Where you pick up Cadence tonight
Cadence has words now - a core idea, long-form, and a social set. Visuals are how those words get seen: covers, social graphics, thumbnails. AI image tools make this fast, but they raise rail two from a footnote to the loudest voice in the room. A great-looking image that borrows a living artist's style, a real face, or copyrighted work is not an asset - it is a liability with your name on it. Tonight is speed and consistency, gated by a rights check that actually has teeth.
On-brand image generation 6 min live
Image models - Midjourney, DALL-E, and design suites - turn a text prompt into graphics, covers, and social visuals. The difference between a lucky one-off and a consistent brand set is the prompt recipe: a fixed structure of subject, style, palette, mood, and aspect ratio, plus a brand style reference so image after image looks like it belongs to the same brand. Random prompts give random results; a locked recipe gives a set.
LiveThe five-part recipe and the style reference3 min▶
Every strong image prompt answers five questions in the same order, so you can change the subject without losing the brand look. Then a style reference - a sample image or a reusable style code the tool supports - carries the look from one image to the next. That is the difference between a graphic and a graphics system.
- Subject. What is literally in the frame - the object, scene, or concept.
- Style + palette. Flat illustration or photo, and the brand colors, stated every time.
- Mood + aspect. The feeling and the exact ratio for the format - 1:1 post, 16:9 cover, 9:16 story.
From lucky shots to a system. A creator kept generating great single images that looked nothing like each other, so the feed felt chaotic. They wrote one recipe - fixed style, palette, and mood - and swapped only the subject. Suddenly a month of visuals looked like one brand. The model did not get better; the recipe did the work.
Self-studyPrompt patterns for consistency2 min read▶
- Fix the style line, vary the subject. Keep style, palette, and mood identical across a batch; change only what is depicted.
- Use the tool's style features. Style references, seeds, or saved style codes are how you carry a look - learn the one your tool offers.
- State the aspect explicitly. Wrong ratio is the most common reason a good image is unusable. Match the format up front.
- Keep a recipe file. Save the working recipe as text so anyone on the team can regenerate on-brand visuals.
This recipe is exactly what you lock in Demo 1. Consistency is a written recipe, not a lucky prompt.
The IP + likeness minefield 5 min live
Read this part twice. AI image generation is the single most rights-sensitive thing in this whole course, and a beautiful image is worth nothing if you did not have the right to make or publish it. Four risks matter, and each has a guardrail. None of these are optional, and "the tool let me" is not a defense.
| Risk | The guardrail (non-negotiable) |
|---|---|
| Copyright + training-data questions on generated images | Treat output as unsettled; do not rely on it for high-stakes use; use rights-clear inputs only |
| Mimicking a living artist's named style | Do not prompt "in the style of [living artist]" - describe the qualities you want instead |
| Likeness / right-of-publicity - real faces | No real person's face without permission; a recognizable face is a legal exposure, not a prop |
| Tool commercial-use terms vary by plan | Read the plan's commercial and ownership terms before any client or paid use |
LiveThe four risks, said plainly3 min▶
This is rail two turned all the way up. Every AI image can carry exposure that a text draft never could, and the human review before publishing is where you catch it. Speed is not a reason to skip it - it is the reason you need it.
- Copyright is unsettled. Who owns AI output, and what the model trained on, are open legal questions. Do not treat a generated image as clearly yours; keep inputs rights-clear.
- Do not clone a living artist. Prompting a named living artist's style mimics their livelihood. Describe the look you want - colors, shapes, mood - not a person.
- Real faces are off-limits. A recognizable person's likeness has right-of-publicity protection. No real faces without explicit permission, full stop.
- Terms + disclosure + provenance. Commercial-use rights vary by tool and plan; disclose AI content where required (FTC); adopt content-credential and provenance practices as they mature.
Self-studyWhy "it looks fine" is not clearance2 min read▶
- The risk is invisible in the pixels. An image can look original and still echo a copyrighted work, a living artist, or a real face. Looking good is not the same as being clear.
- Client and paid use raise the stakes. The more commercial the use, the more the rights question matters. Check the tool terms for exactly that use.
- Disclosure protects you too. Where AI content or endorsements require disclosure, doing it is not a weakness - it is what keeps the work defensible.
- Provenance is coming. Content credentials and watermarking are becoming table stakes; get in the habit now.
The cover that had to be pulled. A team shipped a striking AI cover generated "in the style of" a named living illustrator. It looked great and traveled fast - then the artist noticed. The image came down, an apology went out, and the lesson stuck: the guardrail is cheaper than the cleanup. Describe the look, never clone the person.
Consistency at volume 4 min live
One on-brand image is a nice prompt. A batch that all looks like Cadence, across formats, is a system - and a system needs a locked recipe plus templates. The other half of the discipline: a visual has to serve the message, not just decorate the post. A pretty image that says nothing is slop with better lighting.
LiveLocked recipe + templates2 min▶
Once the recipe works, freeze it - the exact style line, palette, and mood - and build a small set of templates: a post size, a cover size, a story size. Now a batch of visuals stays on-brand because the recipe, not your mood that day, decides the look. And every image earns its place by carrying the message, not by being decoration.
The recipe file that scaled a solo brand. A one-person brand wrote its visual recipe into a single saved text block plus three aspect templates. A month of covers, posts, and stories all read as one brand - made in minutes each, because the decisions were made once. When a teammate joined, they used the same file and the look never wavered.
Self-studyVisuals serve the message2 min read▶
- Lock the recipe. Save the style line, palette, mood, and per-format aspect ratios as a reusable file, not a memory.
- Template the formats. One recipe, a few fixed sizes - post, cover, story - so the batch is consistent by construction.
- Message first, decoration never. If the image does not clarify or carry the point, it is filler. Pretty and pointless is still slop.
This locked recipe plus the rights checklist is what you hand into the b8 pipeline and govern at scale in b10.
Lock a recipe, generate an on-brand set ★ 13 min · everyone builds
Write one prompt recipe and use it to make a small Cadence visual set - a post graphic and a cover - that clearly belong together. Consistency by recipe, not by luck.
Use ChatGPT or Claude to draft a five-part recipe from the Cadence brand: subject slot, style, palette (state real hex), mood, and per-format aspect ratios.
Take the recipe to your image tool. Generate a 1:1 post graphic - keep the style line fixed, fill only the subject slot.
Re-run the same recipe at 16:9 for a cover, changing only the subject. Use the tool's style reference or seed to carry the look.
Put the two side by side. If they do not read as one brand, tighten the style line - not the subject - and regenerate.
Save the working recipe as text. That file is your on-brand generator for every future batch.
Run the rights + brand checklist before publish ★ 8 min · build your own
No AI image ships until it clears a checklist. You build the checklist once, then run each image from Demo 1 through it - IP, likeness, tool terms, disclosure, and a brand fit check. This is rail two with teeth.
Draft a pre-publish checklist with an AI: rights-clear inputs, no living-artist clone, no real faces, tool commercial terms checked, disclosure where required, human review, on-brand.
Take your post graphic from Demo 1 and answer every line honestly. Any "no" or "not sure" stops the image - fix it or drop it.
Run the cover through the same list. Confirm the tool's plan actually permits your intended use.
Only images that pass every line are marked "publish". Save the checklist - it becomes a gate in your b8 pipeline and your b10 governance.
The checklist that caught it at the gate. A creator loved an AI cover until the checklist forced the question "any real faces?" - and the background had a recognizable person. It never published; a five-minute regeneration fixed it. The checklist did what admiration could not: it caught the exposure before the audience did. Human review before publish is the whole guardrail.
Try it yourself - this week ◐ 30-45 min total
- Finish both demos if you did not complete them live - especially locking the recipe. Consistency comes from the written recipe, not a lucky prompt.
- Generate a three-image on-brand set from your recipe and run every image through the rights checklist before you would post it.
- Read your image tool's commercial-use and ownership terms for the plan you are on. Note what it does and does not permit.
- Save your recipe and your rights checklist as two reusable files. You will wire both into the b8 pipeline.
- Optional: find one AI visual you already posted and run it through the checklist retroactively. If it would not pass now, note why.
Official sources covered
Taught from public image-generation curricula and IP/rights canon. Certificates and paid tooling stay official. This session covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · How do you keep a batch of AI visuals on-brand?
Consistency is a written recipe, not luck. Fix the style line and style ref, vary only the subject, and the batch reads as one brand.
2 · Which of these is a real rights risk in AI image generation?
Copyright, a living artist's named style, likeness/right-of-publicity, and commercial-use terms are all live exposures. "The tool let me" is not a defense.
3 · Before an AI image can publish, you must...
Looking good is not clearance. The human review before publish is where you catch the exposure the pixels hide - it is the whole guardrail.