Last updated: 2026-08-19
FLUX is Black Forest Labs' image generation and editing model family, with the flagship FLUX.2 released in November 2025. Its 32-billion-parameter transformer holds a character or product steady across dozens of variations at up to 4 megapixels. An official MCP server connects FLUX.2 to Claude, Cursor, Codex, and Windsurf.
About FLUX
FLUX is the image generation and editing model family built by Black Forest Labs, a Freiburg-based AI company founded in 2024 by Robin Rombach, Andreas Blattmann, Patrick Esser, and Dominik Lorenz, the team behind the original Stable Diffusion. The flagship FLUX.2 generation, released in November 2025, targets the consistency problem that plagued earlier diffusion models, letting a single reference hold its identity across an entire generated set. FLUX.2 runs on a 32 billion parameter latent flow matching transformer paired with a Mistral-3 24B vision-language model, letting it reason about spatial relationships, physical properties, and lighting before it renders pixels. The FLUX.2 [klein] models, added in January 2026, strip this down to 4B and 9B parameter variants for lighter-weight deployment. FLUX is built for developers and creative teams who need API-driven, on-brand image generation at scale: product photography with consistent lighting across hundreds of SKUs, and marketing creative that keeps a mascot or model identical across scenes. Black Forest Labs raised a $300 million Series B in December 2025 at a $3.25 billion valuation, co-led by AMP and Salesforce Ventures with participation from a16z, NVIDIA, and Northzone, bringing total funding past $450 million. FLUX models are already integrated into Adobe, Microsoft Azure AI Foundry, Freepik, and ComfyUI.
Pricing
Pay-per-image, no subscription. FLUX.2 [klein] from $0.014/image, [pro] from $0.03, [flex] from $0.05, [max] from $0.07 (megapixel-based scaling for higher resolutions). New users get 200 free Playground credits (about 12 pro generations). FLUX.2 [klein] (4B/9B) is open weight under Apache 2.0 for free self-hosting.
Key Features
- Multi-reference composition: FLUX.2 generates dozens of consistent variations from multiple reference images in a single pass, keeping characters and products identical across scenes.
- Character and style consistency (Kontext): FLUX.1 Kontext preserves a reference character, object, or art style across iterative edits while making targeted local changes without affecting the rest of the image.
- Sub-second generation with FLUX.2 [klein]: The 4B and 9B [klein] models, released January 2026, generate or edit images in under one second on an NVIDIA GB200 and run on consumer GPUs with around 13GB of VRAM.
- Official MCP server: mcp.bfl.ai exposes the full FLUX.2 toolkit to Claude, Cursor, Codex, and Windsurf via OAuth sign-in, with no API keys required.
- Accurate in-image text rendering: FLUX produces clear, legible typography inside generated images, making it suitable for signs, posters, and book covers where competitors often render garbled text.
- Open weight self-hosting: FLUX.2 [klein] ships under the Apache 2.0 license for commercial self-hosting, while FLUX.2 [dev] is available under a non-commercial license for research.
Pros
- Local self-hosting removes the per-image API cost and any dependency on Black Forest Labs' uptime, a rare option among frontier-quality image models.
- Usage-based pricing means idle months cost nothing, unlike flat monthly subscriptions charged by seat-based competitors regardless of output.
- Being open weight rather than closed puts FLUX.2 [klein] in a small group of frontier-quality image models teams can legally fine-tune or self-host for commercial products.
- Backing from top-tier investors including a16z and NVIDIA gives FLUX more runway than most open-weight labs, reducing the risk the models get orphaned or paywalled.
Cons
- Full FLUX.2 models need around 90GB of VRAM for local deployment, putting them out of reach for most consumer hardware.
- Complex hand gestures and intricate cultural or anatomical details remain weak points compared to top-tier competitors.
- Abstract concepts like 'loneliness' or 'digital chaos' produce literal results, lacking the artistic interpretation Midjourney offers.
- The BFL Playground is a bare-bones web interface with no community gallery or remix features, unlike Midjourney's Discord ecosystem.
Frequently Asked Questions
What are FLUX's pricing plans in 2026?
FLUX bills per image with no monthly subscription. FLUX.2 [klein] starts at $0.014, [pro] at $0.03, [flex] at $0.05, and [max] at $0.07, with megapixel-based pricing that scales up for higher resolutions. The earlier FLUX.1 and Kontext lineup runs $0.04 to $0.08 per image depending on the tier. New users also get free Playground credits, about 12 pro-model generations, and FLUX.2 [klein] can be self-hosted for free under its open license.
Does FLUX have a free plan?
FLUX isn't unlimited for hosted use, but new BFL Playground accounts get a batch of free credits, about 12 pro-model generations, with no third-party sign-up required. FLUX.2 [klein] (4B/9B) is also free to self-host under its open-weight license, though the larger [pro], [flex], and [max] models are hosted-only and billed per image. Local deployment of the biggest FLUX.2 checkpoints needs roughly 90GB of VRAM, which limits truly free use to owners of serious hardware.
What should you use instead of FLUX?
Midjourney still produces the most polished community-driven output if an API is not part of the requirement. Ideogram handles typography inside the image better and hands out more free generations to test with. Adobe Firefly is the enterprise answer to copyright exposure, since its training data is licensed rather than scraped. FLUX earns the slot when API-first pricing, multi-reference consistency, and an open-weight sibling matter in the same decision.
Is FLUX better than Midjourney?
FLUX.2 [pro] generally beats Midjourney on prompt adherence for complex, multi-element scenes and matches it on photorealism and text rendering, though Midjourney still produces a more consistently striking, artistic aesthetic for abstract or stylized concepts. FLUX bills per image with no monthly subscription, while Midjourney requires a $10-$60 monthly plan regardless of usage, and FLUX.2 [klein] adds an open-weight, self-hostable option Midjourney does not offer. Pick FLUX for API integration and open-weight flexibility; pick Midjourney for out-of-the-box artistic quality and its Discord creative community.
How do you set up FLUX?
A free account at bfl.ai or playground.bfl.ai gets you generating in the browser immediately, no card involved. For production, take an API key out of the BFL dashboard and call api.bfl.ai directly, or point an AI coding agent at the hosted MCP endpoint and skip key handling altogether. Either path sends a text prompt plus any reference images to a FLUX.2 model endpoint for the first image or edit.
Top Alternatives
- Midjourney: FLUX bills per image and asks for nothing monthly. Midjourney charges a subscription whether you generate or not, and pays it back in out-of-the-box aesthetic.
- Ideogram: In-image text and logo work is Ideogram's home ground, and its free tier is generous. FLUX answers a different question: consistency across variations, driven from an API.
- Adobe Firefly: Adobe Firefly carries commercial indemnification on licensed training data, which is what legal departments ask about. FLUX offers open weights, self-hosting, and a lower per-image bill instead.
HokAI guides covering FLUX
- Adobe Firefly vs Topaz Photo AI: Which Should You Use in 2026?: Adobe agreed to buy Topaz Labs in June 2026. Here's what Firefly and Topaz Photo AI cost now, who needs which one, and why waiting for the merger is a bad bet.
- Best AI Image Generators in 2026: Pricing, Licensing, and the Real Cost Per Image: Midjourney's cheapest plan legally stops covering you past $1M revenue. A verified price-per-image comparison across 8 AI image generators for 2026 buyers.