Describe-Modify-Imagine Loop
Reverse-engineer an image prompt, change one variable, and regenerate.
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 5
- Confidence
- 99%
The Describe-Modify-Imagine Loop turns an existing image into a prompt-learning and variation workflow. The user uploads a reference through Midjourney's Describe command, receives candidate descriptions of the visual, and studies the language used for subject, style, color, composition, and aspect ratio. One candidate prompt is then passed into Imagine after changing a single meaningful variable, such as the character, setting, or treatment. Holding the other terms stable makes it easier to see how that change affects the output. The generated result can be compared with the source or prior iteration, and the loop repeated. The mechanism reduces blank-page prompting, exposes reusable visual vocabulary, and enables rapid controlled experimentation, although the reverse-engineered prompt may not reproduce arbitrary web images exactly.
Origin
Extracted from Marketing Against The Grain during a demonstration of Midjourney V5's newly launched Describe feature.
Core principles
- 01Start from a concrete visual reference instead of a blank prompt.
- 02Use reverse prompting to expose the image's likely visual ingredients.
- 03Change one variable at a time so the effect remains observable.
- 04Treat generated prompts as learning material, not guaranteed reproductions.
How to run it
- 1
Choose a Reference
Select an image whose visual structure or style you want to understand and adapt.
Pro tip Choose a reference with a clear subject and recognizable visual treatment.
Watch out Do not assume the workflow grants rights to reproduce copyrighted characters or artwork.
- 2
Run Describe
Upload the image to Midjourney and invoke Describe to generate candidate prompts.
Pro tip Look for terms that identify style, perspective, palette, patterns, and aspect ratio.
Watch out The returned language is an interpretation, not the original creator's confirmed prompt.
- 3
Select a Baseline Prompt
Choose the description that best captures the elements you want to preserve.
Pro tip Favor the candidate that explains the image's structure rather than merely naming its subject.
Watch out A weak baseline makes later changes difficult to diagnose.
- 4
Change One Variable
Modify one subject, setting, stylistic, or compositional element while leaving the remaining prompt intact.
Pro tip Use a conspicuous change first so its effect is easy to evaluate.
Watch out Changing several variables at once obscures which instruction caused the result.
- 5
Run Imagine and Compare
Generate the variation with Imagine, compare it with the baseline, and repeat the loop with another controlled adjustment.
Pro tip Save successful phrases as reusable visual vocabulary.
Watch out Some reference types will produce only loose approximations.
In the wild
Kieran uploaded an image of He-Man and Skeletor, obtained a description, reused that prompt in Imagine, and changed the character instruction to make He-Man a leprechaun while keeping the surrounding prompt stable.
→ He produced a rapid character variation while learning how Midjourney described the source image.
A marketer uploads a successful mountain campaign image, selects the strongest Describe output, changes only the location to a coastal setting, and regenerates it. Further iterations adjust one color or prop at a time.
→ The marketer develops related campaign concepts without starting each prompt from zero.
Common mistakes
Changing Everything at Once
Replacing the subject, style, palette, and composition simultaneously prevents useful learning from the output.
Expecting Exact Reconstruction
Describe infers a plausible prompt and may not recreate an arbitrary source image closely.
Ignoring Usage Rights
Technical reproducibility does not establish permission to use the source image, character, or generated derivative.
Is it for you?
Best for
It is best for marketers and creators who have a visual reference and want to produce related concepts quickly.
Not ideal for
It is not ideal when an exact reproduction, protected brand asset, or precisely art-directed final image is required.
From the transcript
“you can basically have a command that is called describe, and describe allows you to upload an image. And when you upload that image, it…”
“then using the imagine command, which basically allows you to recreate the image, I added in the description the prompt and like just changed one…”
“that allows you to quickly learn how an image was created, which you're showing on the screen, but then be able to like modify it…”
From the episode
Midjourney V5 Update: Everything You Need To Know (#107)