Distinguish a single image from an interactive service.
Suggested reading time: about 8 minutes. This is an editorial production guide. Generation, performance and pricing comparison experiments have not been run.
What you will learn
- Explain the different kinds of deliverables.
- Connect each production stage with the people responsible for it.
- Break work assigned to AI into small tasks.
See deliverables as four layers
An asset is a reusable component such as a PNG image, sound or 3D model. A scene describes where those assets are placed and the lighting and camera used to view them. A runtime is an application that changes the scene in response to interaction or time as it runs. A video is a sequence of images and sound exported in a predetermined order. Even when each depicts the same ocean, a background image, a 3D ocean scene, an interactive ocean web app and an ocean video require different technologies and different deliverables.
Read the process before the industry label
Advertising, games, animation, film, social media and web services use different job titles and divide work at different levels. They can still be organized around the same flow: define the purpose → explore ideas → create assets → add movement → finish → publish → observe the response. In an individual project, one person may fill several roles. AI can assist at each stage, but it does not automatically make the right decision about the audience, the message, the selected idea or whether the work is ready to publish.
Turn adjectives into delivery requirements
“A cute 3D character” does not tell the maker whether you want an image or a model that can be rotated. Specify the use and acceptance criteria instead: “For an empty state in a service; transparent PNG; two versions, front-facing and happy; the expression must remain readable at a small size.” A 3D-style image is a 2D image with shading that suggests depth. It does not automatically provide an editable 3D model.
Choose a learning path
Four layers of deliverables
| Element | Meaning | Example | Distinction to retain |
|---|---|---|---|
| Asset | A component used in other production work. | PNG, texture, GLB, sound effect. | A 3D-style PNG is not a rotatable 3D model. |
| Scene | Assets combined with placement, light and a camera. | A space containing a character, floor, lights and camera. | A scene alone does not complete interaction or publication. |
| Runtime | A system that runs on a device and responds to input and time. | A web app with adjustable wave height. | A recorded video does not let the user choose another viewing angle. |
| Video | Images and sound exported in a fixed time sequence. | An MP4 presenting a project. | The export may not include its source scene or editable data. |
How 3D becomes an image
| Element | Meaning | Example |
|---|---|---|
| Geometry | The shape of the object. | A sphere, character mesh or subdivided ocean plane. |
| Material | How a surface should appear. | Matte, metal or water. |
| Texture | Image or numeric data that varies by location. | Color, normals and roughness; it need not be a photograph. |
| Shader | A program that calculates positions, colors and other values on the GPU. | Moving vertices up and down with time or calculating surface reflections. |
| Light | Where illumination comes from and how it lights the scene. | Sunlight, an area light or environmental light. |
| Camera | The viewpoint and framing. | Position, field of view and orientation. |
| Render | Calculate an image from the settings. | One real-time frame or an exported still image. |
| Composite | Combine images or layers to finish the result. | Background compositing, color adjustments, glow or subtitles. |
Increase control in AI production
| Element | Meaning | Example |
|---|---|---|
| Text | Specify meaning and constraints. | Subject, purpose, composition and color. |
| Reference | Provide clues to appearance or structure. | A character reference, a pose or assets you have permission to use. |
| Local editing | Retain selected parts and change the necessary area. | Clothing color, hands or accessories. |
| Manual work and code | Control the parts that need precision. | Typography, exact pixels, UI states or 3D geometry. |
| Verification | Check independently whether the result matches the intent. | Comparison images, an all-around inspection or measurements on the actual device. |
Work and deliverables by industry
Compare industries through their deliverables, success question, AI’s contribution and the checks a person must make.
| Field | Deliverables | Typical roles | Success question | AI use | Human review |
|---|---|---|---|---|---|
| Web and applications | Hero images; Empty-state characters; Interactive 3D; UI motion | Product designer; Front-end engineer; Motion designer | Can users understand the content and operate it? | Explore directions, sketch assets and assist with code. | Readability, state design, performance and accessibility. |
| Games | Sprites; 3D characters; Environments; Real-time VFX | Game artist; Modeler; Rigger; Technical artist | Does it still work when the viewpoint and interaction change? | Concepts, sketches and texture candidates. | Consistency, ground contact, deformation, size and behavior. |
| Animation and film | Design sheets; Shots; Compositing assets; Finished video | Director; Storyboard artist; Animator; Compositor; Editor | Do the characters and story communicate across time? | Storyboard candidates, background concepts and shot assets. | Continuity, performance, editing, sound and rights. |
| Advertising and branding | Key visuals; Product images; Short videos; Adapted assets | Art director; Designer; Copywriter; Producer | Does the message match the value and promise of the brand? | Explore directions and composition candidates, and assist with size variants. | Facts, product accuracy, presentation, rights and medium specifications. |
| Social media and individual publishing | Post images; Short videos; Thumbnails; Process explanations | Creator; Editor; Social media manager | Does the intended audience understand why it is worth watching? | Assist with planning, propose assets and draft subtitles. | Fact-checking, disclosure, provenance and interpreting audience response. |
A shared production process
| Stage | Question | Input | Output | Gate before continuing |
|---|---|---|---|---|
| Planning | Who is this for, and what should it communicate? | Purpose and constraints. | Production brief. | The use case and acceptance criteria are clear. |
| Preproduction | How should it be presented? | Brief and references. | Sketches, composition, designs and storyboards. | Agree on the direction before moving to expensive stages. |
| Asset production | Which components are needed? | Selected concept and specification. | Images, models, sound and code. | Format and quality suit each use case. |
| Assembly and motion | How do placement and timing communicate? | Assets and behavior design. | Scenes, UI and shots. | Consistency and correct behavior. |
| Finishing | Is the result clear and coherent? | Scenes or assets being edited. | Final images, videos or applications. | Typography, color, sound and performance. |
| Publication and evaluation | Did it achieve its purpose? | Finished work and usage terms. | Publication package and improvement record. | Disclosure, rights, actual measurements and the next hypothesis. |
Glossary
- Brief: A short specification covering purpose, audience, constraints, deliverables and acceptance criteria.
- Asset: Reusable material such as an image, sound or 3D model.
- Rendering: Calculating an image from geometry, materials, lighting, a camera and related settings.
- Real-time: Updating output during execution in response to user input or time.
- Offline rendering: Exporting frames while prioritizing the finished image over immediate interaction response.
- Material: Settings that determine a surface’s appearance. Some materials do not use textures.
- Texture: An image or data that assigns values to locations on a surface or elsewhere.
- Shader: A program that calculates geometry, color and other values on the GPU.
- Normal: A direction describing which way a surface faces.
- UV coordinates: Coordinates defining the relationship between a 3D surface and a 2D texture.
- Rig: A system for moving a model, such as bones and control points.
- Bake: Store a calculated appearance or another computed result in a texture or other data.
- Sprite: An image asset or its display unit, mainly used in 2D drawing.
- Keyframe: A position or value specified at an important point in time.
- Composite: Combine images or layers into the finished view.
- Inference: Pass inputs to a trained model to obtain outputs.
- Token: A processing unit handled by a model. Its conversion to characters, words, images or seconds depends on the method.
- Seed: A value used to initialize randomness. Some services do not support it, and it does not guarantee reproduction on its own.
- Prompt: Instructions or context supplied to a model, sometimes combined with conditions such as images.
- Inpainting: Recreate a specified region to fit the surrounding context.
- Latency: The time between a request and its result. It is different from video duration or billed compute time.
- Frame time: The time spent on one frame. Inspect slow frames as well as the average.
Distinguish evidence states
| State | Recorded value | Meaning |
|---|---|---|
| Measured | measured | A result with an execution date, conditions, and logs or artifacts. |
| Illustrative | illustrative | A hypothetical value or demonstration for explanation, rather than a measurement. |
| Not run | not-run | A design or exercise plan with no result yet. |
| Not collected | not-collected | The task ran, but this value was not obtained. |
Keep unknown values as null, and display them as “not collected” or “not run.” Do not treat them as zero.
Continue to the recording conditions and method in Measure quality and cost to improve the next attempt.
Work and responsibilities
- Producer: sets the purpose, budget and deadline.
- Art director: aligns the criteria for visual decisions.
- Designer or artist: creates the assets.
- Engineer or technical artist: connects assets to working behavior.
From input to output
| Input | Workflow | Output |
|---|---|---|
| The service screen or intended use case.; One sentence stating who the work is for and what it should communicate.; The display size and whether interaction is required. | 1. Choose one specific use case.; 2. Classify the deliverable as an asset, scene, runtime or video.; 3. Write down its purpose, format, number of items or duration, and the parts that must remain editable.; 4. Assign responsibilities for generation, manual work and code.; 5. Limit the completion criteria to three items. | A one-page production brief.; Notes describing the inputs and outputs of each stage. |
Quality checks
- It is clear who can open the delivered file format.
- Visual requirements and behavior requirements are specified separately.
- Success is evaluated by more than personal taste.
Diagnose failures
| Symptom | Possible cause | Correction |
|---|---|---|
| The image looks good but cannot be implemented. | The required deliverable is a different kind of artifact. | Define the required format and editable parts before production. |
| Revisions never end. | The use case and acceptance criteria are vague. | Choose and rank the three most important requirements. |
Exercise: Break down an ocean hero screen
Status: not run (not-run). This is a plan with completion criteria. No artifacts, duration, cost, performance or success rate have been recorded.
Plan three versions of an ocean home screen: a still image, a video and an interactive version. For each, write down the inputs, outputs, responsible technologies and workload or runtime load.
Deliverable: Production briefs for all three versions.
Completion criterion: You can explain why similar appearances require different implementations and different ways of measuring cost.
When you run it, separate predictions from results and record conditions and evidence using the experiment notebook.
Continue into practice
Choose a use case and deliverable in the themed template library (Catalog candidates retained for review), then use Turn visual direction into words and specifications to refine the instructions.
Technical references and evidence scope
Technical references were checked on 2026-10-03. They support the technical concepts. The exercises, workflows and evaluation criteria are editorial proposals, rather than experiments performed by the source authors. Recheck changing APIs, prices and publication rules when using them.
- Three.js Mesh — A mesh combines geometry and material.
- Khronos glTF — The role of glTF as a format for delivering 3D assets to runtimes.
Read the production flow
- 1Purpose and audience
- 2Asset specification
- 3Scene
- 4Runtime or exported video
- 5Delivery checks
This is an original conceptual process diagram; arrows show the review sequence, not a claim about an undisclosed model architecture.
MENTAL MODEL / REASONING ORDER
Change one thing at a time.
Decide what the image communicates, its medium and dimensions, and the meaning you want to preserve.
Sources
Publication dates belong to the source; access dates record when it was checked. Community observations are separate from official statements.
01