What Is Flux? Definition + Examples
Flux is Black Forest Labs' family of text-to-image models, including Flux 1.1 Ultra, Kontext, and Klein. Plus how it works, examples, and where to use it in AI workflows.
Flux is a family of text-to-image models built by Black Forest Labs, with three active tiers: Klein (fast and cheap), Flux 1.1 Ultra (photorealistic flagship), and Kontext (reference-guided editing).
Each tier is tuned for a different job. Klein is the volume workhorse at $0.01 per image. Ultra is the quality ceiling at $0.04, capable of photorealism that consistently beats older Midjourney generations in controlled comparisons. Kontext sits at $0.05 and does something the other two don't: it lets you anchor a generation to a reference image and edit specific elements while keeping the rest of the scene intact. All three run on the 8frame canvas, where you can compare them side by side on the same prompt before committing to a run.
How Flux works
Flux is a flow-matching model, not a standard diffusion model. The distinction matters in practice: instead of iteratively denoising from pure noise toward a target image, Flux learns a direct mapping between a noise distribution and the output image distribution. That means fewer inference steps for the same quality, which translates to faster generation and more predictable behavior when you change one variable in a prompt.
Black Forest Labs trained the Flux family with a transformer-based architecture (not a U-Net), which improves how the model handles text rendering, fine-grained detail, and compositional prompts with multiple subjects. You'll notice this most clearly in prompts involving readable text in the image, dense fabric texture, or scenes where two distinct objects need to coexist without blending together.
Kontext adds a second input path: alongside your text prompt, you feed it a reference image. The model uses that reference to condition the output, letting you swap backgrounds, change colors, or isolate and modify a single element without touching the rest. It's the closest thing to non-destructive editing inside a generation model.
When you use each tier
Klein ($0.01) is for high-volume iteration. If you're prototyping a concept, testing a dozen prompt variations, or producing background assets where close-enough is fine, Klein is the right call. The quality is noticeably lower than Ultra on complex scenes, but for flat-lay product shots, simple compositions, and batch generation runs, the 4x cost difference is rarely worth it.
Flux 1.1 Ultra ($0.04) is for final-quality photorealistic output. Editorial portraits, product hero shots, architectural renders, and anything going into a high-resolution placement should go through Ultra. It handles skin texture, specular highlights, and accurate hands better than most models at this price point.
Kontext ($0.05) is for reference-based editing. Use it when you already have an image and need to change part of it without rebuilding from scratch: swap a model's outfit while keeping the face consistent, replace a background without re-generating the subject, or adjust product colorways across a set of shots without losing the lighting.
Examples
Klein, concept iteration. Prompt: "brutalist apartment building, overcast sky, wide angle, muted tones". At $0.01 per image you can run 15 variations in under a minute and pick the two compositions worth refining. The output isn't print-ready but it's fully sufficient for layout mockups or client mood boards.
Flux 1.1 Ultra, product hero shot. Prompt: "glass serum bottle, wet marble surface, macro lens, studio strobe lighting, photorealistic, 8K". Ultra returns a commercially usable still with accurate glass refraction and sharp label detail. At $0.04 it's cheaper than most stock options, and the image is exclusive to you.
Kontext, reference edit. Feed it a photo of a white sneaker on a gray background and prompt: "change colorway to navy blue and coral, keep the same lighting and composition". Kontext updates the shoe colors while leaving the shadow angle, background tone, and sole texture exactly where they were.
Related concepts
A full head-to-head across Klein, Ultra, and Nano Banana with side-by-side outputs and a cost-per-quality breakdown is in Nano Banana vs Seedream vs Flux.
For prompt patterns that consistently get sharp photorealistic results out of Ultra specifically, see Flux 1.1 Ultra prompts for photorealism.
Ready to run it? Open the 8frame canvas, pick your Flux tier, and compare it against the rest of the lineup on the same prompt. Klein to Ultra to Kontext, all in one view.