A family of image generation AI models developed by Black Forest Labs in Germany. Founded by former core developers of Stable Diffusion, the company released the first FLUX.1 (12 billion parameters) in August 2024, drawing attention for its strong prompt adherence, accurate in-image text rendering, and generation speed. Its defining trait is a dual distribution strategy: commercial models served via API, and open-weight versions you can run on your own GPU. The lineup has since evolved through FLUX 1.1, the image-editing-focused FLUX.1 Kontext, and the FLUX.2 generation in 2025, and in July 2026 the company announced FLUX 3, a multimodal model handling image, video, and audio in a single architecture. The models are also easily accessible through inference platforms such as fal.ai and Replicate.
Key Features
- Strong prompt adherence and text rendering: Handles complex composition instructions and in-image typography well, making it good at generating “readable” images such as posters and banners. Its photorealistic output is also highly regarded
- Model tiers for different needs: The FLUX.2 generation offers tiers with different quality, speed, and licensing: top-quality [max], production-API [pro], developer-oriented [flex], open-weight [dev], and lightweight Apache 2.0-licensed [klein]
- Open weights for self-hosting: The [dev] weights are published on Hugging Face, allowing you to run and fine-tune (LoRA training) on your own GPU. [klein] is Apache 2.0 licensed and free for commercial use
- FLUX.1 Kontext for image editing: A dedicated model specialized in text-instructed editing of existing images (inpainting, style changes, subject consistency)
- Instant access via inference platforms: Widely adopted by hosting services such as fal.ai, Replicate, and Together AI ─ just sign up and start generating via API
- Going multimodal with FLUX 3: Announced in July 2026, FLUX 3 supports video generation of up to 20 seconds with synchronized audio, alongside a robotics application (FLUX-mimic). Image and open-weight versions are planned to follow
Pricing
Image generation on the official BFL API (bfl.ai) is pay-per-image, with unit prices varying by model tier and output resolution.
| Model | Reference price (per image) | Positioning |
|---|---|---|
| FLUX.2 [klein] | From ~$0.014 | Lightweight, low-cost. Apache 2.0 open-weight version also available |
| FLUX.2 [pro] | From ~$0.03 | Standard tier for production use |
| FLUX.2 [flex] | From ~$0.05 | Developer-oriented, with more parameter control |
| FLUX.2 [max] | From ~$0.07 | Top-quality tier |
| FLUX.2 [dev] | ─ (self-hosted) | Open weights. Free for research/non-commercial use; separate license for commercial use |
Commercial self-hosting licenses are offered as monthly plans (Builder / Platform / Professional / Enterprise) that include fine-tuning and LoRA rights; contact the company for detailed pricing. When using fal.ai or Replicate, each platform’s own pricing applies.
Pricing is as of August 2026. Please check the official website for the latest pricing.
Pros & Cons
✅ Pros
- The choice between API and open weights supports everything from prototyping to production on your own infrastructure
- High prompt adherence and in-image text accuracy make it well suited for practical design assets
- Transparent pay-per-image API pricing ─ pay only for what you use, with no subscription required
- [klein] is Apache 2.0 licensed and can be embedded freely, even commercially, at no cost
- Broad platform support (fal.ai, Replicate, and more) gives you many adoption paths
⚠️ Cons
- It is not a consumer app with a chat UI like ChatGPT; usage is primarily via API and developer tools
- The default license for [dev] open weights is non-commercial; commercial self-hosting requires a paid license
- The many model tiers and license types can be confusing at first
- The latest FLUX 3 (video/audio) is still in early access as of August 2026; general availability is yet to come
Comparison with Similar Services
| Comparison | FLUX | Midjourney | Stable Diffusion | GPT Image |
|---|---|---|---|---|
| Provider | Black Forest Labs | Midjourney | Stability AI | OpenAI |
| Open weights | Yes ([dev] / [klein]) | No | Yes | No |
| Main usage | API / self-hosted | Web / Discord | Self-hosted / API | ChatGPT / API |
| Strengths | Prompt adherence, text rendering, flexible distribution | Artistic styles, community | Rich ecosystem, customizability | Conversational generation and editing |
| Pricing model | Pay-per-image | Monthly subscription | Free (own GPU) / API usage-based | ChatGPT plans / API usage-based |
Who Is It For
- Developers who want to integrate image generation into their own services or apps via API
- Teams that want to run generation infrastructure on their own GPUs, down to fine-tuning
- Creators of banners, posters, and thumbnails who need accurate text inside images
- Anyone who wants the open-weight escape hatch instead of depending solely on closed SaaS
Summary
FLUX stands apart from other image generation models by offering both a commercial API and open weights as twin pillars. Its prompt adherence and text rendering are production-grade, and developers can adopt it incrementally ─ start with the API, then move to self-hosting as you scale. Since the tier and license structure is somewhat complex, check each model’s license terms in the official documentation before embedding it in commercial products.