Flux

Picture this: you're building a product that needs on-brand marketing images with legible text baked in, and you need them through an API, not a Discord bot. You prototype prompts in the browser, wire the winning setup into your backend at a per-image rate, and later move the model onto your own servers once volume justifies it. That's the workflow Flux from Black Forest Labs is built around. The company pitches it as "Start in our playground. Integrate via API. Download our models, customize, and deploy anywhere," and it's one of the few image generators where all three paths come from the same vendor.

The headline release is FLUX 3, described as "one multimodal model" covering image, video, audio and action-prediction for robotics. For images, Black Forest Labs claims "highly-accurate text rendering" and the ability to handle complex prompts across a wide range of styles. Video clips run up to 20 seconds in a single generation, can start from text, an image or keyframes, and can include generated speech, effects and ambience. There's also FLUX 3 Action, an open weights 7B model aimed at robotics that the company says took first place on the RoboLab benchmark. Most readers here will care about the image side, but it's useful to know the lab is shipping across modalities.

Pricing on the API is refreshingly simple: "No subscriptions, no seat fees. Only pay for what you generate." The calculator's default example comes out to $0.048 per image, with resolutions from 768 × 768 up to 4K and several FLUX.2 variants to pick from. If you'd rather self-host, there are open weights licenses tiered by monthly image volume, domains and seats, all with fine-tuning and LoRA rights. The catch is that the license prices aren't shown on the pricing page we reviewed, so you'll need to click through or talk to sales. If you want production-grade image generation with a clear exit to self-hosting, Flux is a strong pick.