Midjourney’s editing and reference workflows evolve with model versions. Use the current web interface and model documentation rather than assuming older V6/V7 buttons or reference parameters still behave the same.
1. Choose a clean source image
Start with the highest-quality source you are allowed to use. Decide whether the job is object replacement, background change, styling, cleanup or composition correction.
2. Write the invariants first
State what must not change: person identity, camera angle, pose, product shape, lighting direction, aspect ratio or color palette.
3. Make one edit request
Change one meaningful variable at a time. “Replace the jacket with a black leather jacket, preserve face, pose and lighting” is easier to judge than five simultaneous changes.
4. Compare identity and geometry
Inspect hands, face, edges, perspective, product logos/text and background continuity. A prettier image is still a failed edit if it changes protected geometry.
5. Iterate from the best controlled result
Use the strongest output as the next base only after it preserves the required invariants. Avoid repeatedly editing a flawed intermediate because errors compound.
Edit only this element: [change]. Preserve the subject identity, pose, camera angle, composition, lighting direction and all other objects. Keep the result photorealistic and do not add text or extra accessories.
Using an old tutorial for a changed model; requesting many edits at once; failing to define what must stay fixed; continuing from an intermediate that already changed identity or perspective.
Finish check
The edited image makes the requested change while the protected identity, composition and geometry remain stable enough for the intended use.