provide provenance signaling and support governance, but they do not prevent unsafe images from reaching users.

Inspect generated media with image-aware safety controls before user display; prompt-only filtering and watermarks address different concerns.

4. Provide a mask identifying the background region: What should the developer add?

1. Provide a mask identifying the background region for the supported image-editing workflow. ✓

   A mask specifies the region to edit, distinguishing selected-region editing from regenerating the entire image. The model’s editing constraints must still be validated.
2. Apply a watermark to the generated output before returning it to the product app.

   Watermarking is a governance or provenance measure; it does not control which image region the editing model changes.
3. Provide another reference image showing the desired background appearance and retain the existing prompt.

   Reference images can guide generation or editing, but adding one does not identify precisely which existing pixels may change. The explicit selected-region requirement calls for a mask.
4. Expand the prompt with detailed instructions describing the background boundaries and unchanged objects.

   A more detailed prompt can communicate intent, but it does not provide the mask-based region control required to target a selected area reliably.

Add a supported mask that identifies the region to edit; prompts and references alone do not provide selected-region control.

5. Select a supported image-editing model: Which implementation should you choose?

1. Apply a watermark after generation and retain the existing prompt, allowing the model to produce another complete image.

   A watermark provides provenance signaling and does not perform localized image editing. Generating another complete image leaves the demonstrated logo and product-change problem unresolved.
2. Add the desired sky description to the prompt and use a supported image-generation model to create a new composition.

   Prompting a new composition is a real generation workflow, but it does not preserve the existing product, logo, and unrelated pixels as required. It also lacks a region-selection mechanism.
3. Select a supported image-editing model, provide the original image and a mask for the selected region, then prompt the replacement. ✓

   A mask identifies the region targeted for editing, while the original image supplies the surrounding context. This directly addresses the requirement to change only the sky; model-specific editing constraints and the output still require validation.
4. Use a supported image-generation model with the original image as a reference, and prompt it to regenerate the complete composition.

   A reference image can guide generation, but complete regeneration does not target only the sky and may alter the logo or product details, as the evaluation already demonstrated.

Use mask-based editing with the original image and a supported editing model to target only the sky.

6. Add a multimodal validation stage for generated images: Which implementation step best satisfies these require

1. Add a multimodal validation stage for generated images, inspect embedded text, apply watermarking, and route flagged outputs for review. ✓

   This applies safety validation and governance controls to the actual media pipeline, while watermarking provides provenance signaling and review handles flagged results.
2. Filter only the text prompt before generation, then publish the generated image without additional media inspection or provenance marking.

   Text filtering does not necessarily inspect image pixels or embedded text, and this action omits the required provenance control.
3. Use a mask to regenerate suspected unsafe regions, then release the revised image without a separate safety validation.

   Mask-based editing is a supported editing approach, but regeneration alone does not establish that the revised output is safe or governed.
4. Apply a visible watermark to every generated image and approve publication when the watermark is present.

   Watermarking supplies provenance signaling but does not replace unsafe-image filtering or validation for embedded-text prompt injection.

Validate generated media multimodally, inspect embedded text, apply provenance marking, and review flagged outputs.

7. Use a supported image-editing model with the reference: Which implementation best meets the requirement?

1. Use a supported image-generation model with the reference image, then prompt it to preserve the unchanged background.

   A reference image can guide generation, but prompting preservation does not select the damaged region as precisely as a supported mask-based edit.
2. Use a supported image-editing model with the reference image and mask covering only the damaged label area. ✓

   A mask targets the selected region while the reference image supplies context. This directly addresses localized editing and requires a model supporting that workflow.
3. Add a watermark and run text-based safety filtering after generating a replacement image.

   Watermarks support provenance signaling and filtering supports safety controls; neither performs the required localized image repair or preserves pixels.
4. Use a supported image-generation model to regenerate the complete image from a detailed product prompt.

   Whole-image regeneration is an available generation action, but it conflicts with the requirement to preserve the existing product and background.

Use a model that supports image editing and apply a mask only to the damaged label region.

8. Pin the supported editing model: Which reproducible rollout step meets all requirements?

1. Supply the product reference image without a region mask, then manually review generated backgrounds before selecting results for publication.

   Reference input can support image editing, but omitting a mask does not explicitly select the background region required by the workflow.
2. Use the prototype text-to-image prompt without a mask, then watermark each generated campaign image before publication.

   Text-to-image generation and watermarking are real actions, but omitting a mask does not implement localized background editing or reliably preserve the product.
3. Regenerate the complete image with the supported model, inspect prompt text for unsafe terms, and add campaign branding afterward.

   Whole-image regeneration changes the requested editing scope, while inspecting prompt text alone does not validate the generated media pipeline.
4. Pin the supported editing model, submit the product reference and background mask, and validate output safety and provenance before release. ✓

   The supported model, reference image, and mask target the selected region. Output validation addresses media safety and provenance before publication.

Use a supported editing model with a reference image and mask, followed by media-safety and provenance validation.

9. Inspect generated images for safety: Which change is required?

1. Use a reference image and stricter prompt instructions, then publish whenever generation completes successfully without inspecting the resulting media.

   Reference media and stronger instructions can guide generation, but successful generation does not demonstrate safe output. The image still requires applicable safety validation.
2. Filter unsafe content only from the original prompt, then watermark and publish the generated image without inspecting its visual content.

   Text-only filtering can miss unsafe visual content and embedded-text prompt injection. A watermark signals provenance but is not an image-safety control.
3. Mask the entire image before editing, then repeat the original text filter before publication leaving generated-media inspection unchanged.

   Masks control selected editing regions; they do not inspect image content. Repeating prompt filtering leaves the observed output-inspection gap unresolved.
4. Inspect generated images for safety, quarantine flagged media, and watermark each image only after it passes approval. ✓

   The pipeline must inspect the generated image, not only its prompt. Quarantine supports review, while watermarking follows approval and does not replace safety validation.

Inspect generated media itself, quarantine unsafe results, and use watermarking only as governance or provenance support.

10. Use an editing-capable model with the product reference: Which implementation should the developer choose?

1. Use an editing-capable model with the product reference image and a mask covering only the background region. ✓

   A mask-based editing workflow targets the selected region while using the reference image as input. The model and editing capability must support this workflow, and outputs still require validation because exact preservation is not guaranteed.
2. Apply a provenance watermark to the approved product photo, then regenerate its background from the original prompt.

   Watermarking provides provenance signaling and does not constrain image editing. Regenerating from the prompt still lacks a mask-based operation for changing only the background.
3. Send the reference image and prompt through text-only safety filtering before generating the replacement background.

   Safety filtering may address some prompt or content risks, but it is not an image-editing mechanism and does not select the background region. Controls must match the actual media pipeline.
4. Use an image-generation model with the product reference image and prompt the model to preserve the foreground.

   A reference image and preservation instruction can guide generation, but this does not explicitly constrain the edited region. It may regenerate the whole image rather than reliably changing only the background.

Use a supported editing model with the reference image and a mask covering the background. Validate every generated result.

11. Add a mask identifying the label region and use: Which implementation change best meets the requirement?

1. Add a mask identifying the label region and use the model’s supported image-editing workflow. ✓

   A mask defines the selected region for a supported editing workflow, reducing the requested edit’s scope. The model’s documented constraints and output still require validation.
2. Apply an unsafe-image filter after generation and retain the existing unmasked editing request.

   Unsafe-image filtering addresses media safety, not spatial edit control. Retaining an unmasked request leaves the observed cause of the unwanted background changes unresolved.
3. Keep the reference image, strengthen the prompt, and request a complete image regeneration.

   A stronger prompt and reference image can guide generation, but complete regeneration does not explicitly constrain the change to the label. It fails the selected-region requirement.
4. Add a watermark to the output and use the reference image without changing the editing request.

   Watermarking provides provenance signaling and does not control which image region is regenerated. The unmasked request can still alter the surrounding package.

Use a mask with a model-supported editing workflow when the requirement targets one selected image region.

12. Compare the edited pixels with approved logo: What additional implementation is required?

1. Check watermark metadata and approve the image when provenance signaling is present, without comparing its branding or identity with approved references.

   A watermark can provide provenance signaling, but it does not verify factual accuracy, safe content, or faithful reproduction of branding and identity.
2. Run unsafe-content filtering on the prompt again before publishing the already filtered image, without validating the generated pixels.

   Prompt filtering can address text-based input concerns, but it does not validate whether generated pixels preserve required logos, text, or identity.
3. Compare the edited pixels with approved logo, text, and identity references, and publish only after those required elements pass validation. ✓

   Reference images and masks guide editing, but they do not guarantee exact reproduction. The actual output must be compared with approved references.
4. Regenerate the complete image from the prompt instead of validating the selected edited region, then publish the regenerated result when it has a watermark.

   Whole-image regeneration changes the workflow and still does not establish exact logo, text, or identity reproduction. The selected region requires output validation.

Validate the rendered output itself because reference inputs and safety controls do not guarantee exact branding or identity reproduction.

Other AI-103 domains

• Implement generative AI and agentic solutions — 429 questions →
• Plan and manage an Azure AI solution — 364 questions →
• Implement text analysis solutions — 169 questions →
• Implement information extraction solutions — 169 questions →
• All 1300 AI-103 questions →