Midjourney is a $30-per-month subscription. Fast. Beautiful outputs. But you don't own the images, the API can disappear, and every prompt burns through a metered allowance. Run Stable Diffusion XL or Flux locally on an M1/M2/M3/M4 Mac, and you eliminate the entire cost structure. Zero subscription. Zero metering. You own every pixel. The trade-off: 30–90 second generation times instead of 10 seconds, and you do the aesthetic editing locally in Affinity Photo (which is also a one-time purchase, not a subscription). For design ideation, mockup backgrounds, hero illustrations, and brand exploration, local diffusion models are indistinguishable from cloud in the output quality—and infinitely cheaper. This is the Aidxn workflow.
Why Local Image Generation Wins
Three things changed in 2026. First, Flux (and SDXL's refinement releases) hit parity with Midjourney for commercial design work. Second, Apple Silicon made inference fast enough — M3/M4 Macs generate a 1024×1024 image in 30–60 seconds, which is acceptable for ideation loops. Third, open-source model licensing got clearer: SDXL and Flux are both commercially licensed. You can generate images, use them in client work, sell them, and Stability AI doesn't claim ownership. That's the permission structure Midjourney doesn't offer.
The real cost of Midjourney isn't just $30/mo—it's lock-in. Generate a hero image. Love it. Six months later, Midjourney changes the model, your prompt breaks, you can't regenerate it. With local models, the weights are on your disk. You version-control them. You own the generation forever. And since there's no API cost, you iterate freely. Need 50 variations of a hero banner for A/B testing? Local costs zero. Midjourney costs $30–50. That delta justifies the setup time in week one.
SDXL vs Flux — Which Model to Run
SDXL (Stable Diffusion 3.5) — Battle-Tested, Predictable
SDXL has been the standard since 2023. It's stable, tuned, and has millions of LoRAs (style adapters) available online. You can generate photorealistic products, stylized illustrations, web design mockups—it handles everything. On M-series Macs, SDXL runs in 40–60 seconds per 1024×1024 image. If you need consistency, SDXL with a fixed seed produces identical results. The aesthetic is slightly softer than Flux, but for mockup backgrounds and brand exploration, it's perfect. Aidxn uses SDXL for 80% of local generation because the ecosystem is massive (LoRAs for specific art styles, character consistency, textures). Model weight: ~6.8GB. Fits on any M1 with 16GB RAM.
Flux (Open-Source Release) — Bleeding Edge, Faster
Flux just launched as open-source in mid-2026. It's faster than SDXL (30–40 seconds), slightly sharper outputs, and handles text-in-image better. The downside: fewer LoRAs exist yet, and you're running a newer model with fewer quirks documented. Flux is the pick if you want speed and don't mind doing more manual editing in post. Model weight: ~24GB for the full version (there's a pruned ~12GB variant that trades quality for speed). Needs M3 Pro / M4 with 20GB+ unified memory.
In practice: start with SDXL. The ecosystem is proven. If you hit iteration speed limits, swap to Flux. Both run on the same tools (Draw Things, ComfyUI, InvokeAI), so switching is one checkbox.
Three Tools — Pick Your Interface
Draw Things (Easiest) — Mac App, Zero Setup
Draw Things is a native macOS app. Download from the Mac App Store, launch it, and generate. It handles model downloading, prompt engineering, LoRA loading, and image export in one UI. Zero Terminal, zero config files. You type a prompt like "minimalist web design mockup, hero section, gradient background, product photo center", adjust the sampler and steps, and hit generate. 60 seconds later, you have an image. For designers who don't want to touch code, Draw Things is the move. You can batch-generate 10 variations and export them to a folder. The app costs $20 (one-time), but you save that in Midjourney fees in the first month. Aidxn uses Draw Things for quick ideation loops.
ComfyUI (Most Powerful) — Node-Based, Infinite Customization
ComfyUI is a node-based diffusion interface. You chain nodes (model loader → prompt encoding → sampler → image output) and dial in every parameter. It's intimidating at first but incredibly flexible. You can load multiple LoRAs, chain upscalers, add control-net conditioning (pose guidance, edge detection), and batch-process hundreds of images in one graph. If you need precision and reproducibility, ComfyUI is it. Setup: clone the GitHub repo, run ./main.py, and navigate to http://localhost:8188. Steep learning curve, but the output control is unmatched. Aidxn uses ComfyUI for batch hero-illustration generation and A/B testing.
InvokeAI (Web UI, Good Middle Ground) — User-Friendly, Feature-Rich
InvokeAI is a web-based diffusion UI. It bridges Draw Things (simplicity) and ComfyUI (power). You get a clean web interface, model management, LoRA loading, in-painting, and batch workflows. Setup: pip install invokeai, run the initialization script, and open http://localhost:9090. It's more capable than Draw Things and less of a learning cliff than ComfyUI. For teams or for people who want to script workflows, InvokeAI is the sweet spot. Aidxn often recommends InvokeAI to designer collaborators because the UI is accessible and it supports every feature they'll need.
Three Real Use Cases
Mockup Backgrounds & Hero Art
Client website. They need a hero section with a custom illustration. Midjourney: "That'll cost $15–30 for iterations and upscaling." Local diffusion: Free. You generate 20 variations in 20 minutes (at 60 seconds each), export the best three, touch them up in Affinity Photo (color grading, content-aware fill for text areas), and hand them over. The client's hero is photorealistic or stylized to spec. Cost: $0 in API fees, 30 minutes of your time. Same output quality as Midjourney, zero ongoing subscription.
A/B Testing Product Photography
E-commerce brand. You need 10 variations of a product photo (same product, different backgrounds, lighting, framing). Midjourney: 10 prompts × $0.50 ≈ $5–10, plus the $30/mo subscription. Local: Free and infinite. Generate 50 variations, pick the top 10, compare conversion rates locally (no cloud upload, zero privacy risk). The legal clarity here is crucial: you own every generated image completely, so using them in production or selling them is straightforward. Midjourney's ToS has gray areas about image ownership; local diffusion has none.
Brand Exploration & Style Ideation
Rebranding a design agency. You're exploring 10 different visual directions (minimalist, maximalist, geometric, organic, etc.). Midjourney costs spiral quickly because you're iterating on aesthetics, not generating final assets. Local diffusion lets you generate 100 images across 10 styles, export them all, and do side-by-side comparisons in Figma. Cost: zero dollars, one evening of exploration. This is where local generation shines—volume and iteration velocity without metering anxiety.
Workflow: Generate → Export → Polish
The modern local-gen workflow at Aidxn is three steps. One: open Draw Things or ComfyUI, write your prompt (be specific: "product photo, studio lighting, white background, 85mm lens, sharp focus"), adjust sampler steps (30–50 gives a good quality/speed ratio), and generate. Two: export the raw image. Three: open Affinity Photo, do selective editing: dodge/burn for lighting, content-aware fill if needed, color grading for brand alignment. The entire loop—ideation to polished export—takes 30 minutes for a hero image instead of 2 hours of Midjourney iteration + upscaling + prompt engineering. Time savings + cost savings = the reason every design team should have local diffusion in their toolkit by 2027.
Setup: 15 Minutes to Your First Image
If you're on M-series Mac: download Draw Things from the Mac App Store (or go free with ComfyUI or InvokeAI). Launch the app. It'll prompt you to download SDXL (6.8GB, takes 5 minutes on WiFi). Once downloaded, type a prompt like "modern minimal website hero section, gradient background, professional product photo center". Set sampler steps to 40. Hit generate. Wait 60 seconds. You now have a locally-generated image. No API key, no subscription, no data leaving your machine. That's it. From zero to first image is genuinely 15 minutes.
Six FAQs
Do these models create as good images as Midjourney?
Yes, for commercial design work. Flux and SDXL both handle photorealism, stylized art, and product photography at Midjourney-quality levels. The differences are subtle: Flux handles text-in-image slightly better, SDXL has more LoRA variety. For mockup backgrounds, hero illustrations, and brand exploration, they're equivalent. The only area where Midjourney still wins is edge-case prompts and niche aesthetics, but that's diminishing as Flux matures.
Isn't local generation slow?
Slower than Midjourney's 10 seconds, yes. A 1024×1024 image on SDXL takes 40–60 seconds. But you can generate in batches while doing other work, and the cost is zero, so it's acceptable. If speed is critical, Flux is 30–40 seconds. And unlike Midjourney, you can generate 100 images overnight without a subscription or rate limits—just queue them in ComfyUI and walk away.
Do I own the images?
Completely. You generate them on your machine, they live on your disk, you can use them in commercial work, sell them, modify them, and license them however you want. SDXL and Flux are both commercially licensed. Stability AI doesn't claim ownership. This is the primary advantage over Midjourney.
What Mac do I need?
M1 or newer. SDXL (6.8GB) runs on M1 16GB. Flux full version (24GB) needs M3 Pro or M4 20GB+. If you have an M2/M3 MacBook Pro with 16GB, you're set for SDXL and Flux pruned. Older Intel Macs won't work well. Apple Silicon is a hard requirement.
Can I use LoRAs to customize the style?
Yes. LoRAs (style adapters) are freely available from Civitai and Hugging Face. They're tiny files (50–500MB) that you load in Draw Things or ComfyUI. Want a specific artist's style, character consistency, or texture? Grab a LoRA, load it, and it shapes all your generations. ComfyUI and InvokeAI have full LoRA support. Draw Things can load them too. This is why SDXL's ecosystem is so rich—millions of LoRAs exist for it.
Should I abandon Midjourney entirely?
Not necessarily. Midjourney's niche strength is speed and UX simplicity. If you're generating images during client calls or need output in seconds, Midjourney wins. But for design teams doing daily ideation, batch work, or ownership-critical projects, local diffusion is cheaper and more flexible. Hybrid: use local generation for exploration and ownership-critical work, use Midjourney for real-time client-facing demos. That minimizes Midjourney fees while keeping the flexibility of both worlds.
The Bottom Line
SDXL and Flux generate design-quality images on your Mac for free. Run them via Draw Things (easy), ComfyUI (powerful), or InvokeAI (balanced). Forget the $30/mo Midjourney subscription and metering anxiety. You own every image. Iteration is infinite. Setup takes 15 minutes. For mockup backgrounds, hero art, brand exploration, and A/B testing, local image generation is the production workflow Aidxn uses every week. The only limit is your Mac's vRAM and your patience for a 60-second generation time—both are way more acceptable than perpetual API bills.
Building a design system that scales with local tools? Start with Aidxn Design services to set up a custom generation pipeline, or explore our other local AI workflows for internal tooling and cost control.