Midjourney V7 Prompt Migration

What You'll Learn

• Which V6 parameters need a V7 replacement or a different expectation
• How to place parameters, version flags, and references in a valid prompt
• When Draft Mode, Omni Reference, and Style Reference help or conflict
• How to troubleshoot a V7 prompt without randomly adding keywords

What changed when Midjourney moved from V6 to V7

Midjourney released V7 on April 3, 2025, and its documentation describes the model as more precise with text and image prompts, richer textures, and more coherent details in areas such as hands, bodies, and objects. V7 also introduced Draft Mode and Omni Reference. The later V8 releases changed the current default, but V7 remains important when a workflow depends on Omni Reference or a saved V7 prompt format. Read the version history in Midjourney's documentation.

The important migration lesson is that a familiar V6 token is not automatically a useful V7 control. Some parameters remain valid with changed ranges, some features have moved to a new parameter, and some workflows still render through an older model behind the scenes. Treat the prompt as a small configuration file. Check the version, feature compatibility, parameter position, and intended output before changing the prose.

Parameter placement is the first prompt migration test

Midjourney's parameter documentation says parameters belong at the end of the text prompt. Keep a space before the two hyphens, avoid punctuation inside the parameter block, and do not add prompt text after the parameters. This rule matters more than adding another descriptive adjective because a misplaced control can be parsed as ordinary text or ignored.

A dependable V7 pattern is: subject, action, environment, composition, lighting, then parameters. For example, editorial portrait of a ceramic artist in a sunlit studio, hands shaping clay, natural documentary photography --ar 4:5 --v 7 --raw. The descriptive clause comes first. The technical controls are grouped at the end, where they are easier to inspect and revise.

Do not write --ar 4:5, with a trailing comma, split the dashes with spaces, or place a new subject after --v 7. If a saved V6 prompt uses punctuation in its parameter block, clean that block before testing image quality. The official Parameter List is the right reference for current names and aliases.

Ten V6 prompt habits that need a V7 review

The phrase “break” should be read carefully. A legacy keyword may be rejected, ignored, or simply fail to produce the effect the writer expected. Midjourney's live documentation is the authority, not a community list of supposedly banned words. Review the following habits one at a time rather than replacing every token in a single edit.

Replacing Character Reference with Omni Reference

Omni Reference is the central V7 replacement for Character Reference. Midjourney says it can place a person, object, vehicle, or non-human creature from a reference image into a new creation. It accepts one image and uses the --oref parameter in Discord. The web interface has a dedicated Omni Reference area in the image panel.

The feature has important constraints. Omni Reference is compatible with V7, costs twice as much GPU time as a regular V7 image according to the documentation, and is not compatible with Draft Mode, Conversational Mode, Fast Mode, or --q 4. Its results are also not directly compatible with Vary Region, Pan, or Zoom Out. If you need those editing tools, load the result into the Editor after removing the image reference and its related parameters.

Use Omni Reference when the identity or physical form of a reference matters. Do not use it as a substitute for a Style Reference. A character or product image provides subject information, while a Style Reference supplies a visual direction such as color, medium, lighting, or texture. The official Omni Reference guide also advises keeping the text prompt clear because the image alone does not describe the entire scene.

Using Omni Weight without overfitting the reference

The --ow parameter controls how much detail from the Omni Reference appears in the result. Midjourney documents a range from 1 to 1,000 with a default of 100, and recommends staying below 400 unless a high stylize value changes the balance. Start near the default and change one variable at a time.

Higher weight is not automatically more accurate. An image may contain freckles, logos, or small text that the model cannot reproduce exactly. State the important physical features in the prose, then use Omni Weight to adjust influence. This is a controlled test, not a promise of pixel-level copying.

Draft Mode is a speed tool, not an Omni Reference replacement

Midjourney lists Draft as a V7 parameter that generates draft images at half the GPU cost. That makes it useful for exploring composition, subject placement, and prompt direction before paying for a final render. It does not make a poor prompt precise, and it is not compatible with Omni Reference according to the Omni Reference documentation.

A practical workflow is to begin with a text-only prompt and --draft, compare the composition, then remove Draft for the final V7 render. Keep the same core wording while changing only the parameter block. If the draft fails because the subject, camera angle, or spatial relationship is unclear, rewrite the text before running a final job.

Draft Mode should not be described as a universal “fast mode.” Midjourney separately documents Fast, Relax, and Turbo GPU modes. Draft is a V7 image-generation setting with its own compatibility rules. That distinction prevents a migration guide from promising a speed or cost result that depends on the account, mode, prompt, and final workflow.

Style Reference migration from V6

Style Reference captures the overall visual feel of an image, including colors, medium, textures, and lighting. It does not copy the people or objects in that image. The feature remains available in V7, but Midjourney notes that Style Reference was updated between V6 and V7, so older style codes may not produce the same appearance.

For a V6 prompt that depends on a familiar style code, test the code with --sv 4 when you need the older V7 style-reference model, or rebuild the look with a new reference image and a short text prompt. The documentation recommends focusing on the desired content rather than writing instructions such as “copy this style.” Use descriptive terms only when they resolve a real mismatch. This measured approach is similar to the evaluation discipline used in reasoning-model comparisons.

Keep --sref and --sw at the end of the prompt. The default Style Weight is 100 and the documented range is 0 to 1,000. A lower weight can help when the reference is overpowering the subject, while a higher value can make a consistent campaign style more visible.

What to do with quality, aspect ratio, and raw controls

Not every V6 control needs to be deleted. Aspect ratio, Seed, Chaos, Raw, Stylize, Tile, Weird, Image Weight, and Style Reference remain part of the documented parameter set. The migration task is to check each parameter's current range and purpose, not to treat every double-hyphen token as obsolete.

Quality deserves special care. The Version documentation shows different quality options across models, and the Parameter List describes quality as a control for detail and processing time. Do not assume that --q 2 has the same cost or output effect across versions. Record the exact version and parameter when comparing results.

Prompt migration examples for common image tasks

For a portrait with a reference subject, replace the old Character Reference pattern with a V7 prompt that names the subject, scene, and style, then adds --oref and a moderate --ow. For a product campaign, use the reference image for the object and describe the material, camera angle, background, and intended crop in text. For a moodboard-led concept, use Style Reference instead of Omni Reference.

For a fast composition test, use a text-only V7 prompt with --draft. For a final render, remove --draft and retain only the controls that improved the test. For a legacy style code that changed, compare a new V7 Style Reference against --sv 4. These examples are deliberately narrow because a migration test should reveal the effect of one change. Creators who are still learning prompt structure can pair this workflow with the site's beginner prompt engineering guide.

Do not promise that a new parameter will reproduce an old image. Model version, reference image, seed, account settings, and feature compatibility all affect the result. The correct outcome is a reproducible test record, not a claim that V7 has a single replacement keyword for every V6 workflow.

A troubleshooting checklist when a V7 prompt fails

Start by checking that the selected model is V7 and that all parameters are at the end. Remove punctuation from the parameter block, confirm the space before each double hyphen, and test the shortest prompt that still describes the task. Then add a single reference or control. This isolates syntax problems from prompt-quality problems. If the failure is happening inside a larger application workflow, compare the symptoms with the site's AI tool troubleshooting guide without assuming the causes are identical.

Next, check compatibility. Omni Reference cannot be combined with Draft Mode, Fast Mode, Conversational Mode, or --q 4. A V7 prompt that uses Multi-Prompt syntax or Character Weight needs a structural rewrite. A style code that looks wrong may need a different Style Reference version rather than a larger style weight.

Finally, compare outputs using the same subject, ratio, and seed where practical. Save the exact prompt, version, reference image, parameter values, and result notes. If the output is still wrong, change the prose to specify the subject relationship, camera, lighting, or unwanted elements. Random keyword accumulation makes the cause harder to identify.

When a V6 prompt should remain on V6

Migration is not always the best choice. If a production archive depends on Character Reference, Multi-Prompt weighting, or a specific V6 style-code behavior, keep the original workflow documented and render it on the version it was designed for. Recreate it in V7 only after you have a test set and a clear acceptance criterion.

Version choice is also affected by the tool chain. Omni Reference is a V7 feature, while some editing operations still use a legacy editor behind the interface. A workflow that requires a particular editing control may need a two-step process or a different reference strategy. Read the current compatibility tables before promising that a single V7 prompt can replace the entire pipeline.

How to document a reliable V7 prompt system

Use a small prompt record with five fields: task description, model version, text prompt, parameter block, and result notes. Add reference-image identifiers when using Omni Reference or Style Reference. Record the reason for each parameter so another editor can remove it without guessing. This turns prompt migration into an auditable editing process. A broader model inventory can be found in the site's AI models guide.

For teams, keep reusable prompt components separate from version-specific controls. The scene description can remain stable, while --v 7, --oref, --ow, --sref, --sv, and --draft stay in a tested parameter profile. Review that profile against Midjourney's live documentation whenever the default model changes.

Conclusion: migrate by testing the workflow, not the keyword list

Midjourney V7 prompt migration works best as a compatibility review. Move Character Reference workflows to Omni Reference, use Draft Mode only for text-led exploration, check Style Reference version behavior, and keep every parameter at the end of the prompt. The safest replacement for a broken V6 keyword is often a clearer description and a controlled test, not another keyword.