ai:midjourney:workflow [mh8 Wiki]

This page is read only. You can view the source, but not change it. Ask your administrator if you think this is wrong.
====== Organizing Workflow ======

==== Three Image States ====

=== 1. Exploration ===
Loose experimentation. Prompt testing, style tests, weird ideas.

=== 2. Candidate ===
An image worth refining. You might upscale, remix, or edit it.

=== 3. Final ===
Something worth exporting, archiving, post-processing, or using in a project.

A simple workflow becomes:

<code text>
Explore → Select → Refine → Finalize
</code>

<WRAP round tip 60%>
**Inside Midjourney, the key is to separate exploration noise from meaningful candidates.**
</WRAP>



==== Use the “Like” System Aggressively ====

Treat it as your **first filter pass**.

When reviewing a grid:

  * Ignore most results
  * Heart anything with **potential**
  * Not only perfect images

Your workflow becomes:

<code text>
All generations → Liked images → Refinement work
</code>

Later you can simply open: **Organize → Likes**

Now you are browsing only the good seeds instead of the chaos.

Think of Likes as your creative shortlist.

==== Use “Organize” Like a Working Table ====

Midjourney’s Organize tab lets you filter:

  * Liked
  * Upscaled
  * Variations
  * Generated images

This is extremely useful if you adopt a habit:

Upscale only images that enter the “Candidate” phase.

That means:
  * **Exploration** → generate grids
  * **Candidate** → upscale
  * **Refinement** → remix / vary
  * **Final** → export

Then Upscaled images automatically become your shortlist.

Filtering by Upscaled becomes almost like a portfolio view.

==== Always Work From One “Seed Image” ====

A common beginner mistake is jumping prompts constantly.

Instead:
  * Generate a grid
  * Pick the strongest image
  * Build from it

=== Workflow: ===
  * **Draft** → Pick best composition
  * **Vary/Remix Strong** → If re-seed is needed on the draft
  * **Vary Subtle** → Explore undirected variations
  * **Remix** → Explore directed variations
  * **Editor** → Surgically regenerate parts
  * **Upscale** → Add detail and create high resolution "hero"

This creates image families instead of random outputs.

<WRAP round info 60%>
**Don’t Upscale Too Early**

Upscaling everything creates clutter and consumes credits.

**Rule of thumb:**

Grid → shortlist → upscale only the best
</WRAP>


==== Controlled Experimentation ====

Instead of random prompts you change one gene at a time.

Example Base prompt:

<code text>
ancient forest shrine
foggy forest
soft volumetric moonlight
cinematic fantasy concept art
wide angle composition
</code>

Then test only lighting:
  * **Test A:** ''soft volumetric moonlight''
  * **Test B:** ''bioluminescent glow''
  * **Test C:** ''sun rays through fog''


Now you understand exactly **what changed** in the image.
==== Use Prompt Comments in the Prompt Bar ====

Midjourney lets you add text that doesn't affect generation if you use separators.

=== Example: ===

<code text>
forest shrine at dusk, cinematic lighting --ar 3:2
// exploration phase

forest shrine at dusk, cinematic lighting --ar 3:2
// fairy version test
</code>

This way the prompt itself becomes documentation when browsing later.





==== Maintain One “Hero Image” ====

For each exploration session, try to end with **one hero image**

That’s the one you:
  * refine
  * upscale
  * export
  * archive

If you create 50 images but only 3 heroes, your archive stays clean.

==== Weekly “Creative Garbage Collection” ====

Once a week open Likes and ask:
  * still interesting?
  * worth refining?

Remove likes from anything that no longer excites you.

This **keeps your shortlist alive**.

=== A Simple MidJourney Workflow Blueprint ===

<code text>
1. Exploration
   generate grids

2. First filter
   heart interesting images

3. Candidate phase
   upscale promising ones

4. Refinement
   vary / remix / editor

5. Final
   hero image selected
</code>

This keeps the system manageable even with thousands of generations.