Overview
FLUX is, in our evaluation, the open image model that finally gave Midjourney a real run for its money — without the Discord, without the subscription lock-in. Built by Black Forest Labs (the team behind the original Stable Diffusion), it arrived in variants — Schnell for speed, Dev for quality, and Pro/1.1 via API — and reset the bar on photorealism, hands, and in-image text. Because the weights are open, you can run FLUX on your own hardware or call it through hosts like Replicate, which means the “free tier” is real in a way closed tools cannot match.
The honest trade-offs are weight and licensing. The full model is heavy to run locally — you need serious GPU VRAM or a hosted endpoint — and commercial use of the Dev/Pro weights requires a license, which trips up teams that assume “open” means “free for business.” Our evaluation is that FLUX is the right tool when you want Midjourney-grade output with self-hostable control and excellent prompt adherence, and the wrong tool when you need zero-setup convenience or unrestricted commercial rights out of the box. It is less stylized out of the box than MJ but more controllable than vanilla Stable Diffusion, and it is the backbone of most modern ComfyUI setups.
Key Features
- High photorealism and detail. Sharper hands, skin, and textures than SDXL; rival to closed models on faces and product shots.
- Strong prompt following. Renders spatial relationships and counts reliably, closer to DALL·E-level adherence than older open models.
- Open and Pro model variants. Schnell (fast, Apache-2), Dev (quality, non-commercial), and Pro/1.1 via API for best results.
- Great for portraits and scenes. The default aesthetic is clean and photographic, ideal for product and people photography.
- Widely available via APIs and apps. Hosted on fal, Replicate, and integrated into editors like getimg.ai and Leonardo AI.
Pricing
| Plan | Price | For |
|---|---|---|
| Open (Schnell) | $0 | Self-host, Apache-2 |
| Dev (self-host) | $0 | Non-commercial, needs GPU |
| Pro / 1.1 API | Usage-based | Best quality, commercial ok |
Tier details: The open weights cost nothing to download, but you pay in VRAM and electricity if you self-host, or in API credits if you use a host. Schnell is free even commercially; Dev is free but non-commercial without a license; Pro/1.1 through Black Forest Labs or partners is usage-based and clears commercial use. For most teams, the pragmatic path is a hosted endpoint — you get FLUX quality without a 24GB GPU. Watch the license: shipping Dev weights in a commercial product without permission is the easy mistake to make.
How It Compares
vs. Midjourney: MJ is more stylized and zero-setup; FLUX is controllable, self-hostable, and license-flexible (with the right tier). Pick FLUX when you need the weights.
vs. Stable Diffusion: FLUX is the spiritual successor — cleaner, better hands and text, but heavier. SD remains lighter and more battle-tested for niche controlnets.
vs. Adobe Firefly: Firefly is the enterprise-clean, indemnified choice; FLUX is the open, technically superior-but-license-aware alternative.
Getting Started
- Try FLUX on a host (Replicate or getimg.ai) before self-hosting — confirm the look fits your need.
- If self-hosting, start with Schnell on modest GPU; jump to Dev only when you have the VRAM.
- Run Flux.1 Dev with a LoRA for your subject and a short negative prompt — the base is so clean you rarely need heavy prompting.
- Use explicit spatial language; FLUX follows it better than older open models, so “left,” “behind,” “two” actually work.
- For product shots, pair with a controlnet or precise inpainting in ComfyUI for repeatable frames.
- Check the license tier before any commercial launch — Schnell is safe; Dev is not.
- Keep prompts concise; unlike MJ, FLUX rewards clarity over florid description.
Hands-on Verdict
Our honest evaluation: FLUX (by Black Forest Labs) is the open image model we run for photoreal results that rival Midjourney without the Discord. The Schnell/Dev split gives a real speed-or-quality trade-off, and it has become the backbone of most ComfyUI setups — sharper hands and text than SDXL. It is less stylized out of the box than Midjourney and more controllable than vanilla Stable Diffusion.
Who it is for: self-hosters, product photographers, and developers who want weights they control. Practical tip — run Flux.1 Dev with a LoRA for your subject and a short negative prompt; the base model is so clean you rarely need heavy prompting, unlike older SD checkpoints. The only real friction is the GPU requirement and the commercial-license caveat, both solved by a hosted Pro endpoint. For open-weight realism, FLUX is the one to build on. For teams, the hosted Pro endpoint removes both the GPU requirement and the licensing worry at once, which is the path we suggest over wrestling Dev weights in production — you keep the control story without the infrastructure bill.