Why creative is where speed meets real legal risk
Text you can proofread; an image you have to reason about. AI image tools - Midjourney, DALL-E / GPT image, Nano Banana, Canva - make gorgeous marketing creative fast, but the same speed makes it easy to ship something that copies a living artist's style, mimics a real person, or leans on a trademark. Tonight you build a small on-brand Cadence visual kit, then treat the rights check as a required step, not an afterthought. The rail here is loud on purpose: human review before anything publishes.
On-brand image generation 6 min live
The difference between one lucky image and a consistent brand look is the prompt recipe. A strong image prompt is not a wish - it is a structure: subject, style, palette, mood, and aspect ratio, plus a brand style reference so every image belongs to the same family. Nail the recipe once and you can reproduce the look across an ad, a social post, and a banner.
LiveThe recipe: subject, style, palette, mood, aspect3 min▶
Vague prompts give you a slot machine - occasionally brilliant, never repeatable. A structured prompt gives you a recipe you can rerun. Say what is in frame, how it should look, the exact brand colours, the feeling, and the size, then anchor it with a reference so the family holds.
- Be specific about each ingredient. "A laptop on a warm desk, clean flat-illustration style, deep-pink and cream palette, calm and focused mood, 16:9" beats "a nice picture of a note-taking app".
- Use brand style references. Most tools let you feed a reference image or a saved style. That is how you stop every image looking like a different company made it.
- Know the tool for the job. Midjourney for stylised concepts, GPT image / DALL-E for prompt-following and text-in-image, Nano Banana for edits and consistency, Canva for on-brand templated production. Match the tool to the task.
The IP, likeness & rights minefield 6 min live
This is the rail that can end a campaign, so read it twice. AI images can raise real copyright, likeness, and trademark problems - and "the AI made it" is not a defence. Before any AI visual goes near a public channel, it gets a rights check and a human sign-off. No exceptions.
An AI image is an input you are responsible for, not a finished clearance. The tool does not guarantee you own the output, that it is free of someone else's copyright, or that it is safe to use commercially. That responsibility sits with you and your brand. Treat every generated image as unverified until you have checked it.
LiveThe three risks: copyright, likeness, trademark3 min▶
Three separate landmines, each with its own guardrail. Learn to spot all three before you prompt, not after you publish.
| Risk | Guardrail |
|---|---|
| Copyright - output trained on or mimicking copyrighted work, or a named living artist's style | Do not prompt "in the style of [living artist]"; use generic style words; prefer rights-clear inputs and your own references |
| Likeness / right of publicity - real people, celebrities, deepfakes, recognisable faces | No real or identifiable people without written consent; never generate a public figure; no synthetic "customer" faces presented as real |
| Trademark - logos, brand marks, another company's product or packaging in frame | Keep other brands' logos and trade dress out of the image; do not imply endorsement you do not have |
| Commercial-use terms - the tool's own licence for paid/business use | Read the tool's terms; confirm your plan permits commercial use and check who owns the output |
| Disclosure - where AI-generated imagery must be labelled | Disclose AI-generated visuals where the platform or law (FTC / ASA) requires; never pass a synthetic image off as a real photo of a real event |
Self-studyRights-clear inputs and the "would this hold up" test2 min read▶
- Use rights-clear inputs. Your own photos, licensed stock, brand assets you own - not images scraped off the web fed in as references.
- Avoid naming living artists. "in the style of [famous living artist]" invites a copyright and moral-rights problem. Describe the look in generic terms instead.
- No real faces without consent. Do not generate identifiable real people, and do not create synthetic people you then present as genuine customers.
- Check commercial-use terms per tool. Free tiers and business tiers differ on ownership and commercial rights. Read them for every tool you use.
- Disclose where required. Follow platform and FTC / ASA rules on labelling AI-generated content; never deceive.
Brand consistency at volume 4 min live
One on-brand image is a demo; fifty on-brand images is a system. The way you scale without drifting is presets, reference sets, and human art direction - and a rule you carry from b2: the visual serves the message, it does not replace it.
LivePresets, do/don't refs, human art direction2 min▶
Consistency at volume comes from reuse and judgement, not from generating more and hoping. Save what works and steer what does not.
- Save the recipe as a preset. Lock the style, palette, mood, and reference so anyone on the team produces the same family.
- Keep do/don't reference boards. Show examples of on-brand and off-brand output so the tool - and the team - has a target.
- Art-direct, do not just accept. The human picks, crops, adjusts, and rejects. AI proposes; you direct. That is where the brand actually lives.
LiveThe visual serves the message2 min▶
A striking image that says nothing about Cadence is decoration, not marketing. Tie every visual back to the positioning from b2 and the message it supports. Pretty is the floor; on-message is the point.
Craft an on-brand Cadence visual set ★ 13 min · everyone builds
Build a small, coherent visual set - one ad image and a couple of social images - from a single consistency recipe, so they read as one brand rather than three random pictures.
Write the recipe first, in a chat AI: subject options for Cadence, plus a fixed style, palette (deep pink and cream), mood, and the aspect ratios you need (16:9 ad, 1:1 and 4:5 social). Save it as your preset.
Generate the ad image in your image tool using the recipe. Then generate the social images by keeping style, palette, and mood fixed and changing only the subject and aspect ratio.
Use a brand style reference if the tool supports it - a saved style, a reference image you own, or a first accepted image used as the anchor for the rest.
Art-direct the set. Reject anything off-palette or off-mood and regenerate. Confirm all three look like they came from the same brand.
Check each against the message: does this ad or post actually support Cadence's positioning, or is it just a nice picture? Cut the decoration.
The set that finally looked like a brand. A team kept generating one-off images that were each fine but never matched - a different vibe every post. Once they locked a single recipe and reused it, their feed suddenly read as one confident brand instead of a stock-image grab bag. Consistency, not cleverness, was what made it look professional.
Run the rights & brand checklist before "publish" ★ 8 min · build your own
The gate that keeps the brand safe. Take each image from Demo 1 and run it through an IP, likeness, and rights check plus a brand check, before anything is treated as ready to publish.
Build the checklist with a chat AI, tailored to your tools: one line each for copyright, likeness / real people, trademark / other logos, the tool's commercial-use terms, disclosure, and the two brand questions.
Run every image through it, honestly. Any "no" or "not sure" blocks the image until it is fixed or replaced. A "not sure" on rights is a no.
Regenerate anything that fails - drop a named artist reference, remove a stray logo, swap a too-realistic face for a clearly illustrated one - and re-run the check.
Record the human sign-off. Note who reviewed each image and that it passed. That record is what turns "we should be careful" into an actual control.
Try it yourself - this week ◐ 30-45 min total
- Finish both demos. If you only do one, do the checklist - the rights habit matters more than any single image.
- Build a real on-brand visual set for a project of your own using one locked recipe, and confirm the images read as one family.
- Run every image through your IP, likeness, and rights checklist, and note the human sign-off for each. Keep the record.
- Read the commercial-use terms for each AI image tool you actually use. Write down, in one line, whether your plan allows commercial use and who owns the output.
- Optional: find one image you were tempted to publish that fails the check on likeness or a stray logo. That near-miss is the lesson. Bring it to session b9.
Official sources covered
Taught from public AI image-generation practice and the IP, likeness, and rights canon in the course map. Certificates and paid tooling stay with the official programs. This session covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · How do you get a consistent, on-brand look across many AI images?
Consistency is a recipe, not luck. Fix the style, palette, mood, and a reference; vary only the subject. Reusing the recipe is what makes a set read as one brand.
2 · Which of these is a real legal risk with AI-generated marketing images?
All three are live risks, plus the tool's commercial-use terms and disclosure. "The AI made it" is not a defence - responsibility for the output sits with you and the brand.
3 · An AI ad image looks great. What has to happen before it publishes?
Human review before anything publishes is the loud rail. Check copyright, likeness, trademark, commercial terms, disclosure, and brand fit, and record the sign-off. A "not sure" on rights is a no.