AI for Everyone study guide · Lesson 5 of 14
Make with Multiple Media
Prompt an Image · Choose the Canvas · Plan Video as Shots · Work with Audio
This is the lesson's study guide from the FlashFeed app, plus its key facts and a self-test.
What you'll be able to do
- Prompt and review image, video, and audio systems with attention to composition, continuity, consent, and fit.
Prompt an Image
An image prompt works like a compact creative brief. Name the subject and action, then setting, composition, camera viewpoint, light, color, medium, mood, and anything that must be absent. Put the most important visual relationship in plain language rather than a pile of aesthetic keywords.
Generate, inspect, and revise. Change one dimension—framing, palette, subject placement, or texture—while preserving the rest. For identity or product consistency, use reference images where the tool supports them and check permissions for every reference.
Three ideas to retain:
- Start with the Visual Subject: State what must appear, what it is doing, and the relationship among important elements before adding style language.
- Composition Directs Attention: Viewpoint, framing, scale, subject placement, depth, and negative space determine how the eye moves through an image.
- Iteration Beats Keyword Piles: Inspect a result and revise a specific visual dimension. Many fashionable adjectives can compete without clarifying the composition.
Key term: Image subject means “the main person, object, or scene.”
Choose the Canvas
Aspect ratio is width divided by height. A square canvas supports balanced product or profile compositions; portrait ratios serve posters and phone screens; landscape ratios fit banners, presentations, and wide scenes. The destination should determine the canvas before you refine the composition.
Changing ratio is not neutral cropping. A subject centered in a square may need to move in a wide banner to leave space for a headline. Generate for the target ratio when possible, protect essential details from edges, and preview the asset inside the real layout.
Three ideas to retain:
- Aspect Ratio Shapes Composition: The width-to-height relationship changes where subjects, motion, text space, and visual balance can fit.
- Design for the Destination: Choose the canvas for the feed, slide, poster, thumbnail, or banner where the asset will actually appear.
- Cropping Can Remove Meaning: Converting one ratio to another may cut faces, hands, products, or context. Preview each required crop rather than assuming it is safe.
Practice by sketching the target placement before generation. Choose the destination, select its canvas, place the subject and negative space, then preview the result in the actual layout instead of trusting an isolated image.
Plan Video as Shots
Video adds time. A useful prompt describes not only what a frame contains but what changes: camera movement, subject action, duration, pacing, and the transition into the next shot. A storyboard or shot list makes those decisions visible before generation.
Continuity remains hard. Lock character details, wardrobe, props, lighting, and screen direction; reuse reference frames when available. Generate short shots, select the best takes, and edit them together rather than asking one long generation to solve every beat.
Three ideas to retain:
- Video Prompts Describe Change: Action, camera movement, duration, pacing, and transitions matter because the output unfolds over time.
- Storyboards Externalize the Plan: A sequence of frames or shot descriptions lets you inspect narrative beats and composition before generating footage.
- Continuity Needs Deliberate Control: Characters, props, lighting, geography, and direction can drift between clips unless they are specified and checked.
Key term: Shot list means “an ordered plan of camera setups.”
Work with Audio
Audio systems can transcribe speech, separate speakers, translate, synthesize a voice, generate music, or create sound effects. Each task has different quality checks. A transcript needs names, numbers, and speaker turns reviewed; a synthetic voice needs explicit authority and disclosure appropriate to its use.
Voice can function as identity. Short samples may support convincing impersonation, so never assume familiar sound proves a caller is genuine. Obtain consent for cloning, restrict access to voice assets, and verify urgent requests through a trusted second channel.
Three ideas to retain:
- Audio Tasks Are Not Interchangeable: Transcription, speaker labeling, translation, voice synthesis, music, and effects use different inputs and require different reviews.
- Transcripts Need Human Checks: Names, numbers, technical terms, accents, noise, and overlapping speakers are common places for speech recognition to fail.
- A Voice Can Be an Identity Signal: Synthetic speech can imitate a person, enabling accessibility and creativity as well as impersonation and fraud.
Practice by sorting the six named moves in this section, then choose one real task and explain which move fits it. For consequential work, keep the evidence, permissions, and final judgment with an accountable person.
Sources
Key terms
- Creative brief
- A concise statement of a visual project's goal, subject, audience, and constraints.
- Storyboard
- A sequence of images or panels that plans how a video will unfold.
Key concepts
Transcription
- Turns recorded speech into written text.
- Names and numbers need special review.
- Its output is a transcript.
Diarization
- Works with recordings containing multiple speakers.
- Answers the question of who spoke when.
- It labels speaker turns.
Speech translation
- Changes spoken language for another audience.
- It combines recognition and translation.
- Its goal is speech in another language.
Voice synthesis
- Produces spoken audio from text or controls.
- It may use a generic or imitated voice.
- Consent matters when identity is copied.
Music generation
- Creates rhythmic and tonal audio.
- Prompts may name instrumentation, tempo, and mood.
- Its output is music.
Sound-effect generation
- Creates short nonmusical audio events.
- Prompts name source, environment, and duration.
- Footsteps or a door close are examples.
Key facts
Composition Directs Attention
Viewpoint, framing, scale, subject placement, depth, and negative space determine how the eye moves through an image.
Iteration Beats Keyword Piles
Inspect a result and revise a specific visual dimension. Many fashionable adjectives can compete without clarifying the composition.
Design for the Destination
Choose the canvas for the feed, slide, poster, thumbnail, or banner where the asset will actually appear.
Cropping Can Remove Meaning
Converting one ratio to another may cut faces, hands, products, or context. Preview each required crop rather than assuming it is safe.
Storyboards Externalize the Plan
A sequence of frames or shot descriptions lets you inspect narrative beats and composition before generating footage.
Continuity Needs Deliberate Control
Characters, props, lighting, geography, and direction can drift between clips unless they are specified and checked.
Transcripts Need Human Checks
Names, numbers, technical terms, accents, noise, and overlapping speakers are common places for speech recognition to fail.
A Voice Can Be an Identity Signal
Synthetic speech can imitate a person, enabling accessibility and creativity as well as impersonation and fraud.
By the numbers
- Square canvas
- 1 width ÷ height
- Portrait 3:4 canvas
- 0.75 width ÷ height
- Landscape 4:3 canvas
- 1.333 width ÷ height
- Vertical 9:16 canvas
- 0.5625 width ÷ height
- Widescreen 16:9 canvas
- 1.778 width ÷ height
- Classic photo 3:2 canvas
- 1.5 width ÷ height
Matched pairs
Each item on the left goes with the one on the right.
- Image subject
- The main person, object, or scene
- Negative prompt
- A description of elements to avoid
- Shot list
- An ordered plan of camera setups
- Continuity
- Consistency of visual details across shots
Step by step
Revise a generated image
- Identify the one visual problem that matters most
- Name the desired change in composition, light, color, or subject
- Preserve elements that already work
- Generate a small set of variants
- Check details, rights, and intended use before publishing
Keep the revision specific enough that you can tell whether it succeeded.
Plan a generated video sequence
- Write the audience and single message
- Break the message into visual beats
- Specify each shot's subject, action, camera, and duration
- Lock continuity details and reference frames
- Generate short takes, review, and edit
Sound, captions, rights, and accessibility belong in the production plan too.
Self-test
Answer each question from memory, then check your work against the answer key.
1. What should an image prompt name first?
- ATwenty style buzzwords
- BA random camera brand
- CAn invisible watermark
- DThe essential subject
2. A portrait feels cramped. Best revision?
- ATighten the camera framing
- BAsk for more negative space
- CIncrease text length
- DAdd more art styles
3. Which canvas is tallest relative to width?
- A1:1 square
- B4:3 landscape
- C9:16 vertical
- D16:9 widescreen
4. What does a video prompt add beyond a still?
- AChange over time
- BA single aspect ratio only
- CGuaranteed continuity
- DHigher image resolution
5. Best way to control a long sequence?
- AIgnore screen direction
- BUse one vague prompt
- CRemove all references
- DGenerate and edit short shots
Go deeper
- WatchNot a Neutral Crop A product sits centered in a square image. In a wide banner, that product may need to move to leave space for a headline.
- StudyAI for Everyone in the FlashFeed app
Answer key
- 1. D The essential subjectThe required visual content should lead the brief.
- 2. B Ask for more negative spaceNegative space directly addresses the framing problem.
- 3. C 9:16 verticalNine-by-sixteen has the smallest width-to-height value in this set.
- 4. A Change over timeMotion and duration are defining dimensions of video.
- 5. D Generate and edit short shotsShort, planned clips make selection and continuity checks easier.
Sources
- OpenAI Academy — Image generation — openai.com/academy/image-generation/
- Google Cloud — Configure image aspect ratio — cloud.google.com/vertex-ai/generative-ai/docs/image/configure-aspect-ratio
- Rombach et al. — High-Resolution Image Synthesis with Latent Diffusion Models — arxiv.org/abs/2112.10752
- Radford et al. — Robust Speech Recognition via Large-Scale Weak Supervision — arxiv.org/abs/2212.04356
- FTC — Approaches to address AI-enabled voice cloning — www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/04/approaches-address-ai-enabled-voice-cloning
- Google Cloud — Generate videos with Veo — cloud.google.com/vertex-ai/generative-ai/docs/video/generate-videos
- Google Cloud — Speaker diarization — docs.cloud.google.com/speech-to-text/docs/multiple-voices