Skip to main content
Back

Midjourney's Edit Model Now Applies Reusable Style Controls to Existing Images

Midjourney's Edit model now accepts Moodboards, Style Codes, and Style References as drop-in style inputs, no text prompt required.

  • Midjourney's Edit model now accepts Moodboards, Style Codes, and Style References as direct inputs.
  • Attach an image plus a style carrier under "attach to prompt" and generate with no text prompt required.
  • Composition stays locked to the source image while the aesthetic swaps to match the reference.
  • Works on V8.1 and V8.2, stackable with Image Prompts, Personalization, and --hd.
  • Fixes prior friction where srefs and moodboards needed explicit prompt scaffolding inside Edit.
  • Live now for subscribers via web and Discord, no new pricing tier.

Midjourney has expanded its Edit model so Moodboards, Style Codes, and Style References can directly control the look of an existing image. In the web interface, users can place a source image under Attach to prompt, add a style input, and generate without writing a text prompt.

Midjourney’s announcement also demonstrates composition preservation. The edited result retains the source image’s overall arrangement while adopting a different aesthetic, making it easier to compare styles against a fixed layout.

Style inputs now work directly

The Edit model handles retexturing, inpainting, outpainting, and compositions assembled from multiple reference images. It uses an existing image as the structural starting point, then changes selected regions, extends the canvas, or applies a new visual treatment.

Earlier versions treated attached images mainly as editing references. Earlier guidance warned that moodboards and Style References could require explicit prompt language to produce their intended look. The update allows those reusable style inputs to drive an edit more directly.

Three ways to carry a look

Input What it carries Typical use
Style Reference One or more reference images supplied through --sref. Transfer visual traits from a specific image.
Style Code A reusable numeric identifier, such as --sref 2847361. Repeat or share a discovered Midjourney style.
Moodboard A style profile built from a curated image collection. Apply broader art direction drawn from several examples.

A source-and-style workflow

  1. Attach the source. Add an existing image and select Attach to prompt.
  2. Add the style. Choose a Style Reference, Style Code, or Moodboard. Multiple style inputs can be combined.
  3. Describe the target when needed. A phrase such as hand-drawn pencil sketch can reinforce the intended treatment and improve consistency.
  4. Generate and compare. Keep the source fixed while changing style inputs, prompts, or weights.

On Discord, the Edit model accepts --edit [urls] with up to four image URLs on V8.1 and V8.2. The edit path can also combine Image Prompts, Style References, Moodboards, Personalization, and --hd.

Fixed layouts make variants cheaper

  • Restyle product shots and hero images while retaining their composition.
  • Test one character or scene across several visual directions.
  • Compare style codes against the same layout during visual A/B testing.
  • Apply shared moodboards or brand styles during an edit pass instead of regenerating the scene.
  • Blend multiple Moodboard, Style Reference, and Personalization inputs into a reusable treatment.

Where edits can still drift

Composition preservation targets the overall layout, and pixel-level geometry can still vary between runs. Current Edit model workflows require V8.1 or V8.2 and do not support Remix or tile mode.

Numeric style codes are model-dependent. A code saved under an older version can produce a different look after a style-system update. Production workflows can reduce surprises by pinning the model version and storing representative outputs alongside each code.

Style strength remains a manual control. With --sw, useful starting ranges are 150 to 200 for image-based references and 120 to 150 for numeric codes when the effect is weak. Values from 50 to 80 can preserve more of the source when a style overwhelms the subject. Model version and source-image complexity can shift those ranges.

Access and GPU cost

The expanded style controls are available to Midjourney subscribers through the standard web and Discord interfaces, with no separate add-on. Each edit consumes GPU time, and combining Omni Reference with an sref costs twice the usual GPU allocation. The announcement describes web and Discord workflows; no public API change was announced.