Black Forest Labs closed $300M Series B at $3.25B post-money on December 1, 2025 — Salesforce Ventures and AMP co-leading, with a16z, NVIDIA, Bain Capital, and Figma Ventures alongside. That round confirmed FLUX as one of three serious open-and-closed labs left standing after the multimodal-frontier pile-on. The 25 companies on this list ($1.65B disclosed) are now in clear consolidation.
Who's still credible?
Black Forest Labs ships FLUX as the leading open-weight image model on Hugging Face and the most-deployed enterprise model on Fal.ai, Replicate, and Together. Midjourney shipped v7 (still bootstrapped, still #1 on aesthetic benchmarks) and now extends into video. Ideogram (Toronto, $96.5M Series A) shipped 3.0 with the strongest in-image text rendering — the wedge that originally won them a buyer like Microsoft Copilot. Stability AI's $101M Series A keeps the open-weight Stable Diffusion lineage alive but the lab is no longer setting the frontier.
The design-suite tier
This is where Recraft (UK, $30M Series B), Leonardo.Ai (Australia, $31M Series A), Krea AI ($11M Series A), and Magnific (Spain, acquired by Freepik in 2024) compete on workflow rather than raw model quality. Recraft has been winning enterprise design contracts on brand-consistency and vector output. Krea ships real-time generation; Magnific specializes in upscaling and stylization. Adobe Firefly (corporate, inside Creative Cloud) is the existential threat to all of them.
Why the long pause, and why it broke
Twelve months of zero disclosed rounds preceded the FLUX raise. Investors had concluded standalone image generation was a feature, not a category — GPT-image, Gemini-image, and Claude-image inside chat absorbed casual prompting. The Dec 2025 BFL round and an active Q1 2026 (Midjourney v7, Ideogram 3.0, Krea Series A talks reported) suggest the consolidation phase is ending. Surviving labs now defend on text rendering, brand control, real-time editing, or open weights for regulated buyers — not on absolute prompt fidelity.
Regulatory pressure on training data is tightening in parallel. The EU AI Act's transparency obligations, the Getty v. Stability suit still grinding through the UK High Court, and active US class-action discovery mean every new model trained in 2026 is built with provenance and opt-out tooling baked in.