FLUX

Overview

FLUX by Black Forest Labs (the original Stable Diffusion creators) is the open-source challenger to Midjourney. FLUX.1 Pro produces photorealistic images with excellent prompt adherence and can be used free via Hugging Face or cheaply via Replicate and Fal.ai APIs.

My Experience with FLUX

I started testing FLUX not long after Black Forest Labs launched it, mostly out of curiosity about whether an open-source model could genuinely compete with Midjourney's polish. A year of real project use later, my answer is: yes, for certain jobs, unambiguously — and for others, no, not close. Here's the honest breakdown of where FLUX has earned a permanent spot in my workflow, and where I still open Midjourney instead.

Built by the people who built Stable Diffusion

The backstory matters more than most reviews give it credit for. Black Forest Labs was founded by the researchers behind Stable Diffusion — the people who originally proved open-source image generation could be a serious category. When they left to build FLUX, they applied everything they'd learned about what made Stable Diffusion good and what held it back. You can feel that lineage in the output: FLUX doesn't have the "obviously open-source" rough edges of earlier community models — hands are right, text rendering is dramatically better, and busy scenes hold together in a way Stable Diffusion rarely managed.

The per-image cost math that changes everything at scale

This is the reason FLUX is in my regular rotation. Running FLUX.1 Pro through Fal.ai costs around $0.003 per image in my typical batch jobs, with higher-end configurations topping out around $0.055. Compare that to Midjourney, where a subscription works out to roughly $0.01 or more per image at realistic usage, and the gap becomes obvious once you're generating anything beyond a handful of images a month. For a blog needing a dozen header images a week, or a product generating a thumbnail per user action, that difference compounds fast. I've moved every high-volume, programmatic image job I have onto FLUX for exactly this reason.

Photorealism and prompt adherence that actually hold up

Where FLUX consistently impresses me is photorealistic output and how closely it follows a detailed prompt. Ask for a specific number of objects, a particular lighting setup, or a precise pose, and FLUX gets noticeably closer to what I actually asked for than earlier open models did — competitive with Midjourney on this specific axis. For product mockups, realistic portraits, and scene compositions where accuracy matters more than stylization, I trust FLUX to interpret the prompt literally rather than "interpreting" it into something prettier but different from what I wanted.

Running it completely free with ComfyUI

With a decent GPU — I've run this comfortably on an RTX 3080 and above — you can skip the API entirely and generate images locally through ComfyUI at effectively zero marginal cost. No per-image billing, no rate limits, no sending prompts or outputs to a third party. This is what I point people toward when privacy is the actual requirement: client work involving sensitive concepts, early-stage product designs that shouldn't leak, or just generating volume without watching a usage meter. The tradeoff is upfront setup time and the GPU cost itself, but once running, it's the cheapest option here by a wide margin.

A fine-tune and LoRA ecosystem that's still exploding

Because the base models are open, the community around FLUX has taken off faster than I expected. New fine-tunes and LoRAs targeting specific aesthetics — anime styles, particular photography looks, brand-specific product shots — show up on a weekly basis. Practically, if the base model's default aesthetic doesn't match what you need, someone has probably already trained a LoRA that gets you most of the way there, and you can layer it into your ComfyUI pipeline without retraining anything yourself. Midjourney has no equivalent — you get its house style and official variations, full stop.

FLUX vs Midjourney: two very different tools

I don't think of these as direct competitors so much as tools for different jobs, and my own usage splits cleanly along that line. Midjourney is still my choice for artistic, exploratory work — moodboards, illustration, anything where I want the model's aesthetic judgment to surprise me, generating one image at a time in a browser tab. FLUX is my choice the moment the job becomes photorealistic, high-volume, or programmatic — anything running through an API, anything where cost per image matters, anything needing the output to match my exact prompt rather than a house style.

Honest weaknesses

FLUX is not beginner-friendly. There's no polished, single web interface the way Midjourney has its Discord-turned-web-app experience — you're choosing between Hugging Face Spaces, a Replicate or Fal.ai API, or setting up ComfyUI yourself, and none of those are one-click for someone who just wants to type a prompt and get an image. For a casual, one-off image, Midjourney's interface will get you there faster with less thinking. And for pure artistic quality — the striking, stylized "wow" image people share — Midjourney still leads. FLUX's strength is accuracy and photorealism, not artistic flair.

Is FLUX worth it? The pricing verdict

For high-volume or API-driven use, FLUX is close to a no-brainer — at $0.003 to $0.055 per image via Fal.ai or Replicate, it's dramatically cheaper than a Midjourney subscription once you're generating more than a handful of images regularly, and running it locally through ComfyUI takes the marginal cost to essentially zero if you already own the GPU. For casual, occasional use where you want the least friction possible, Midjourney's subscription is still the more sensible spend. I keep both: FLUX for anything programmatic or photorealistic, Midjourney for anything artistic or one-off.

Pricing