What V8.2 Actually Changes
Midjourney V8.2 became the default model on July 24, replacing V8.1. The core focus of this release is aesthetic quality and personalization accuracy — outputs are described as more creative, bold, and sophisticated, with a significant reduction in the low-quality results that could appear unpredictably for certain prompts.
If you have been using Midjourney for a while, the most noticeable change will be that V8.2 consistently produces more refined results without the occasional rough drafts that were part of earlier versions. The personalization system — Midjourney learning your aesthetic preferences over time — also works more accurately in V8.2, especially for users who have built up large rating histories.
How V8.2 Differs From V8.1
V8.1, which remains available as a non-default option, brought native 2K HD output without a separate upscaling step — generation speed 50% faster and 25% cheaper per image. Text rendering also improved meaningfully in V8.1. V8.2 builds on that foundation with better aesthetic judgment and personalization rather than introducing new generation mechanics.
For users who rely heavily on style references or mood boards, an important caveat: features like –sref, –cref, –hd, and –q 4 run 4x slower on V8 and consume 4x the GPU time. So while base generation costs dropped in V8.1, users who apply style references frequently may find their effective capacity lower than the headline price suggests.
Draft Mode: The Workflow Change Worth Knowing
Draft Mode, available since V7 and improved in V8.1, renders at approximately 10x normal speed with GPU costs halved. It generates 24 low-resolution sketches in the time a standard render produces one full-quality image. This changes how you approach composition: rather than refining a single prompt iteratively, you can generate a broad set of variations quickly, identify the strongest direction, and upscale only that one.
On the web interface, the conversational workflow lets you adjust images by describing changes in plain language — say change the background to a nighttime scene and Midjourney applies it to the current generation without regenerating from scratch. Voice input including Chinese is supported for these adjustments.
The Image Editing Model: V8.2 Most Significant Capability Addition
Released August 28, the V8.2 image editing model is a major new capability that breaks from Midjourney text-only generation history. Four core features: Instruction tuning — send a text modification instruction for an existing image rather than regenerating from a new prompt, changing style, elements, or scene without starting over. Multi-image reference generation — combine up to 4 reference images simultaneously for style, element, and lighting fusion, with compatibility reportedly over 60% higher than previous approaches. Inpainting with Vary Region — precisely select any area of an image for targeted modification without disturbing surrounding areas. Outpainting — extend canvas dimensions in any direction without the hard edges and content discontinuities that plagued earlier versions.
Critically, V8.2 image editing is fully compatible with existing personalization profiles and style presets — you do not need to rebuild your style library for the new model.
How to Access Everything
V8.2 applies automatically as the default when you submit a prompt — no parameters needed. Image editing features are accessible on both the web interface and Discord. If you want to stay on V8.1, you can explicitly select it in model settings. For image editing specifically, check that your current plan includes access to the editing module — this feature rolled out progressively and may require a specific plan tier depending on your subscription level.