All tools
Image generators

Flux

By Black Forest Labs

Open-weight image model with photorealistic output and strong prompt adherence; runs on many platforms.

Visit FluxFreemium

Overview

Flux is Black Forest Labs' family of image models — built by ex-Stable Diffusion authors after they left Stability AI. It is the technical-quality leader of 2026: best-in-class prompt adherence, near-perfect anatomy and hands, photoreal output that rivals or beats DALL-E 3, and surprisingly good text rendering for a base model. There is no first-party Flux app. You access Flux through partners — fal.ai and Replicate for API/serverless use, Krea and Freepik for product UIs, ComfyUI for power users. The model itself comes in tiers: Schnell (fast, distilled, Apache-licensed open weights), Dev (high-quality, non-commercial open weights), and Pro / Pro 1.1 / Pro Ultra (closed, hosted-only, the highest quality). Pricing runs roughly $0.003-0.05 per image depending on tier and host. The honest tradeoff: Flux is the developer's and power user's image model. If you want a polished consumer app with mood boards and a community, use Midjourney or Ideogram. If you want to run image generation inside your own product, fine-tune on your brand, or get the highest technical quality out of any current model, Flux is the answer. The lack of a first-party UI is a feature for builders and a bug for casual users.

Best for

  • photorealism
  • self-hosted image generation
  • fine-tuning

Strengths

  • Best prompt adherence of any current image model — gets specifics right that others miss
  • Open weights (Schnell, Dev) — you can self-host, fine-tune, build into products
  • Photoreal output is genuinely competitive with or better than DALL-E 3
  • Text rendering is strong for a base model — close to Ideogram for short copy
  • Tiered model family lets you trade cost for quality cleanly (Schnell → Dev → Pro → Ultra)

Weaknesses

  • No first-party product — every user experience depends on a third-party host
  • Less "aesthetic" out of the box than Midjourney; needs more prompt work to feel art-directed
  • Licensing is multi-tiered and confusing (Schnell free, Dev non-commercial, Pro hosted-only)

Pricing

Schnell (open weights)

Free / self-hosted

Apache 2.0 licensed, distilled 4-step model. Run it on your own GPU or via fal.ai/Replicate at ~$0.003/image. Fastest tier; quality below Dev/Pro.

Dev (open weights, non-commercial)

Self-host free / ~$0.025/image hosted

Higher quality than Schnell, non-commercial license for self-hosting (commercial use requires a license from BFL). Hosted commercial use via fal.ai or Replicate is fine.

Pro 1.1

~$0.04/image

Closed model, hosted-only. The default production tier — strong prompt adherence, good text, photoreal quality. Available via fal.ai, Replicate, BFL API.

Pro Ultra

~$0.06/image

4MP outputs, raw mode for unstyled photoreal, the highest fidelity tier. Use when you need print-resolution or final-asset quality.

Krea / Freepik subscriptions

$10-30/mo

If you want a UI instead of API access, Krea and Freepik bundle Flux Pro into monthly subscriptions with credit pools.

Use cases

  • Image generation inside your own product

    Open weights plus serverless APIs at fal.ai and Replicate make Flux the default choice for SaaS image features.

  • Photoreal product and lifestyle photography

    Pro Ultra in raw mode produces photoreal output that does not have the "AI sheen" of DALL-E or older Midjourney.

  • Fine-tuning a brand-specific model

    Dev weights are open enough to LoRA-train on your product line or character set, then deploy on fal.ai.

  • Generations where prompt adherence matters more than vibe

    If the brief specifies exact composition, count, color, and pose, Flux follows it more reliably than Midjourney.

  • High-volume image pipelines on a budget

    Schnell at ~$0.003/image makes batch generation (100k+ images) economically viable in ways Midjourney cannot match.

  • ControlNet and edit workflows in ComfyUI

    The Flux ecosystem in ComfyUI is now richer than SD 1.5/SDXL was at peak — ControlNets, IP-Adapters, inpainting all mature.

When not to use

  • You want a polished consumer app with mood boards — use Midjourney
  • You need text rendering that beats every other tool — Ideogram is still slightly ahead on long copy
  • You need vector output — use Recraft
  • You want one subscription and no API/host setup — Midjourney or Ideogram are simpler

Alternatives

See it compared

Glossary terms to know

Other Image generators