# Who makes FLUX *Two of the three picture models on [The same eighteen pictures, card by card](https://research.strata2signal.com/the-same-eighteen-pictures-card-by-card/) come from Black Forest Labs (BFL), a company in Freiburg and San Francisco whose models are called FLUX. This is its history as the public record tells it, as of 2026-09-25 (UTC): the papers, the Stability AI years, the launch, the money as reported, the licences model by model, and the turn to video and robots. A history, not a review.* *Published 2026-09-26 (UTC) · A small (human) team and a fleet of AI agents.* **the short version:** three of the researchers whose papers became Stable Diffusion were reported to have left Stability AI in March 2024; with Patrick Esser and others they launched Black Forest Labs and FLUX.1 on 2024-08-01, with a seed round BFL put at $31 million. By 2026-09-25 BFL said it was valued at $3.25 billion and had raised more than $450 million in all. It sells its strongest models through an API and releases others as open weights: FLUX.1 [schnell] and FLUX.2 [klein] 4B under Apache 2.0; FLUX.1 [dev] and klein 9B under a non-commercial licence whose words on the pictures are quoted below, without a legal reading; FLUX 3 Action under a licence with a revenue test. Every figure is as reported or as BFL states it. 3,070 words · about 14 minutes (at 220 words/min) · 1 table · data kit: no https://research.strata2signal.com/who-makes-flux/ --- *[strata→signal](https://strata2signal.com) is a small workshop that runs its own machines and writes up what it measures. This page reads rather than measures; a bracketed number names a sentence's source, and every source was read on 2026-09-25. The workshop has no business relationship with Black Forest Labs, which did not see this page before it went up; the models the bench ran came from Hugging Face under the licences below.* ## The words this page leans on {#the-words-this-page-leans-on} - **Open weights, and the API** — model files anyone can download and run, against models that run only on BFL's servers, paid for per picture. - **Distilled** — trained to copy a bigger or slower model's results with less work; BFL calls FLUX.1 [dev] "guidance-distilled" [8]. - **The autoencoder (VAE)** — the part that packs a picture into the small form the model works on, and unpacks it into pixels. - **Post-money** — a company's value counting the money just raised. - **A non-commercial licence, and outputs** — BFL's [dev] licences grant use "solely for your Non-Commercial Purposes", which the text says excludes "revenue-generating activity" and "direct interactions with or that has impact on end users". Of the pictures they say "We claim no ownership rights in and to the Outputs" and "You may use Output for any purpose (including for commercial purposes), except as expressly prohibited herein"; a restrictions clause bars using the model "or any data produced by" it "for any commercial or production purposes". This page quotes both clauses and gives no legal reading [10]. ## Before the company {#before-the-company} In December 2020 Patrick Esser and Robin Rombach posted VQGAN with Björn Ommer at Heidelberg University [1]. In 2021 Rombach "moved with his research team to LMU Munich", says Germany Trade & Invest, a German federal agency [29]; that December Rombach, Andreas Blattmann, Dominik Lorenz, Esser and Ommer posted the latent diffusion paper [2]. Stable Diffusion, released on 2022-08-22 [4], "builds upon" that work, the CompVis lab's README says, crediting "a generous compute donation from Stability AI and support from LAION" [3]. BFL's launch post counts among its team's later work "Stable Diffusion XL, Stable Video Diffusion, Rectified Flow Transformers" and "Adversarial Diffusion Distillation" [8]. The SD3 paper of 2024-03-05 lists seventeen authors under one affiliation, "Stability AI" [5]; our own reading: all ten people the federal agency later named as BFL's co-founders (Robin Rombach, Andreas Blattmann, Jonas Julius Müller, Sumith Kulal, Tim Dockhorn, Axel Sauer, Dominik Lorenz, Patrick Esser, Frederic Boesel and Harry Saini) are on that list [5, 29]. On 2024-03-20 SiliconANGLE reported that Rombach, Blattmann and Lorenz had left Stability AI, and that "Stability AI confirmed one of the three researchers' departure" [7]. ## Launch, and Grok {#launch-and-grok} The company enters the record on 2024-07-03, when the Munich commercial register, via North Data, shows an existing company take the name BFL GmbH, with Blattmann, Rombach and Axel Sauer as managing directors [6]. On 2024-08-01 BFL announced itself, "to develop and advance state-of-the-art generative deep learning models for media such as images and videos", with "our Series Seed funding round of $31 million" led by Andreessen Horowitz [8]. The lead investor wrote that BFL "is led by Robin Rombach, Patrick Esser, and Andreas Blattmann — the original co-creators of the Stable Diffusion series of models", and gave the plan: "offer paid API access and custom enterprise solutions" while committing to "release core models with permissive licenses" [9]. Three models shipped that day, and "All public FLUX.1 models" were "scaled to 12B parameters" (twelve billion), BFL said [8]: FLUX.1 [pro] through the API only; FLUX.1 [dev], "Directly distilled from FLUX.1 [pro]", as open weights; and FLUX.1 [schnell] under Apache 2.0 [8]. The post promised "a more detailed tech report in the near future" and, as "Up Next", "SOTA Text-to-Video for All" [8]; "the company did not mention where it obtained the training data", Ars Technica noted on 2024-08-02 [45]. Two days after the launch a third party, Comfy-Org, posted an all-in-one fp8 file of FLUX.1 [dev], the file the workshop's bench later ran [43]; BFL published an fp8 of its own on 2026-01-01 [44]. On 2024-08-13 xAI wrote: "In collaboration with Black Forest Labs, we are experimenting with their FLUX.1 model to expand Grok's capabilities" [11]; The Verge reported the next day that Grok's image maker, "powered by Black Forest Lab's Flux 1 AI model", let users publish pictures to X "with seemingly few guardrails in place to prevent abuse" [12]. On 2024-09-20 TechCrunch reported, from unnamed sources, "A $100 million round at a $1 billion valuation is the amount we are hearing", not final [13]. On 2024-12-09 xAI announced an image model of its own, "code-named Aurora" [15]. ## Building a business {#building-a-business} FLUX.1 Kontext [dev], an editing model, went out on 2025-06-26; with it, BFL wrote, the [dev] licence "requires content filters or manual review", and "Commercial Licenses to our open weights models can now be purchased" through a self-serve portal [17]. The Kontext report of 2025-06-17 is the only BFL paper found that describes FLUX.1's architecture [16, 37]. Bloomberg reported on 2025-09-09 that Meta had signed "a multi-year contract worth more than $100 million" (its headline says $140 million); both companies declined to comment [18]. Fortune reported on 2026-02-17 that BFL "declined to comment on exact revenue numbers" and that "Rombach says the mix is roughly 50/50 between usage‑driven products and classic enterprise licensing" [24]. ## FLUX.2, the round, and klein {#flux-2-the-round-and-klein} On 2025-11-25 BFL released FLUX.2, with FLUX.2 [dev] as open weights: a "32B open-weight model" that "couples the Mistral-3 24B parameter vision-language model with a rectified flow transformer" [19]. Its autoencoder went out under Apache 2.0, and a smaller FLUX.2 [klein] was promised as "Open-source, Apache 2.0" [19]. Six days later BFL announced "our Series B of $300M at a $3.25B post-money valuation", co-led by Salesforce Ventures and Anjney Midha's AMP [20]; the press release disclosed "a previously unannounced Series A led by Andreessen Horowitz", "bringing total funds raised to over $450M USD", with no date, size or valuation for it [21]. FLUX.2 [klein] arrived on 2026-01-15 at 4B and 9B; the name "comes from the German word for 'small'" [22]. Only the 4B models went out under Apache 2.0; the 9B carries the FLUX Non-Commercial License, a split BFL says was a release decision informed by outside safety testing [23, 25]. BFL says klein can "Generate or edit images in under 0.5s on modern hardware" and that it "outperforms Z-Image while supporting both text-to-image generation and (multi-reference) image editing in a unified model"; both are BFL's claims [22]. BFL's safety post of 2026-02-24 says outside testers checked each of its five open-weight releases before release, and that its evaluation "could not directly measure robustness to adversarial modification" such as fine-tuning [25]. ## FLUX 3, the data question, and the critics {#flux-3-the-data-question-and-the-critics} In June 2026 BFL named Martin Scorsese an adviser; BFL's page quotes him: "cinema is a young medium, only around 125 years old, so we have to be open to how it can evolve" [26]. The Art Directors Guild called his endorsement "a betrayal of the collaborative nature of cinema", Deadline reported on 2026-06-10 [27]. On 2026-06-18 BFL published a speech it says Rombach gave to G7 leaders, urging them to back open innovation [28]. On 2026-07-23 BFL put FLUX 3 into early access under the heading "Real World Models": a model that "jointly learns from images, videos, and audio within a unified architecture" [30], with "faster and open-weight versions of FLUX 3 later this year" promised [31]. On 2026-08-01 came two training-data statements. One, under a California law (AB 2013) and covering the whole FLUX family, says the family "was trained on a proprietary mix of textual, image and video data, including data privately acquired from third parties, data provided by data-labeling services and paid contractors, usage data, synthetic data, and data generated internally by Black Forest Labs", and "We started collecting data in approximately 2024 and continue to collect data today"; BFL's page lists it as superseded since 2026-08-01 and showed no newer version on 2026-09-25 [32]. The other, an EU AI Act summary for FLUX 3 only, counts "More than 1 billion images", answers "Yes" to publicly available datasets and "No" to crawlers used "by the provider or on behalf of" it, and says BFL signed the EU's code of practice [33]. BFL's release notes of 2026-08-04 call FLUX 3 "our first video model" [40]; its blog that day made video generation "generally available for the first time" [34], where the launch post had listed "SOTA Text-to-Video for All" as "Up Next" [8]. FLUX 3 Action, "A 7B World Action Model for Robot Control", was announced on 2026-09-23 as open weights under a new licence, from repositories created the day before [35, 36]. ## The licences, model by model {#the-licences-model-by-model} - **FLUX.1 [schnell]**, 2024-08-01: Apache 2.0; its model card says it "can be used for personal, scientific, and commercial purposes" [42]. - **FLUX.1 [dev]**, 2024-08-01: the FLUX.1 [dev] Non-Commercial License, its clauses on the pictures quoted in the words section [10]. From 2025-06-26 it "requires content filters or manual review", BFL wrote [17]; from 2025-11-25 one text, v2.0, covers "the models denoted as FLUX.x [dev]" [41]; on 2026-01-15 BFL said the licence "has been renamed to 'FLUX Non-Commercial License'" with "No material changes" [22], while its legal page still served v2.0 under the old name on 2026-09-25 [41]. Three texts are in circulation; which governs a download is a legal question this page does not answer [10, 38, 41]. - **FLUX.2 [dev]**, 2025-11-25: the FLUX [dev] Non-Commercial License v2.0 [41]; the FLUX.2 autoencoder, Apache 2.0 [19]. - **FLUX.2 [klein] 4B**, 2026-01-15: Apache 2.0, "available for commercial use"; BFL's card says "we encourage deployers to implement" its filters. **klein 9B**: the FLUX Non-Commercial License, "Filters or manual review must be used" [23]. - **FLUX 3 Action**, 2026-09-23: the FLUX Kommunity License v1.0 (BFL's spelling), free for "Non-Commercial Purposes"; commercial use of outputs "solely if you are a Qualifying User", one whose annualized revenue is "less than US$5,000,000", who may also run the model "in a production environment" to make them; its bans on "military purposes", "surveillance" and "biometric processing" repeat the [dev] licences [36, 10]. Our reading: the first BFL licence we found in which a user's revenue decides whether the outputs may be used commercially; tighter for large firms, looser for small ones. ## The timeline in one table {#the-timeline-in-one-table} Each date is the source's own; UTC where the source gives a time. | Date | What happened | Source | |---|---|---| | 2020-12-17 | VQGAN: Esser, Rombach, Ommer, Heidelberg | 1 | | 2021-12-20 | The latent diffusion paper | 2 | | 2022-08-22 | Stable Diffusion released | 4 | | 2024-03-05 | The SD3 paper, seventeen authors, all "Stability AI" | 5 | | 2024-03-20 | Rombach, Blattmann and Lorenz reported to have left Stability AI | 7 | | 2024-07-03 | BFL GmbH in the Munich register; Blattmann, Rombach and Sauer managing directors | 6 | | 2024-08-01 | BFL launches: FLUX.1 [pro], [dev], [schnell]; a $31 million seed, BFL says | 8, 9 | | 2024-08-13 | xAI "experimenting with" FLUX.1 in Grok | 11 | | 2024-10-02 | FLUX1.1 [pro] and "the general availability of the beta BFL API" | 14 | | 2024-12-09 | xAI announces its own image model, Aurora | 15 | | 2025-06-26 | FLUX.1 Kontext [dev] open weights; the [dev] licence requires filters or review | 17 | | 2025-09-09 | A Meta contract "worth more than $100 million", reported by Bloomberg | 18 | | 2025-11-25 | FLUX.2: [dev] open weights at 32B; the autoencoder Apache 2.0 | 19, 41 | | 2025-12-01 | Series B, $300 million at $3.25 billion post-money, BFL says; a Series A disclosed | 20, 21 | | 2026-01-15 | FLUX.2 [klein] 4B (Apache 2.0) and 9B (non-commercial) | 22, 23 | | 2026-06-03 | Martin Scorsese named an adviser; the Art Directors Guild objects, reported 2026-06-10 | 26, 27 | | 2026-07-23 | FLUX 3 in early access | 30, 31 | | 2026-08-01 | Training-data statements: California, whole family; EU, FLUX 3 only | 32, 33 | | 2026-08-04 | FLUX 3 Video on the API; "our first video model", BFL's release notes say | 34, 40 | | 2026-09-22 / 09-23 | FLUX 3 Action open weights, the FLUX Kommunity License | 35, 36 | ## The company today, and what the record does not say {#the-company-today-and-what-the-record-does-not-say} As of 2026-09-25 BFL's own pages give a company founded in 2024, valued at $3.25 billion with $450 million raised and "75M+" model downloads on Hugging Face, whose counter put FLUX.1 [dev] alone at 33,723,390 that day [38, 39]. Its pages name two companies, BFL GmbH in Freiburg and Black Forest Labs Inc. in San Francisco [33, 39]. Head counts differ by source and week: "a 90-person frontier AI lab" (BFL, 2026-06-18 [28]), "70 employees" (the federal agency, 2026-07-16 [29]), "the 100-person team" (BFL, 2026-07-23 [31]); none says what it counts. Not found in any source read, as of 2026-09-25 [37, 38]: the Series A's terms; a revenue figure BFL confirms; the FLUX.1 report promised at launch; what FLUX.1 and FLUX.2 were trained on beyond the family-wide statement; a confirmed end to the Grok work; the size of the FLUX 3 backbone, and the date and licence of "FLUX 3 Dev", the open-weight backbone BFL promised [30]; which of three texts governs a FLUX.1 [dev] download; a lawsuit naming BFL over its training data. A search that finds nothing is not proof of nothing. ## What their models did on our cards {#what-their-models-did-on-our-cards} FLUX.1 [dev], in that third-party fp8 file, and FLUX.2 [klein] 4B were two of the three models on [The same eighteen pictures, card by card](https://research.strata2signal.com/the-same-eighteen-pictures-card-by-card/), beside Z-Image-Turbo from the "Z-Image Team, Alibaba Group" [46]: the same kinds of picture at several sizes and step counts, timed on several of the workshop's graphics cards running identical software. It asks which model sits whole on a smaller card, what a lower power cap costs, and how many pictures a stretch of time buys. It times the pictures rather than judging them, and it publishes none made by FLUX.1 [dev], which it ran under the non-commercial licence above, for timing only. The numbers live on that page; this one carries none. ## Who wrote this, and thanks {#who-wrote-this-and-thanks} Written from a fact sheet of seventy-five dated rows and twelve licence rows, every source read on 2026-09-25 and checked again by a separate agent; a claim that check could not confirm was corrected, said as reported, or left out. Thanks to Black Forest Labs for the open weights and the public pages a history can be read from; to Hugging Face and arXiv; and to the reporters at SiliconANGLE, The Verge, TechCrunch, Ars Technica, Bloomberg, Deadline and Fortune. A human signed the sentences. ## Sources {#sources} All read 2026-09-25 (UTC). 1. Esser, Rombach, Ommer, "Taming Transformers for High-Resolution Image Synthesis", arXiv, 2020-12-17. https://arxiv.org/abs/2012.09841 2. Rombach, Blattmann, Lorenz, Esser, Ommer, "High-Resolution Image Synthesis with Latent Diffusion Models", arXiv, 2021-12-20. https://arxiv.org/abs/2112.10752 3. CompVis, stable-diffusion README, GitHub, undated, read 2026-09-25. https://github.com/CompVis/stable-diffusion 4. Stability AI, "Celebrating one year(ish) of Stable Diffusion … and what a year it's been!", 2023-10-03. https://stability.ai/news-updates/celebrating-one-year-of-stable-diffusion 5. Esser, Kulal, Blattmann et al., "Scaling Rectified Flow Transformers for High-Resolution Image Synthesis", arXiv, 2024-03-05. https://arxiv.org/abs/2403.03206 6. North Data, BFL GmbH, Amtsgericht München HRB 291820, register publications 2024-03 to 2026-01, read 2026-09-25. https://www.northdata.com/BFL%20GmbH,%20Freiburg%20i%C2%B7%20Breisgau/Amtsgericht%20M%C3%BCnchen%20HRB%20291820 7. SiliconANGLE, "Three of Stable Diffusion's original developers reportedly leave Stability AI", 2024-03-20. https://siliconangle.com/2024/03/20/three-stable-diffusions-original-developers-reportedly-leave-stability-ai/ 8. Black Forest Labs, "Announcing Black Forest Labs", 2024-08-01. https://bfl.ai/blog/24-08-01-bfl 9. Andreessen Horowitz, "Investing in Black Forest Labs", 2024-08-01. https://a16z.com/announcement/investing-in-black-forest-labs/ 10. Black Forest Labs, LICENSE-FLUX1-dev and its history (v1.0 to v1.1.1), GitHub, commits 2024-08-01 to 2025-06-27. https://github.com/black-forest-labs/flux/commits/main/model_licenses/LICENSE-FLUX1-dev 11. xAI, "Grok-2 Beta Release", 2024-08-13. https://x.ai/news/grok-2 12. The Verge, "xAI's new Grok-2 chatbots bring AI image generation to X", 2024-08-14. https://www.theverge.com/2024/8/14/24220127/grok-ai-chatbot-beta-image-generation-x-xai-update 13. TechCrunch, "Grok's image generator, Black Forest Labs, is raising $100M at a $1B valuation, say sources" (originally headlined "Black Forest Labs, the company that powers Grok's image generation, is raising another $100M on a $1B valuation, say sources"), 2024-09-20. https://techcrunch.com/2024/09/20/grok-image-generator-black-forest-labs-raising-100m-at-1b-valuation/ 14. Black Forest Labs, "Announcing FLUX1.1 [pro] and the BFL API", 2024-10-02. https://bfl.ai/blog/24-10-02-flux 15. xAI, "Grok Image Generation Release", 2024-12-09. https://x.ai/news/grok-image-generation-release 16. Black Forest Labs et al., "FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space", arXiv, 2025-06-17. https://arxiv.org/abs/2506.15742 17. Black Forest Labs, "FLUX.1 Kontext [dev] - Open Weights for Image Editing", 2025-06-26. https://bfl.ai/blog/flux-1-kontext-dev 18. Bloomberg, "Meta to Pay $140 Million to Use Black Forest Labs AI for Images", 2025-09-09. https://news.bloomberglaw.com/artificial-intelligence/meta-to-pay-140-million-to-use-black-forest-labs-ai-for-images 19. Black Forest Labs, "FLUX.2: Frontier Visual Intelligence", 2025-11-25. https://bfl.ai/blog/flux-2 20. Black Forest Labs, "Laying the Foundations for Visual Intelligence—Our $300M Series B", 2025-12-01. https://bfl.ai/blog/our-300m-series-b 21. Black Forest Labs press release, "UPDATE -- Black Forest Labs Announces Series B Investment to Accelerate Frontier Visual Intelligence", GlobeNewswire, 2025-12-01. https://www.globenewswire.com/news-release/2025/12/01/3197140/0/en/UPDATE-Black-Forest-Labs-Announces-Series-B-Investment-to-Accelerate-Frontier-Visual-Intelligence.html 22. Black Forest Labs, "FLUX.2 [klein]: Towards Interactive Visual Intelligence", 2026-01-15. https://bfl.ai/blog/flux2-klein-towards-interactive-visual-intelligence 23. Black Forest Labs, FLUX.2 [klein] 4B model card and LICENSE.md, Hugging Face, revision 2026-02-24. https://huggingface.co/black-forest-labs/FLUX.2-klein-4B 24. Fortune, "The AI startup that has quietly become one of Europe's most valuable AI companies", 2026-02-17. https://fortune.com/2026/02/17/ai-startup-that-has-quietly-become-one-of-europes-most-valuable-companies/ 25. Black Forest Labs, "Capable, Open, and Safe: Combating AI Misuse", 2026-02-24. https://bfl.ai/blog/capable-open-and-safe-combating-ai-misuse 26. Black Forest Labs, "Martin Scorsese × Black Forest Labs", index date 2026-06-03. https://bfl.ai/martin-scorsese-bfl-advisor 27. Deadline, "Art Directors Guild Calls Martin Scorsese's AI Endorsement 'A Betrayal'", 2026-06-10. https://deadline.com/2026/06/art-directors-guild-martin-scorsese-ai-betrayal-1236953110/ 28. Black Forest Labs, "Our co-founder and CEO urges G7 leaders to back open innovation", 2026-06-18. https://bfl.ai/blog/our-co-founder-and-ceo-urges-g7-leaders-to-back-open-innovation 29. Germany Trade & Invest, "From The Forest to The Stratosphere", 2026-07-16. https://www.gtai.de/en/invest/service/publications/markets-germany/from-the-forest-to-the-stratosphere-2011208 30. Black Forest Labs, "FLUX 3 - Real World Models: Towards Multimodal Flow Models as the Backbone of Visual Intelligence", 2026-07-23. https://bfl.ai/blog/flux-3 31. Black Forest Labs press release, "Black Forest Labs Unveils FLUX 3, A New Multimodal Frontier Model For Visual Intelligence", GlobeNewswire, 2026-07-23. https://www.globenewswire.com/news-release/2026/07/23/3332364/0/en/black-forest-labs-unveils-flux-3-a-new-multimodal-frontier-model-for-visual-intelligence.html 32. Black Forest Labs, Transparency Hub (California AB 2013 statement, model "All"), last updated 2026-08-01. https://bfl.ai/transparency 33. Black Forest Labs, "Public Summary of Training Content for General-Purpose AI models: FLUX 3, v1.0", 2026-08-01. https://cdn.sanity.io/files/2gpum2i6/production/f9dbcc8cc160256d7102a365f42be4ae33286c3c.pdf 34. Black Forest Labs, "FLUX 3 Video, Part 1: Generation", 2026-08-04. https://bfl.ai/blog/flux-3-video 35. Black Forest Labs, "FLUX 3 Action: A 7B World Action Model for Robot Control", page data 2026-09-22, index date 2026-09-23. https://bfl.ai/models/flux-3-action 36. Black Forest Labs, flux-3-action-base model card and LICENSE.md ("FLUX Kommunity License v.1.0"), Hugging Face, repository created 2026-09-22. https://huggingface.co/black-forest-labs/flux-3-action-base 37. Black Forest Labs, Research index, undated, read 2026-09-25. https://bfl.ai/research 38. Hugging Face Hub API, black-forest-labs model metadata (licence links, download counters), live data, read 2026-09-25. https://huggingface.co/api/models?author=black-forest-labs 39. Black Forest Labs, Imprint, About and Careers pages, undated, read 2026-09-25. https://bfl.ai/legal/imprint · https://bfl.ai/about · https://bfl.ai/careers 40. Black Forest Labs, API Release Notes, entry 2026-08-04, read 2026-09-25. https://docs.bfl.ml/release-notes.md 41. Black Forest Labs, "FLUX [dev] Non-Commercial License v2.0", last revised 2025-11-25, read 2026-09-25. https://bfl.ai/legal/non-commercial-license-terms 42. Black Forest Labs, FLUX.1 [schnell] model card, Hugging Face, last modified 2024-08-16, read 2026-09-25. https://huggingface.co/black-forest-labs/FLUX.1-schnell 43. Hugging Face Hub API, Comfy-Org/flux1-dev commit history (flux1-dev-fp8.safetensors, 2024-08-03), live data, read 2026-09-25. https://huggingface.co/api/models/Comfy-Org/flux1-dev/commits/main 44. Black Forest Labs, FLUX.1-dev-FP8 model card, Hugging Face, repository created 2026-01-01, read 2026-09-25. https://huggingface.co/black-forest-labs/FLUX.1-dev-FP8 45. Ars Technica, "FLUX: This new AI image generator is eerily good at creating human hands", 2024-08-02. https://arstechnica.com/information-technology/2024/08/flux-this-new-ai-image-generator-is-eerily-good-at-creating-human-hands/ 46. Z-Image Team, Alibaba Group, "Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformer", arXiv, 2025-11-27. https://arxiv.org/abs/2511.22699