Capstone: Build Your Creative Production Workflow
- Build a prompt formula that keeps aesthetic and subject constant across a full image set
- Use Draft Mode for the iteration phase before committing to Standard Mode for finals
- Maintain a library of style codes, reference images, and personalization profiles across projects
The Problem with an Ad-Hoc Approach
Most people start using Midjourney the same way: open the app, type something, see what happens. That works for exploration. It doesn't scale for production.
When you need to deliver a consistent set of images for a project — a brand campaign, a book's illustrations, a social media content calendar — an ad-hoc approach produces inconsistent output, burns Fast GPU hours inefficiently, and leaves you searching through hundreds of images for the ones you actually want.
This capstone lesson shows how to turn the tools from this track into a system that's repeatable, efficient, and produces consistent results.
Step 1: Define the Visual Language Before You Generate
Before opening the prompt bar, answer three questions:
- Subject: What specific subjects will appear? (A specific character? A product? A type of environment?)
- Aesthetic: What should the images look and feel like? (Warm film photography? Clean product studio shots? Dark painterly illustration?)
- Output specs: What aspect ratio, resolution, and format does the final output need?
The first two questions determine which tools you'll use. Consistent subjects → you'll need --oref. Consistent aesthetic → you'll need --sref and possibly personalization. Output specs → you'll set --ar from the start.
Step 2: Build Your Reference Library
Spend 10–15 minutes in Explore finding images that match your target aesthetic. Like them, add them to a moodboard named for the project. This becomes your --sref source for the whole project.
If the project involves a recurring subject — a character, a product, a mascot — get or create one clean reference image. This becomes your --oref source. Test it at a few different weight values (100, 300, 500) to understand how it behaves before committing to a workflow.
Rate your personalization profile if you haven't recently. 200 ratings takes 15 minutes and will improve every generation in the project.
Step 3: Build a Prompt Formula
For any project that requires multiple related images, build a base prompt formula with the shared elements locked in:
[subject + scene description], [lighting], [style/medium] --ar [ratio] --oref [URL] --ow [weight] --sref [URL] --sw [weight] --p --stylize [value]
Example formula for a product photography project:
a leather wallet on [surface], [lighting description], editorial product photography --ar 4:5 --oref https://product-ref.jpg --ow 350 --sref https://aesthetic-ref.jpg --sw 100 --p --stylize 150
The bracketed parts change per image. Everything else stays constant. This is what makes the output consistent across a full set.
Step 4: Use Draft Mode for Prompt Development
Before running the full formula in Standard Mode, add --draft and run 5–10 variations at 10x speed and half the cost. Check:
- Is the subject consistent enough?
- Is the aesthetic matching the moodboard?
- Is the aspect ratio working with the composition?
Adjust your formula based on what you see. This is the phase to test and iterate. Running 10 drafts costs the same as about 5 standard generations — use it freely.
Step 5: Generate Finals, Edit, Deliver
Once the formula is working in Draft Mode, switch to Standard Mode (remove --draft). Generate the full set. For each image:
- Upscale the best of the four
- Use Vary Region for any specific fixes (wrong expression, unwanted element)
- Use Zoom Out or Pan if the framing needs adjusting
- Use Upscale (Subtle) for the final high-resolution version
- Add to the project collection
Maintaining the Library
After the project is complete:
- Save the style codes for the best-performing images from each project (visible in image details). These become the basis for future
--srefinputs that are guaranteed to match your project's aesthetic. - Note the
--owweight that worked for your--orefsubject. Write it down alongside the reference image. - Save your personalization shortcode if you trained a new profile for this project.
Over time, you build a library of prompts, reference images, style codes, and personalization profiles that make the next similar project dramatically faster to set up.
The Formula at a Glance
- A prompt formula with locked-in --sref, --oref, --ar, and --p produces consistent results across a full project set
- Define the visual language before generating — decisions made upfront eliminate most revision cycles
- Draft Mode for iteration + Standard for finals + Vary Region for fixes is the core production loop
Progress is saved in this browser only — clear your cache or switch device and it's gone. Create a free account and it follows you everywhere.