L1
·
Quiz
·
Lab
L2
·
Quiz
·
Lab
L3
·
Quiz
·
Lab
L4
·
Quiz
·
Lab
Module Test
Module 2 · Lesson 1

The Generative Image Landscape

Midjourney, DALL·E, Stable Diffusion, Firefly — mapping the tools that are rewriting visual production
Which generative image tool is right for a professional design workflow, and why does it depend on the job?

In early 2023, Heinz ran a global campaign called "A.I. Ketchup." The brand prompted several AI image generators with the phrase "ketchup" — no brand name, just the word. Every generator, across every style, defaulted to a bottle that looked unmistakably like Heinz. The campaign ran across 18 markets and earned over 1.3 billion impressions organically. The creative director cited this as proof that brand equity had become encoded into the training data of generative models themselves.

Why the Landscape Feels Overwhelming

Between 2022 and 2025 more than a dozen serious generative image platforms launched or reached maturity. Each makes slightly different tradeoffs: image quality versus speed, artistic range versus brand controllability, ease of use versus depth of customization. For designers, the instinct to pick one tool and master it is understandable — but it leaves capability on the table.

This lesson maps the major platforms, explains the architectural differences that produce different results, and gives you a professional framework for deciding which tool belongs in which part of a real project pipeline.

The Four Major Platforms
Midjourney
Discord-native, cloud-hosted. Known for high aesthetic output with minimal prompting. Closed weights. Best for editorial, mood, concept art.
DALL·E 3
OpenAI, integrated into ChatGPT. Strongest at literal prompt fidelity and text rendering within images. Best for client presentations and comps.
Stable Diffusion
Open-weight, runs locally or via API. Maximum customizability via LoRA, ControlNet, fine-tuning. Best for brand-specific or proprietary pipelines.
Adobe Firefly
Trained on licensed Adobe Stock. Commercially safe outputs with IP indemnification. Best for client deliverables requiring clean provenance.
How They Differ Under the Hood

All four use diffusion models — a process that starts from random noise and iteratively refines it toward an image matching a text description. But the training data, model architecture, and inference controls differ significantly.

Midjourney uses a proprietary model trained on a curated dataset that skews toward fine art, photography, and design. Its aesthetic quality comes partly from heavy curation of what it learned from. It gives users relatively few direct parameters but rewards experienced prompt engineers who understand its aesthetic vocabulary.

DALL·E 3 uses a GPT-4-based system to rewrite your prompt before sending it to the image model — meaning it aggressively interprets your intent. This makes it very good at literal fidelity but can override stylistic choices you make deliberately.

Stable Diffusion (specifically SDXL and SD 3.x) exposes its weights publicly, enabling the entire ecosystem of ControlNet (pose/depth control), LoRA (lightweight fine-tuning), and img2img workflows that make professional production pipelines possible.

Adobe Firefly was trained exclusively on Adobe Stock images, openly licensed content, and public domain material — giving it a unique commercial safety guarantee that competitors cannot currently match at scale.

Key Distinction

Closed-weight models (Midjourney, DALL·E) are faster to use but impossible to customise at the model level. Open-weight models (Stable Diffusion) require more infrastructure but allow brand-specific fine-tuning that closed models cannot replicate.

Key Terms
Diffusion ModelAn image generation architecture that learns to reverse a noise-adding process; at inference, it starts from noise and denoises toward a target image.
LoRALow-Rank Adaptation — a lightweight fine-tuning method that teaches a base diffusion model specific styles, faces, or objects without retraining the full model.
ControlNetA Stable Diffusion extension that conditions generation on structural inputs — pose skeletons, depth maps, edge maps — giving designers layout control over outputs.
IP IndemnificationA legal guarantee from a vendor (notably Adobe Firefly) that they will defend users against intellectual property claims arising from AI-generated outputs.
Prompt FidelityHow accurately a model translates a text prompt into a matching image; DALL·E 3 scores highest; Midjourney deliberately interprets creatively.
Designer Takeaway

No single tool wins across all criteria. A mature design pipeline in 2025 typically uses Firefly for client-deliverable assets requiring legal clarity, Midjourney for concept ideation, DALL·E for precise comps, and Stable Diffusion for bespoke brand fine-tuning. The skill is knowing when to switch.

Lesson 1 Quiz

The Generative Image Landscape — 4 questions
1. What architectural feature makes Stable Diffusion uniquely suited to brand-specific fine-tuning that closed-weight models cannot replicate?
Correct. Open weights are what make Stable Diffusion the foundation for brand fine-tuning pipelines — LoRA, ControlNet, and img2img workflows all require access to model internals that closed models never expose.
Not quite. The defining advantage is its open weights, which enable LoRA fine-tuning, ControlNet, and other deep customisations unavailable on closed models like Midjourney or DALL·E.
2. Adobe Firefly's commercial safety guarantee is distinct from competitors because its training data consists primarily of what?
Correct. Training exclusively on Adobe Stock, openly licensed, and public domain material is what allows Adobe to offer IP indemnification — a promise competitors relying on scraped internet data cannot make.
That's not correct. Firefly's legal safety comes from training on Adobe Stock, openly licensed content, and public domain material — enabling the IP indemnification guarantee that competitors cannot match.
3. DALL·E 3 uses a GPT-4-based system to rewrite prompts before image generation. What is the practical implication of this for designers?
Exactly right. GPT-4 prompt rewriting aggressively interprets intent, which produces high literal fidelity but can override subtle stylistic decisions a designer makes deliberately. Knowing this helps you work with DALL·E rather than against it.
Not quite. The key implication is a fidelity/control tradeoff: literal accuracy improves but GPT-4 may override deliberate stylistic choices you embed in your prompt.
4. The 2023 Heinz "A.I. Ketchup" campaign demonstrated which phenomenon relevant to generative image tools?
Correct. When every AI generator defaulted to a Heinz-looking bottle given only the word "ketchup," it demonstrated that Heinz's visual identity had been absorbed into model training data — a significant insight about brand dominance in the AI era.
That's not the lesson. The campaign showed that dominant brand equity gets encoded into model training data — so generators produced Heinz-style imagery without the brand name ever appearing in the prompt.

Lab 1 — Tool Selection Strategy

Discuss generative image tool selection for real design scenarios with your AI coach

Your Mission

You'll receive three real-world design scenarios. For each, work through which generative image platform (Midjourney, DALL·E 3, Stable Diffusion, or Adobe Firefly) is the best choice — and why. Your coach will push you to justify decisions based on capability, legal requirements, and workflow fit.

Start by asking for your first scenario, or describe a real project you're working on and let's figure out the right tool together.
AI Design Coach
Lab 1
Welcome to Lab 1. I'm your design coach for this session. We're going to work through generative image tool selection — one of the most consequential practical decisions in a modern design workflow.

Tell me: do you want me to give you a scenario to analyse, or do you have a real project brief you'd like to work through together?
Module 2 · Lesson 2

Prompt Engineering for Visual Output

The craft of writing prompts that produce usable, on-brief images — not beautiful accidents
What separates a prompt that generates a mood from a prompt that generates a deliverable?

When Coca-Cola used DALL·E 3 and Stable Diffusion for their 2024 "Masterpiece" campaign extensions, the creative team published internal notes describing their prompt development process. Initial prompts produced images that were aesthetically impressive but unusable — wrong aspect ratios, brand colour drift, inconsistent bottle silhouettes. The team spent three weeks developing a prompt library of 47 tested templates before achieving production-quality consistency across 200+ assets. The insight: prompt engineering at production scale is a repeatable craft, not creative inspiration.

Why Prompts Fail — and How to Fix Them

Most designers approach AI image prompts the way they'd write a creative brief to a human illustrator: descriptive, evocative, relatively loose. This produces aesthetically interesting results but rarely production-ready ones. Generative models need a different kind of input — one that balances subject specification, style parameters, technical constraints, and negative space (what you don't want).

The anatomy of a professional prompt has five layers. Mastering them is the difference between a designer who occasionally gets lucky with AI and one who uses it reliably for client deliverables.

The Five-Layer Prompt Framework
  • 1Subject + Action: What is in the image and what is it doing? Be specific. "A woman" becomes "a mid-30s South Asian woman in a tailored navy blazer, standing at a floor-to-ceiling window, looking outward." Specificity reduces model interpretation.
  • 2Environment + Lighting: Where is the scene? What is the quality of light? "Golden hour, soft diffused natural light from the left" gives the model constraints that affect realism, mood, and compositional balance simultaneously.
  • 3Style Reference: What visual language does this belong to? Reference specific photographers, art movements, film stocks, or rendering styles. "Editorial photography, Vogue aesthetic, 85mm lens, shallow depth of field" is more actionable than "professional photo."
  • 4Technical Parameters: Aspect ratio, resolution target, colour palette constraints, camera angle. For Midjourney: --ar 16:9 --style raw --chaos 0. For DALL·E: specify within the prompt text. For Stable Diffusion: CFG scale, sampler, steps.
  • 5Negative Prompts: What to exclude. "No watermark, no text, no cartoon, no distorted hands, no chromatic aberration" — particularly critical in Stable Diffusion where negative prompts have direct model influence.
Midjourney vs. DALL·E — Style Prompt Behaviour

Midjourney responds strongly to emotional and aesthetic vocabulary: "ethereal," "desolate," "brutalist opulence." DALL·E 3 responds better to structural description: "a wide-angle photograph of a concrete building at dusk, dramatic shadows, no people." Knowing which register each model responds to saves hours of iteration.

Prompt Iteration — A Production Workflow

Professional prompt engineering is iterative, not single-shot. The production workflow used by studios like BUCK and Superside typically follows three stages: Discovery (broad, high-chaos prompts to explore the solution space), Refinement (narrowing style and subject with lower chaos/temperature), and Production Lock (a fixed prompt template + seed that can generate consistent variants).

Seed locking — specifying a fixed random seed in Midjourney or Stable Diffusion — is particularly powerful for campaign work because it allows stylistically consistent images of different subjects using the same visual language.

Key Terms
Negative PromptText that tells the model what to exclude from the image; particularly effective in Stable Diffusion where it has direct denoising influence.
CFG ScaleClassifier-Free Guidance scale in Stable Diffusion; higher values make the model follow the prompt more strictly at the cost of image naturalness.
SeedA fixed random number that determines the starting noise pattern; locking the seed with a consistent prompt produces stylistically matched variants.
Chaos (Midjourney)The --chaos parameter (0–100) controls how much Midjourney deviates from the prompt; high values produce unexpected results, low values produce predictable ones.
Style Raw (Midjourney)A mode that reduces Midjourney's aesthetic processing, giving photographers and designers more direct control over the literal content of the output.
Designer Takeaway

Treat your best prompts as design assets — version-controlled, documented, and shared across your team. The Coca-Cola team's 47-template prompt library was as much a brand asset as their style guide. Build yours deliberately.

Lesson 2 Quiz

Prompt Engineering for Visual Output — 4 questions
1. In the Five-Layer Prompt Framework, what is the role of Negative Prompts, and which platform benefits most directly from them?
Correct. Negative prompts exclude unwanted elements and have direct influence on the denoising process in Stable Diffusion — more so than in Midjourney or DALL·E, where their effect is more indirect.
Not quite. Negative prompts specify what to exclude, and they have the most direct technical effect in Stable Diffusion, where they directly influence the denoising/sampling process at the model level.
2. Locking a seed in Midjourney or Stable Diffusion is particularly valuable for campaign production because it allows what?
Exactly right. A fixed seed with a consistent prompt template produces variants that share the same visual language — critical for campaign coherence when generating many different hero images that must feel like a set.
Not correct. Seed locking controls the starting noise pattern, which means the same prompt + seed produces the same stylistic foundation — enabling consistent visual language across a campaign's multiple subjects or scenes.
3. Midjourney's --style raw parameter is most useful for which type of professional user?
Correct. --style raw reduces Midjourney's tendency to apply its own aesthetic interpretation, making it respond more literally to the prompt — particularly valuable for photographers and designers with specific visual requirements.
Not quite. --style raw reduces Midjourney's aesthetic processing, giving photographers and designers more literal control. Higher --chaos values are what you'd use for creative unpredictability.
4. The Coca-Cola "Masterpiece" campaign prompt development process revealed which key insight about prompt engineering at production scale?
Correct. Three weeks and 47 tested templates to achieve production consistency across 200+ assets — this is the industrialisation of prompt engineering. It's a craft discipline, not a creative shortcut.
That's not the key insight. The key learning was that consistent, production-quality AI imagery requires systematic prompt template development over weeks — it's a craft, not a quick creative process.

Lab 2 — Prompt Engineering Workshop

Build and refine image prompts using the Five-Layer Framework with your AI coach

Your Mission

You'll practice the Five-Layer Prompt Framework by building, critiquing, and refining prompts for real design scenarios. Your coach will analyse your prompts, identify which layers are weak, and help you revise toward production quality.

Share a prompt you'd use for an image you need — or describe the image you want and I'll help you build the prompt from scratch using the framework.
AI Design Coach
Lab 2
Welcome to Lab 2. We're doing hands-on prompt engineering today — specifically using the Five-Layer Framework: Subject, Environment/Lighting, Style Reference, Technical Parameters, and Negative Prompts.

Give me either a rough prompt you've written, or describe an image you need for a real or hypothetical brief. I'll help you build it into something production-ready.
Module 2 · Lesson 3

Image-to-Image, Inpainting & ControlNet

Beyond text-to-image — the advanced controls that make generative tools genuinely useful in a production design pipeline
How do professional studios use reference images, masks, and structural controls to get consistent, composable AI assets?

When director Wes Anderson style became a dominant AI aesthetic trend in 2023, the studio Curious Refuge documented how they built full AI short films. Their key finding: text-to-image alone produced beautiful single frames but no temporal or compositional consistency. Their production pipeline used ControlNet with OpenPose to lock character body positions across frames, img2img at 0.4–0.6 strength to restyle photographic references into the target aesthetic, and inpainting to swap elements within otherwise-locked compositions. The combination turned a single-shot tool into a production-capable system.

The Limits of Text-to-Image Alone

Text-to-image is the entry point to generative tools, but it has a fundamental limitation for design work: stochastic composition. Every generation produces a random compositional arrangement within the prompt's constraints. For editorial work, this is acceptable. For campaign assets, packaging, or multi-image series requiring visual consistency, it is a blocking problem.

The three techniques in this lesson — img2img, inpainting, and ControlNet — solve this by giving designers control over composition, content, and structure that text prompts alone cannot provide.

Image-to-Image (img2img)

Img2img takes an existing image as input alongside a text prompt. The model uses the reference image's structure as a starting point, then rewrites it toward the prompt. The denoising strength parameter (0.0–1.0) controls how much the model departs from the reference: 0.0 produces an identical copy, 1.0 produces a purely text-driven result. The productive range for most design applications is 0.35–0.65.

Practical applications include: restyling a mood board photograph into an illustrated look; converting a rough sketch into a polished render; adapting a competitor's aesthetic into a new visual language; creating dark/light mode variants of the same scene.

Practical Denoising Strength Guide

0.2–0.35: Preserve structure almost entirely, subtle style change only. 0.4–0.55: Major style shift while keeping composition and subject. 0.6–0.75: Significant departure — shape and rough composition survive. 0.8+: Only faint traces of the original remain.

Inpainting

Inpainting allows a designer to mask a specific region of an image and regenerate only that region while the rest remains pixel-identical. This is one of the most practically powerful tools in the generative pipeline: it solves the "almost perfect" problem that plagues text-to-image outputs.

Common design applications: removing AI-generated hands and regenerating them (the persistent hand failure mode); swapping background environments while keeping a hero subject; replacing product labels or colours; adding a missing design element to an otherwise-locked composition; and cleaning up compositional artefacts without restarting generation.

Inpainting is available natively in Stable Diffusion (via AUTOMATIC1111, ComfyUI), in Adobe Firefly's Generative Fill, and in Photoshop's AI features from version 25.0 onwards. Midjourney added a limited inpainting tool ("Vary Region") in late 2023.

ControlNet

ControlNet conditions a Stable Diffusion generation on a structural input rather than (or in addition to) a text prompt. The structural input can be a pose skeleton (OpenPose), a depth map, an edge map (Canny), a surface normal map, or a segmentation mask. This allows designers to specify the exact spatial layout of an image before generation begins.

The most important ControlNet modes for designers: OpenPose for human figure positioning (critical for fashion, lifestyle, and character work); Canny Edge for preserving product shapes and architectural layouts; Depth for controlling spatial depth while allowing surface texture and style to vary; MLSD for architectural line preservation in interior and exterior visualisation.

Key Terms
Denoising StrengthIn img2img workflows, the parameter (0–1) controlling how much the model departs from the reference image; 0 = identical copy, 1 = pure text-to-image.
InpaintingGenerative technique that regenerates only a masked region of an image, leaving surrounding pixels unchanged.
OpenPoseA ControlNet conditioning mode that extracts a skeleton from a reference image and uses it to control the body position of generated figures.
Canny EdgeA ControlNet mode that extracts edge maps and uses them to preserve product shapes and structural outlines in generation.
Vary Region (Midjourney)Midjourney's inpainting-equivalent feature, introduced late 2023, allowing regeneration of a selected sub-region of an existing Midjourney output.
Designer Takeaway

The professional pipeline is almost never a single text-to-image generation. It's: reference → img2img (style transfer) → inpainting (fix problems) → ControlNet (lock structure for variants). Master this sequence and you control AI outputs rather than accept them.

Lesson 3 Quiz

Image-to-Image, Inpainting & ControlNet — 4 questions
1. A designer wants to restyle a lifestyle photograph into a hand-drawn illustration aesthetic while keeping the composition nearly identical. Which denoising strength range would be most appropriate?
Correct. The 0.4–0.55 range produces major aesthetic shifts while keeping the source image's compositional structure largely intact — ideal for converting photo references to illustrated looks.
Not quite. 0.4–0.55 is the sweet spot for major style shifts with composition preservation. Lower values barely change anything; higher values lose the original composition entirely.
2. Inpainting is described as solving the "almost perfect" problem. In practice, which of the following is its most common application in professional design workflows?
Exactly right. Inpainting surgically fixes specific regions — distorted hands, wrong backgrounds, misplaced elements — without disturbing the rest of the image. It's the production polish step in most professional pipelines.
Not correct. Inpainting regenerates a masked region only — its primary value is fixing specific problems (hands, backgrounds, elements) within an otherwise-acceptable image without restarting generation.
3. The Curious Refuge studio's AI film pipeline used ControlNet with OpenPose to solve which specific technical problem?
Correct. OpenPose extracts a pose skeleton and uses it to condition generation — ensuring the character occupies the same body position across different frames, solving the temporal consistency problem that pure text-to-image cannot address.
Not quite. OpenPose ControlNet locks body positions by conditioning generation on a pose skeleton — Curious Refuge used this specifically to maintain character position consistency across film frames.
4. Which ControlNet conditioning mode is most appropriate for preserving the exact outline shape of a product in a commercial packshot scenario?
Correct. Canny Edge extracts an edge map that preserves the precise structural outline of a product — essential for packshot work where the product silhouette must remain accurate regardless of style changes.
Not quite. Canny Edge is the right mode for product silhouette preservation. It extracts an edge map of the product outline and uses it to constrain the shape of the generated result.

Lab 3 — Pipeline Design

Plan a multi-step generative pipeline using img2img, inpainting, and ControlNet for a real scenario

Your Mission

You'll work through designing a complete generative production pipeline for a scenario requiring visual consistency across multiple assets. Your coach will help you decide which advanced techniques to use at each stage and why.

Describe a project where you need multiple consistent AI-generated images — a campaign, a product line, a character set — and we'll design the full pipeline together.
AI Design Coach
Lab 3
Welcome to Lab 3. We're moving beyond single-image generation into pipeline design — the sequence of tools and techniques that produces consistent, composable AI assets for real production work.

Describe a project needing multiple consistent images, and we'll map out exactly which techniques to use at each step: where does img2img apply? Where do you need inpainting? Where does ControlNet become necessary?
Module 2 · Lesson 4

Copyright, Ownership & Ethical Use

What designers actually own, what they don't, and how to stay on the right side of a rapidly evolving legal landscape
If an AI generated it, who owns it — and can you actually use it commercially?

In February 2023, the US Copyright Office issued formal guidance ruling that AI-generated images produced without sufficient human creative authorship are not eligible for copyright protection. The immediate case involved Kris Kashtanova's graphic novel "Zarya of the Dawn," created with Midjourney. The Office granted copyright to the text and overall arrangement but cancelled protection for all individual AI-generated images within it. The ruling established the human authorship threshold that now governs every commercial AI image use in the United States.

The Human Authorship Threshold

The US Copyright Office's 2023 Zarya ruling established a principle that has since been echoed in guidance from copyright offices in the EU, UK, and Australia: copyright protection requires human creative expression. Entering a text prompt and accepting a generated result does not meet this threshold. The copyright question hinges on how much human creative decision-making shaped the final output.

Factors that may establish stronger authorship claims: significant post-production editing in Photoshop or Illustrator; extensive use of img2img with a designer's own photography as reference; detailed ControlNet conditioning that reflects specific creative decisions; or substantial selection and arrangement of multiple AI elements into a designed composition.

This area of law is actively evolving and varies by jurisdiction. US guidance as of 2025 suggests that works where AI generation is merely a component of a larger human-authored design work are more defensible than works where the AI output is used without modification.

Practical Copyright Baseline

Currently: pure AI-generated images (prompt → accept → deliver) are unlikely to be copyright-protected in the US. Your own post-production additions may be copyrightable separately. The underlying AI output itself is in a contested legal space. Always consult your legal team before claiming copyright in AI-assisted commercial work.

Training Data Litigation — What It Means for Designers

Parallel to copyright ownership questions, major litigation is underway about whether training AI models on copyrighted images constitutes infringement. In 2023, three stock photography plaintiffs (including Getty Images) filed suit against Stability AI. Getty's UK case, filed in January 2023, alleged that Stability AI scraped and used 12 million Getty images without licence. As of 2025 these cases remain unresolved but have produced two immediate practical effects.

First, Adobe Firefly's "trained on licensed data" proposition became a significant commercial differentiator — prompting enterprises including Publicis Groupe and WPP to mandate Firefly for client deliverables requiring legal clarity. Second, Midjourney and Stability AI both added opt-out mechanisms for creators whose work appeared in training data, and several jurisdictions are legislating mandatory opt-out rights.

Platform Terms of Service — What You Actually Own

Each platform grants different usage rights through its Terms of Service, independent of copyright law:

Platform Commercial Use Ownership Grant Key Restriction
Midjourney (paid) Yes (paid plans) Non-exclusive licence; generated images not copyright-protected Enterprise plan required for companies >$1M revenue
DALL·E 3 (OpenAI) Yes OpenAI assigns all rights to user; no OpenAI claim on outputs Cannot use to deceive or misrepresent
Stable Diffusion Yes (CreativeML OpenRAIL-M licence) User owns outputs; model itself is licensed not sold Cannot use for specific prohibited applications
Adobe Firefly Yes User retains full rights; Adobe IP indemnification applies Must have active Creative Cloud subscription
Ethical Dimensions for Designers

Beyond legal compliance, professional designers face ethical questions that law has not yet addressed: disclosure (should clients know AI was used?), representation (who should be depicted in AI imagery, and with what care?), displacement (what obligations exist toward illustrators and photographers whose work may have trained the models?), and environmental cost (large-scale AI image generation carries significant compute energy costs).

Several major design industry bodies — including the AIGA and the Design Council UK — published AI ethics frameworks in 2023–2024. These consistently recommend: proactive client disclosure of AI tool use; intentional representation review of AI outputs before publication; and support for opt-out mechanisms for living artists.

Key Terms
Human Authorship ThresholdThe copyright standard requiring demonstrable human creative expression; AI-generated images meeting this standard must show human creative decision-making beyond prompt entry.
IP IndemnificationAdobe Firefly's legal guarantee to defend users against IP claims arising from their AI-generated outputs — uniquely valuable for enterprise commercial use.
CreativeML OpenRAIL-MThe licence governing Stable Diffusion outputs; permits commercial use of outputs while restricting certain harmful applications of the model itself.
Training Data LitigationOngoing legal cases (including Getty Images v. Stability AI) contesting whether training AI models on copyrighted images constitutes infringement.
Designer Takeaway

For client commercial deliverables requiring legal certainty: use Adobe Firefly and document it. For internal ideation or work where copyright ownership is not critical: any platform is workable under its ToS. For large enterprise accounts: understand the revenue thresholds in Midjourney's Enterprise plan. And for all AI work: build disclosure into your client relationship from the start — the law will catch up, and transparency protects you.

Lesson 4 Quiz

Copyright, Ownership & Ethical Use — 4 questions
1. The US Copyright Office's 2023 Zarya of the Dawn ruling established which key principle for designers using generative AI tools?
Correct. The Copyright Office granted copyright to the text and arrangement of "Zarya of the Dawn" but cancelled protection for all individual AI-generated images — establishing the human authorship threshold that now governs US AI copyright.
Not correct. The ruling determined that AI-generated images without sufficient human creative expression do not qualify for copyright protection — regardless of subscription type or platform.
2. Which platform offers IP indemnification, and why did this become a commercial differentiator leading enterprises like Publicis Groupe and WPP to mandate its use?
Correct. Firefly's licensed training data and IP indemnification guarantee are what led major holding companies to mandate it for client deliverables — no other platform can make this legal commitment at scale.
Not quite. Adobe Firefly is the only platform offering IP indemnification — a legal defence guarantee that became the deciding factor for enterprise agencies needing legal certainty on client AI imagery.
3. Under Midjourney's paid terms of service, which condition applies specifically to companies earning over $1 million in annual revenue?
Correct. Midjourney's Terms of Service explicitly require an Enterprise plan for commercial use by companies with more than $1 million in annual revenue — a threshold that catches many professional studios and agencies.
That's not correct. Companies earning over $1M annually must hold an Enterprise plan for commercial use of Midjourney outputs — the standard paid plan does not cover them.
4. The AIGA and Design Council UK AI ethics frameworks consistently recommended which designer practice regarding AI-generated imagery?
Correct. Both bodies consistently recommend disclosure, representation review before publication, and supporting mechanisms that allow artists to opt out of training data — going beyond legal compliance into professional ethics.
Not quite. The frameworks recommend three things: proactively disclosing AI use to clients, reviewing AI outputs for representation issues before publication, and supporting opt-out rights for artists whose work may have trained the models.

Lab 4 — Ethics & Legal Scenarios

Work through real copyright and ethical dilemmas in AI image use with your AI coach

Your Mission

You'll analyse specific scenarios involving copyright, platform terms, and ethical decisions in AI image use. Your coach will present situations that require you to apply what you've learned about the Zarya ruling, platform ToS differences, and industry ethics frameworks.

Ask for a scenario, or bring a real situation you've encountered — a client request, a tool choice, a representation question — and we'll analyse it together.
AI Design Coach
Lab 4
Welcome to Lab 4 — our final lab in this module. We're applying the legal and ethical framework to real scenarios: copyright questions, platform ToS conflicts, client disclosure decisions, and representation considerations.

I can give you a scenario to analyse, or you can bring a real situation from your practice. Where would you like to start?

Module 2 Test

Generative Image Tools for Designers — 15 questions · 80% to pass
1. Which generative image platform is best suited for a deliverable requiring IP indemnification for a large enterprise client?
Correct. Adobe Firefly is the only platform offering IP indemnification — essential for enterprise client deliverables requiring legal certainty.
Adobe Firefly is the correct answer — it's the only platform that offers IP indemnification, backed by its licensed-data training.
2. What does a LoRA (Low-Rank Adaptation) enable in a Stable Diffusion workflow?
Correct. LoRA is a lightweight fine-tuning method that teaches a base diffusion model new styles or subjects without requiring full model retraining.
LoRA is a lightweight fine-tuning technique — it adapts the base model to learn specific styles, faces, or brand objects without full retraining.
3. DALL·E 3 uses GPT-4 to rewrite user prompts. What is the primary practical consequence of this for design work?
Correct. GPT-4 prompt rewriting improves literal accuracy but may override subtle stylistic choices you make deliberately — a key workflow consideration.
The GPT-4 rewrite system produces high literal fidelity but can override deliberate stylistic decisions the designer made in the original prompt.
4. In the Five-Layer Prompt Framework, which layer is most directly responsible for ensuring correct camera perspective and output dimensions?
Correct. Technical Parameters covers aspect ratio, camera angle, resolution targets, and platform-specific settings like Midjourney's --ar flag.
Technical Parameters is the correct layer — it handles aspect ratio, camera angle, resolution targets, and platform-specific settings.
5. Seed locking in generative image tools is most valuable for which production scenario?
Correct. A fixed seed + consistent prompt template produces variants that share the same visual language — essential for campaign coherence across multiple hero images.
Seed locking controls the starting noise pattern, allowing stylistically matched variants across different subjects — key for campaign visual consistency.
6. What is the denoising strength range most appropriate for img2img style transfer when you want major aesthetic change but need to preserve the source composition?
Correct. 0.4–0.55 produces major style shifts while keeping the source image's compositional structure largely intact — the professional sweet spot for style transfer.
0.4–0.55 is the sweet spot — major style change while preserving the original composition. Lower values barely change anything; higher values lose the composition.
7. Which ControlNet mode would you use to maintain exact human body positions across multiple generated images for a fashion campaign?
Correct. OpenPose extracts a skeleton from a reference image and conditions generation on it, locking body position across multiple outputs — essential for fashion campaign consistency.
OpenPose is correct — it extracts and applies a body position skeleton, ensuring consistent figure positioning across generated variants.
8. Inpainting's primary value in a professional design pipeline is best described as which of the following?
Correct. Inpainting surgically regenerates only masked regions — solving the "almost perfect" problem by fixing hands, backgrounds, or elements without disturbing the rest of the image.
Inpainting's value is surgical repair — regenerating only the masked problem area while leaving the rest of the image pixel-identical.
9. The 2023 US Copyright Office ruling on "Zarya of the Dawn" determined that AI-generated images were not copyright-protectable because they lacked what?
Correct. The ruling established the human authorship threshold — copyright requires human creative expression, and entering a text prompt and accepting a generated result does not meet this standard.
The ruling turned on human authorship — copyright law requires demonstrable human creative expression, which a simple text-prompt-to-image workflow does not provide.
10. Under Midjourney's Terms of Service, which revenue threshold triggers the requirement for an Enterprise plan for commercial use?
Correct. Companies earning over $1 million annually must use Midjourney's Enterprise plan for commercial use — a threshold that catches many professional studios and agency groups.
The correct threshold is $1 million annual revenue — above this, the standard paid plan is insufficient for commercial use under Midjourney's ToS.
11. The Heinz "A.I. Ketchup" campaign, in which every AI generator defaulted to Heinz-style imagery given only the word "ketchup," demonstrated which phenomenon?
Correct. Heinz's visual dominance was so complete that the word "ketchup" alone triggered Heinz-like outputs across all tested models — demonstrating that brand equity lives inside training data.
The correct insight is that strong brand equity gets encoded into AI training data — the brand's visual identity became the model's default for the entire category.
12. Which ControlNet mode is most appropriate for preserving architectural line geometry in interior visualisation work?
Correct. MLSD (Multi-Line Segment Detection) preserves architectural straight-line geometry — specifically designed for interior and exterior architectural line preservation.
MLSD is the correct mode — it's designed specifically for architectural straight-line detection and preservation in interior/exterior visualisation.
13. Adobe Firefly's "trained on licensed data" proposition became commercially decisive for enterprises after which event?
Correct. Training data litigation — particularly Getty Images v. Stability AI — made legal provenance of training data a critical enterprise concern, directly benefiting Firefly's licensed-data positioning.
The triggering event was training data litigation (including Getty v. Stability AI), which made legal data provenance suddenly critical for enterprise brand safety.
14. A designer using Midjourney's --chaos 0 parameter and locking a seed is most likely in which stage of the production workflow?
Correct. --chaos 0 plus seed locking produces highly predictable, reproducible outputs from a fixed template — the hallmarks of Production Lock stage for consistent campaign asset generation.
--chaos 0 with a locked seed means maximum predictability and reproducibility — that's Production Lock, not exploration or refinement.
15. According to both the AIGA and Design Council UK AI ethics frameworks, which set of practices do professional designers have a responsibility to follow when using AI-generated imagery?
Correct. Both bodies consistently recommend disclosure to clients, representation review before publishing, and supporting opt-out rights for artists — the three pillars of ethical AI image practice for designers.
Both frameworks recommend three things: proactive client disclosure, representation review of AI outputs before publication, and supporting opt-out mechanisms for artists whose work trained the models.