Midjourney claims new edit model improves character sheets

• Midjourney says its Edit Model has replaced OmniReference, according to midjourney's reply about Omni, while the official Edit Model documentation calls it the replacement for both Omni Reference and Character Reference.
• Character-led prompts can take up to four reference images, and midjourney directs requests for consistency to the new system in midjourney's consistency reply.
• The rollout includes character-sheet examples, as midjourney's character-sheet reply points out, alongside claimed scene transfers and editable text.
• Early feedback has not closed the gap: techhalla asked for character and spatial consistency from a reference image, and its later follow-up said that request followed testing the current tool.

Midjourney's Edit Model documentation lays out a four-image reference stack, targeted inpainting, and canvas expansion in the Editor. A community portrait test is using it to add hands and make subtle pose changes to archive photography.

Midjourney's position on Omni is direct.

The Edit Model accepts up to four uploaded reference images on the web, or image URLs after --edit in Discord. It works with Midjourney versions 8.1 and 8.2, and the Omni Reference docs now direct V8 users to Edit.

Four input slots turn the reference control into a small casting board: the documentation says creators can combine multiple elements into one new composition, alongside Image Prompts, Style References, Moodboards, Personalization, and --hd.

Midjourney pointed users asking about character sheets to this launch example.

The documentation names one multi-character failure mode: attributes can get mixed between characters. Its suggested workaround is to combine those characters into a single reference image, making a sheet or lineup a useful composite input rather than only a presentation asset.

Midjourney also claimed its launch examples already surpassed OmniReference, in midjourney's comparison with Omni, and asked for counterexamples where the older system performs better.

The Edit Model changes Midjourney's prompt posture. The docs allow direct instructions such as changing a subject, rather than only describing a desired scene; the Editor handles masked changes through inpainting and expanded frames through outpainting.

Midjourney says characters can be moved into different scenes and text can be added or edited, while also saying text is not perfect. The docs add three useful mechanics:

• The first uploaded image supplies the default aspect ratio, unless the prompt sets --ar.
• --raw reduces Midjourney's added creative flair, giving prompt text more influence.
• A prompt can request a new angle, including front and rear views of a character or object.

A request for layers drew a narrower answer from Midjourney.

The company asked whether the request meant a background and primary foreground, then said those might be prompted as separate images with the current model. That sits alongside the documented four-reference composition workflow, rather than a new layer-based project structure.