What’s new
- Fast and open: 4-step distilled inference gets you sub-second generation, and it runs on a consumer GPU (the 9B build needs roughly 13GB VRAM and up).
- One model for generation and editing: no need to switch models for editing — the same architecture handles text-to-image, image editing and multi-reference input.
- Small model, big-model results: Black Forest Labs says the 9B variant matches or exceeds models five times its size on text-to-image, single-reference editing and multi-reference generation, in under half a second.
- Positioned as the "spiritual successor" to FLUX.1 [schnell]: built for speed, editing and local deployment rather than chasing the highest possible quality ceiling.
What it’s good for
- Low-latency, interactive generation and editing — think real-time previews inside a design tool.
- Local deployment for developers who don't want to depend on a cloud API.
- Fast, capable image editing (background swaps, local changes) where top-tier quality isn't the priority.
Prompting tips
- Don't expect prompt expansion: Black Forest Labs says klein has no prompt upsampling, so writing one or two vague words will get you a weaker result — spell it out yourself.
- Same structure as FLUX.2 Pro: subject, then action, then style, then context, with the most important details first.
- No negative prompts here either: describe what you want; there's no supported way to say what to avoid.
- Be explicit about the edit target: when you supply a reference image, state exactly what should change and how — not just "make it look better".
Replace this product photo's background with a plain light-gray studio backdrop, keeping the product's
position, angle and color unchanged. Add a soft contact shadow on the ground, cool-toned overall lighting,
and keep the output at the same aspect ratio as the source image.
Known limits
- The 9B build Nomi uses ships under the FLUX Non-Commercial License (only the 4B build is fully open under Apache 2.0), so it can't be used commercially as-is.
- No negative prompts and no prompt auto-expansion, so results are more sensitive to how detailed your prompt is.
- It's built for speed first — the quality ceiling is lower than the Pro or Max tiers.
Using it in Nomi
On the generation canvas, add a image card, pick “FLUX.2 Klein” in the model picker and choose a mode (Text to image). Add your references, write the prompt and generate. In the storyboard you can pick it per shot. The first time, connect any one of the providers above under Settings → Models.
