Midjourney Web Editor Inpainting & Reference Layers vs Stable Diffusion 4

“The inpainting actually produces coherent output — fix a hand, swap a background element, adjust a face without nuking the rest of the composition. That's the hard problem other inpainters fumble. The reference layer system is the real unlock: stack a character ref on top of a style ref and the model holds both with real fidelity, not a mushy average. The editing surface is brush-based with adjustable hardness, which is the right call — it matches how illustrators already think about masking. The one failure is the layer stack has no blend mode controls, so if your references fight each other, you can't arbitrate who wins.”

“The output question is everything here, and without a public gallery of SD4 video outputs I can't score the taste layer blind — but the improved prompt adherence claim is the right problem to fix, because SD3's notorious text-in-image failures made it genuinely unusable for real creative briefs. The taste layer is fully delegated to the user, which is the correct call for an open-weights model: Stability isn't trying to impose an aesthetic, they're giving fine-tuners the primitive to build one. The fingerprint concern is real though — 60-second video from a diffusion model still has the motion-texture-smoothness signature that screams AI to anyone who's seen more than ten generated clips, and no distillation trick fixes that. What earns the ship is the editing surface: open weights means LoRA, ControlNet, and every community extension will land within weeks, giving creators the iteration depth that closed-API tools like Runway will never offer.”

“The inpainting brush tool is actually designed — there's a clear mask preview in a distinct overlay color, an undo stack that doesn't blow away your full session, and the strength slider gives you real feedback as you drag, not just after you regenerate. What's missing is any visual hierarchy between the reference layer panel and the generation controls; they sit at the same visual weight and the eye has nowhere to land when you're deciding what to adjust next. The empty-state handling is also lazy — drop into a blank editor with no image loaded and you get a generic placeholder instead of a guided first action. Strong fundamentals, unfinished information architecture.”

“This is genuinely Midjourney catching up to Stable Diffusion workflows that have existed in ComfyUI and Automatic1111 for two years — credit where it's due for packaging it without requiring a local GPU and a PhD in node graphs. The specific scenario where this breaks is complex product photography: multi-layer references with fine texture like fabric or intricate logos still drift noticeably after inpaint cycles, which means professional retouching workflows aren't fully replaced yet. What kills this tool in 12 months isn't a competitor — it's Adobe Firefly and the Photoshop generative fill team, who now have a direct target to match feature-for-feature. Midjourney wins if their model quality gap holds; right now it does.”

“The direct competitors here are Wan2.1, CogVideoX, and Runway Gen-4 — so the market is not empty and Stability is not early. The scenario where this breaks is enterprise production: 60-second video at acceptable quality likely requires VRAM that most teams don't have on-prem, and the distilled mode probably trades quality for speed in ways that matter for commercial work. The 12-month prediction: this wins the hobbyist and fine-tuning community outright because it's open-weights and nobody else in that tier ships native video at this length — but Stability's monetization problem remains unsolved, and the API business stays under pressure from cheaper hosted alternatives. To be wrong about the ship, Stability would need to collapse operationally before the community forks and maintains the model independently — and at this point, the community would carry it regardless.”

“The thesis here is that non-destructive, multi-reference generative editing becomes a standard primitive in all creative software — not a specialty feature but a baseline expectation, the way layers were after Photoshop 3.0. Midjourney stacking inpainting and reference layers in the same session is a bet that the editing and generation workflows converge into a single surface, eliminating the round-trip between generator and editor that currently fragments creative pipelines. The second-order effect that matters: if this works at quality, it transfers creative leverage from production designers who own the toolchain to art directors and clients who only own taste — and that's a real power shift in agency workflows. The dependency that has to hold is Midjourney's model quality advantage over commodity diffusion endpoints; the moment that gap closes, the web editor is just a UI wrapper.”

“The thesis SD4 bets on is specific and falsifiable: by 2028, the majority of generative video production for indie creators and small studios will run on locally-deployed open-weights models rather than cloud APIs, because compute costs fall faster than API margins. The dependencies are two: consumer GPU VRAM continues its trajectory past 24GB at the $500 price point, and no foundation lab releases a comparably capable open-weights video model in the next 18 months. The second-order effect that matters most isn't the video itself — it's that open-weights video generation hands fine-tuning leverage to IP holders and brands who will never put their training data into a third-party API, unlocking a commercial fine-tuning market that closed-model providers structurally cannot serve. Stability is on-time to the open-weights image trend but genuinely early to the open-weights video trend — Wan2.1 is the only real prior art, and SD4's prompt adherence improvement is the specific technical delta that could make this the training base the community actually adopts.”

“The primitive here is a unified diffusion backbone that handles both image and video generation in a single model weight, which is actually a meaningful architectural decision rather than a bolted-on video pipeline. The DX bet is clear: put complexity at the hardware layer and keep the inference API surface identical to SD3, so existing ComfyUI workflows and diffusers integrations don't break. The moment of truth is pulling the weights from Hugging Face and running the distilled inference mode — if the 40% speed claim holds on a 4090 without quantization tricks, that's a genuine win. The weekend-alternative test is real: you can't replicate a 60-second native video model with three API calls and a Lambda, so the open-weights moat is legitimate. What earns the ship is that Stability actually put the weights on Hugging Face instead of hiding them behind an API — that's the specific decision that respects the developer.”