Lesson 10. Complete AI visual workflow: brief → prompt → iteration → export
There is a big difference between knowing prompting techniques and professional AI visual creation. In this final lesson, we combine everything learned into a single sequential workflow: from brief to finished image.
Topic breakdown
A professional AI workflow consists of 5 stages: Brief analysis → Prompt preparation → Generation and iteration → Selection and post-processing → Export and use.
Brief analysis: what is needed (product, portrait, illustration), for which platform, what mood/style, what goal (sales, branding, content).
Prompt preparation: make a list of key visual elements, choose a model, set aspect ratio, prepare negative prompt.
Iteration: evaluate the first generation (mood, composition, quality). Save the seed of the best option. Make minor changes. Stop when you achieve the goal.
What you'll learn
- Fully analyze a brief and translate it into a prompt plan
- Systematically conduct an iteration strategy
- Evaluate and select images by quality
- Prepare an AI image for post-processing and export
Deep dive
At the center of a workflow is not the prompt itself, but the sequence of decisions. When the choice of model, parameters, and prompt flows from the brief, iterations rarely go off track, and the team reaches consensus faster.
Professional iteration always has a reason: soften the light, simplify the background, center the product. Aimlessly producing dozens of options wastes time but does not guarantee quality.
Before final delivery, a technical checklist is also important: format, safe zone for crop, space for text, file naming, archiving the original prompt and seed. Such order accelerates the next campaign.
Lesson plan
Brief analysis: right question — right image
A brief is goal, platform, mood, and constraints. Clearly defining all four elements lets you insert them into the prompt.
Iteration strategy
The first generation is a starting point, not the finale. Take the seed of the closest option. Change one element. Evaluate. Repeat.
Selection and quality control
With each generation you get 4 images. Evaluation criteria: alignment with goal, technical quality, aesthetics, usability (does it need edits).
Post-processing and export
An AI image often requires minor refinement: cropping, color correction, removing unwanted details. Canva Magic Eraser or Adobe Firefly Generative Fill do this quickly.
Weak vs strong prompt
(without brief, immediately) a spa product image
Brief: spa brand, Instagram story, calm and premium → Prompt: 'luxury honey and lavender spa serum bottle, soft natural materials, warm cream and gold palette, bokeh background, morning natural light, clean minimalist lifestyle photography, premium brand aesthetic --ar 9:16 --q 2 --seed 73441'
Prompt based on brief: platform (9:16), brand mood (luxury, calm), color (cream/gold), lighting (morning natural), seed (for iteration) — all key elements are accounted for.
Ready visual prompt template
Copy and adaptWORKFLOW TEMPLATE: 1. Brief: [goal | platform | mood | constraints] 2. Model: [Midjourney / DALL·E / SD / Leonardo] 3. Main prompt: [subject | style | lighting | color | composition] 4. Parameters: [--ar | --seed | --q | --v] 5. Negative: [quality blocks | context blocks] 6. Iteration: [save seed → minor change → evaluate]
Why it works
Brief analysis — 4 questions: What? Where? What mood? What constraints?
Model selection: aesthetics → Midjourney, accuracy → DALL·E, control → SD, budget → Leonardo.
Iteration rule: change only 1-2 elements at a time. Many changes obscure what exactly had an impact.
Selection: out of 4-6 options take the closest to the goal, save seed, continue based on it.
Export: right format (JPEG for product, PNG for transparency, WebP for web), desired size.
Practice
- Take this brief: 'For a spa brand based on natural honey — Instagram story (9:16). Mood: calm, natural, premium. Color: cream and gold.'
- Choose a model (recommendation: Midjourney or Leonardo).
- Write a prompt: subject, style, lighting, color, and composition.
- Add parameters: --ar 9:16, --q 2.
- Generate, save seed, do 2 iterations.
- Select the final image and determine the export format.
Checklist
Common mistakes
- moving to generation without analyzing the brief
- stopping at the first good image without checking if the next iteration could be better
- changing too many elements in one iteration
- forgetting to save the seed of the best base
- ignoring export format (difference between JPEG/PNG/WebP)
Lesson FAQ
Is post-processing always needed for AI images?
Not always, but often. Minor edits: crop, color balance, sometimes fixing hands or background. Canva Magic Eraser or Adobe Firefly Generative Fill handle this quickly.
What are the rights for AI images?
Depends on the model and jurisdiction. In Midjourney Pro and Enterprise, images can be used commercially. DALL·E — per OpenAI terms. Stable Diffusion (open source) gives more freedom. We recommend reading terms of use.