Level 0 exam — Midjourney & Visual AI

Exam — fresh questions

12 questions to validate the fundamentals with Sonia: how AI generates, which tool to pick, which pro uses, and how to stay legally clean.
Answer everything, then click Check: your score and explanations will appear.

0.1 · Recall

1. What is a diffusion model?

A camera plugged into AI A model that starts from random noise and denoises it step by step, guided by the text A filter that upscales the pixels of a photo A bank of images from which pieces are cut out

Random noise → denoising guided by prompt embeddings → emerging image.

0.1 · Application

2. Sonia reruns exactly the same prompt twice and gets two different images. What to conclude?

The tool is broken, reinstall it The prompt is too precise, shorten it The AI copied two different photos This is normal: each generation starts from different random noise — generation is probabilistic

Same prompt ≠ same image: to iterate seriously, freeze the seed and change one thing at a time.

0.1 · Application

3. Sonia wants to measure the exact effect of the word 'stormy' in her prompt. Which method?

Change the whole prompt on each run Freeze the seed and change only that word, then compare Rerun at random until she likes the result Ask a colleague to guess

Frozen seed + a single change at a time: otherwise you never know what produced the improvement.

0.2 · Recall

4. Match each tool to its strength per the guide's landscape.

All four tools do exactly the same thing Midjourney = local control; Stable Diffusion = aesthetics DALL·E 3 = Photoshop retouching; Firefly = briefs Midjourney = aesthetics; DALL·E 3 = briefs and text; Stable Diffusion = free local control; Firefly = Adobe production

Four real philosophies: aesthetics, brief obedience, local control, Adobe integration.

0.2 · Application

5. Sonia must produce a poster with a strict brief and a perfectly readable title. Which choice?

Use DALL·E 3, strongest on complex briefs and text in the image Entrust the text to local Stable Diffusion Generate only in Photoshop with no AI Draw lots between the four tools

DALL·E 3 follows complex instructions better and handles readable text — the constraint commands the tool.

0.2 · Arbitration

6. A colleague claims: 'Last year's comparison says tool X is the best, no need to test'. What verdict?

Reject: the market evolves very fast and the right choice depends on use — check the official sites and test the same prompt on your own case Accept: rankings are final Reject: you must pay the most expensive subscription in all cases Accept only for beginners

No absolute best tool: dominant constraint + official check + comparison on your case.

0.3 · Application

7. Sonia sorts her orders: mood-board, article illustration, photo of the product to sell. Which sort?

Refuse all AI on principle Generate the product photo and photograph the mood-board AI for the mood-board and illustration (idea); real photo for the product (proof) Generate everything with AI, including the product photo

AI illustrates an idea; it does not document a reality — the product photo promises what the client will receive.

0.3 · Application

8. A client asks Sonia for a generated 'fake client' to endorse his product. What to answer?

Refuse only if the client does not pay enough Refuse: a generated testimonial is a fake testimonial — people and facts belong to the real Accept: it is an illustration use like any other Accept if the image is beautiful

Real people and facts: exclusion zone — with AI disclosure and a written internal framework before scaling up.

0.3 · Arbitration

9. Management wants to deploy visual AI everywhere with no rules, 'to go faster'. What verdict?

Reject: AI is never useful in business Accept: speed justifies everything Accept if you also generate the product photos Reject: with no framework (uses, verification, AI disclosure), the image incident costs more than the time saved — combined workflow first

Upstream accelerator + written internal framework: explore and mock up with AI, finish with graphics tools.

0.4 · Recall

10. What are the three legal questions to distinguish?

Discord, the web and local Seed, chaos and style Training data, generated-image ownership, represented content Prompt price, speed and language

Disputed training, ownership (ToS + local law), content: three topics, three reasonings.

0.4 · Application

11. Sonia spots in her series a visual 'in the style of' a named living artist, with a lookalike logo. What to do?

Hide the logo in a corner of the image Rename the file to erase the trace Rule out or regenerate: describe style by components (light, palette, medium) and neutralize the logo Deliver as is: the subscription covers everything

No named-artist style, no brand: the ToS settle the file, never the right to show just anything.

0.4 · Arbitration

12. For a major campaign, Sonia hesitates: 'The guide is enough as legal advice, right?' What verdict?

Reject: law varies by country and evolves fast — up-to-date ToS + validation by a local lawyer for big stakes Accept if the campaign stays internal Reject: you must drop AI for campaigns Accept: the guide replaces the lawyer

Reference points yes, legal advice no: kept prompts, AI disclosure, and a lawyer for packaging, campaigns and brands.

What's next?

Score ≥ 80%: move on to Level 1 — first Midjourney prompts, with Sonia.
Score 50–80%: revisit the failed section(s) (0.1 diffusion, 0.2 tools, 0.3 uses, 0.4 rights), reread the explanations above, then retake the test.
Score < 50%: reread all of level 0 applying each 48h challenge (variability test, two-tool comparison, idea/reality grid, content filter) before retrying.
And after? Keep your Sonia tool comparison and delivery checklist: level 1 enriches them with the first prompts.