Practical guide. Provider claims are attributed to the sources below. This page does not claim a new independent DARC benchmark.
For: Creators and developers shipping generated visual assets
Before you start: A visual brief · Rights to the reference material
⚡ TL;DR — the 30-second version
- Define acceptance criteria before generating.
- Preserve model and edit provenance.
- Check privacy before upload and inspect the final responsive crop.
Image generation is not finished when the output looks impressive at thumbnail size. The asset still needs to meet dimensions, content, consistency, accessibility and rights requirements. A repeatable review process is more useful than a universal claim about which model has the most taste.
Keep the exact model and editing workflow in the project record. Midjourney's current documentation names V8.2 as the default and describes a new Edit Model; an old tutorial can send you looking for controls that have moved or changed.
“A good generation is a candidate. A publishable asset is a candidate that survived the brief, the rights check and the final layout.”
Field guide / 01
01 / Write a brief that can be checked
Specify use, dimensions, safe areas, subject, palette and prohibited elements. Decide whether text belongs inside the generated asset or should be added as real typography afterward. For a website, live text often improves accessibility and responsiveness.
Prepare reference images only when you have the right to upload and use them. Record which elements the reference is supposed to control—composition, identity or style—so reviewers know what consistency means.
Field guide / 02
02 / Keep an asset ledger
Save source inputs, prompt, model version, edit steps and selected output. Record any human retouching. The same product name can route to different underlying models over time, so preserve the served model when the API exposes it.
Avoid regenerating the entire scene to fix a small issue if a controlled edit can preserve the approved parts. Recheck neighboring details after every edit; a local instruction can have wider visual effects.
Field guide / 03
03 / Inspect at delivery size
Check text, logos, hands, reflections, repeated patterns and physical plausibility. Place the image in its final layout and test crops at mobile and desktop sizes. An attractive full canvas can fail when the responsive crop removes the subject.
Write useful alternative text from what the image actually communicates. Do not copy the generation prompt into alt text; the prompt describes intent, not necessarily the delivered asset.
- → Typography and brand marks are correct.
- → The chosen crop works at every target size.
- → People, objects and reflections have no distracting artifacts.
- → The image does not falsely imply a real event or endorsement.
Field guide / 04
04 / Check visibility and provenance
Midjourney's editor documentation warns that V8.X edited images can appear publicly unless the appropriate privacy mode is enabled. This matters before uploading unreleased product photography or confidential client work.
Keep applicable labels, consent records and provenance metadata. Watermarks and credentials can help disclose synthetic media, but they do not prove you hold the rights to every source image or depicted identity.
Before you ship
0/5Use this as a review aid. Checks reset when you leave this page.
Sources & verification
Last checked 23 Sept 2026. Publication dates and access conditions belong to the linked originals.
- 01Midjourney — current model versionsMidjourney↗
- 02Midjourney — editor and image visibilityMidjourney↗
- 03Runway — model routers for video, image and audioRunway↗
- 04Gemini 3.8 text-to-speech says helloGoogle↗