FLUX.2
Production-grade AI image generation from Black Forest Labs delivering 4MP photorealistic outputs with multi-reference control and open weights.
Overview
FLUX.2 is the flagship model from Black Forest Labs, founded by ex-Stability AI researchers who built much of the original Stable Diffusion architecture. Launched in January 2026 after BFL's $300M Series B, FLUX.2 sits at the top of image generation leaderboards for photorealism, prompt adherence, and structural accuracy. It already powers production image pipelines at Meta, Canva, and Adobe.
What sets FLUX.2 apart is production readiness. The model generates native 4-megapixel images — sharp enough for print without upscaling — and includes built-in control mechanisms for depth, canny edges, inpainting, and outpainting. The multi-reference feature lets you condition outputs on several source images simultaneously, giving art directors and product teams precise control over style and composition without cobbling together external tools.
BFL offers three model tiers: FLUX.2 Pro for maximum quality via API, FLUX.2 Dev with open weights for developers who want to self-host or fine-tune, and FLUX.2 Schnell as a fast, lightweight variant for real-time applications. The open-weights approach has made FLUX.2 a favorite in the ComfyUI and diffusion community, where it's rapidly replacing older Stable Diffusion checkpoints as the default backbone.
Key features
4MP Photorealistic Output
Generates native 4-megapixel images with sharp detail and accurate textures. Print-ready quality without needing a separate upscaler.
Multi-Reference Control
Condition image generation on multiple reference images simultaneously — combine style, pose, and composition cues from different sources in a single output.
Structural Controls
Built-in depth maps, canny edge detection, inpainting, and outpainting. Guide composition precisely without relying on third-party extensions.
Open Weights (Dev & Schnell)
FLUX.2 Dev and Schnell variants ship with open weights. Self-host on your own GPUs, fine-tune on proprietary data, or integrate into custom pipelines.
Pricing
Free tier: Open-weight Dev and Schnell models can be self-hosted at no cost
| Plan | Price | What's included |
|---|---|---|
| API (Pro) | ~$0.05/image | Maximum quality, 4MP output, all control modes, commercial license |
| API (Dev) | ~$0.025/image | Near-Pro quality, faster generation, suitable for prototyping |
| API (Schnell) | ~$0.01/image | Fastest variant, good for real-time and high-volume use cases |
| Self-hosted | Free | Open weights for Dev & Schnell — run on your own hardware, fine-tune freely |
| Enterprise | Custom | Dedicated capacity, SLA, priority support, custom licensing |
Maximum quality, 4MP output, all control modes, commercial license
Near-Pro quality, faster generation, suitable for prototyping
Fastest variant, good for real-time and high-volume use cases
Open weights for Dev & Schnell — run on your own hardware, fine-tune freely
Dedicated capacity, SLA, priority support, custom licensing
Pros & cons
Pros
- ✓Native 4MP resolution produces print-quality images without upscaling
- ✓Open weights let you self-host, fine-tune, and inspect the model
- ✓Built-in structural controls (depth, canny, inpainting) without third-party tools
- ✓Strong prompt adherence — handles complex multi-subject scenes accurately
Cons
- ×Pro model is API-only with no open weights
- ×Requires beefy hardware for self-hosting (24GB+ VRAM recommended)
- ×Less stylized aesthetic than Midjourney — optimized for photorealism over artistic flair
- ×Smaller community ecosystem than Stable Diffusion despite growing fast
How it compares
| Tool | Best for | Pricing | Score |
|---|---|---|---|
| FLUX.2 | — | API ~$0.01-$0.05/image + free open weights | 9.2/10 |
| Midjourney | — | From $10/mo | 9.4/10 |
| DALL-E 3 | — | Freemium | 9.1/10 |
| Stable Diffusion | — | Free | 9/10 |