Stability AI

Stability AI is a generative artificial intelligence company best known for helping fund and release the Stable Diffusion family of image models. Its principal UK trading entity, Stability AI Ltd, was incorporated on November 4, 2019 and is wholly owned by Stability AI, Inc., a Delaware corporation. The group develops and distributes models for image, video, audio, language, code, and 3D generation through hosted services, self-hosted licenses, custom training, and distribution partners.^[1][2]

Stable Diffusion was a collaborative project, not the work of Stability AI alone. Stability AI's 2022 launch announcement credited Patrick Esser of Runway and Robin Rombach of the Machine Vision and Learning group at Ludwig Maximilian University of Munich with leading the work, which built on the CompVis latent diffusion research and received support from Stability AI, LAION, EleutherAI, and other contributors.^[3][4][5]

The company calls many of its releases open source, but "open-weight" is a more precise description for much of its current portfolio. Stability AI publishes weights for selected models, while its Community License imposes use conditions and requires a paid Enterprise License for commercial users whose annual revenue exceeds $1 million. Those conditions differ from unrestricted open-source AI licenses.^[6][7]

Prem Akkaraju has been chief executive since a June 2024 recapitalization, with Sean Parker as executive chairman. Under them the company has shifted toward products for professional creatives and now describes itself as building AI tools for music, gaming, and entertainment. In August 2026 it announced a $76 million Series B whose new investors included Electronic Arts and the three major record companies, Sony Music Group, Universal Music Group, and Warner Music Group. The company did not disclose a valuation.^[11][38][39]

Corporate history and financing

Legal entities and early history

Stability AI Ltd was incorporated in England and Wales in 2019. Its filed accounts identify it as the main trading entity of the Stability AI group and say that its first revenue-generating activities began in August 2022 with the Stable Diffusion series.^[1][2]

The history of the US parent is partly contested in litigation. A June 2026 Delaware Court of Chancery opinion recounts that Emad Mostaque, Cyrus Hodes, and a third co-founder incorporated Stability AI, Inc. in October 2020, with Mostaque receiving 70 percent of the equity and the other two receiving 15 percent each. The opinion cautions that this account was drawn from submissions on a discovery dispute and was not a formal finding of fact.^[8]

Hodes sold his one million shares to Mostaque in two transactions in 2021 and 2022 for a total of $100. He later alleged fraud, negligent misrepresentation, and breach of fiduciary duty, among other claims. The court dismissed unjust-enrichment and quantum-meruit counts but allowed the three other counts to proceed into discovery. Its June 2026 opinion decided only whether particular communications were protected by spousal privilege, not whether Hodes's underlying allegations were true.^[8]

Funding and the 2024 recapitalization

In October 2022, Stability AI announced a $101 million funding round led by Coatue, Lightspeed Venture Partners, and O'Shaughnessy Ventures.^[9] TechCrunch, citing a Bloomberg source, reported that the round valued the company at $1 billion post-money.^[42] In March 2024, Mostaque resigned as chief executive and from the board. Chief operating officer Shan Shan Wong and chief technology officer Christian Laforte became interim co-chief executives.^[10]

The company announced another investment round in June 2024 and appointed Prem Akkaraju as chief executive and Sean Parker as executive chairman. The announcement named Greycroft, Coatue, Sound Ventures, Lightspeed, O'Shaughnessy Ventures, Parker, Eric Schmidt, Akkaraju, and Robert Nelsen among the investors but did not disclose the amount.^[11] Stability AI Ltd's later accounts say the parent group raised more than $80 million and renegotiated onerous contractual commitments. The UK group recorded a separate $50 million recapitalization from its parent during 2024.^[2]

TechCrunch, citing The Wall Street Journal, reported at the time that the new investors had committed $80 million, that suppliers had agreed to forgive about $100 million in debt, and that Stability AI had been released from about $300 million in future obligations, largely to cloud infrastructure providers. The Information and Reuters had reported beforehand that the round could value the company below its earlier valuation.^[43]

2026 Series B

On August 25, 2026, Stability AI announced a $76 million Series B, describing itself as "the leader in purpose-built AI products for professional creatives across music, gaming, and entertainment." The company said the round brought the total raised under Akkaraju since June 2024 to $232 million, a figure that includes two equity rounds and convertible notes. Some coverage reported the $232 million as the company's overall fundraising total, but the announcement ties it to the period under the new leadership. The announcement did not name a lead investor or disclose a valuation.^[38][39][40]

New investors were Electronic Arts, Sony Music Group, Universal Music Group, Warner Music Group, AMD Ventures, and Pacific Alliance Ventures. Coatue, Greycroft, Kadmos Capital, Parker, and Eric Schmidt also invested, in what the company called their second consecutive round under the new leadership. The announcement listed its other backers as Lightspeed Venture Partners, Mantis Capital, Sound Ventures, and WPP, along with individual investors including James Cameron, Kevin Mayer, Mark Burnett, Patrick Whitesell, Akkaraju, Reuben Brothers, Robert Nelsen, and Vivi Nevo. Electronic Arts, Universal Music Group, and Warner Music Group were already strategic partners of the company. Stability AI said the capital would fund its product suite for creative production, its applied research, and an expanded professional services arm.^[38][40]

In October 2026, TechCrunch, summarizing an interview Parker gave to The Information, reported that Sony, Warner, and Universal had also licensed their catalogs for training as part of the deal. Parker described the aim as making Stability AI the go-to AI toolmaker for music professionals and said a coming update would let users steer generation by humming a melody or beatboxing a drum pattern.^[44]

Tech.eu reported that the company, which was founded in London, had closed its London office and shifted its focus to the United States, with an office in Los Angeles.^[41] The Series B announcement carried a Los Angeles dateline, while Companies House continued to list Stability AI Ltd's registered office in London.^[38][1]

Filed financial results

Stability AI Ltd's accounts for the year ended December 31, 2024 distinguish reported revenue from recurring operations. The group reported $55.0 million in total revenue, but classified $43.6 million as non-underlying. That amount included a one-time $31.0 million dataset-license transaction, $12.2 million from a discontinued managed-services line, and about $0.4 million of intragroup ClipDrop sales. Underlying revenue was $11.4 million.^[2]

The accounts compare the full 2024 year with a nine-month 2023 period, so direct year-over-year comparisons require care. They show that the group remained loss-making, with a $47.5 million loss in 2024, down from a $119.5 million loss in the shorter prior period. At year end, it had $23.9 million in cash and cash equivalents and $52.2 million in net liabilities.^[2]

The directors prepared the accounts on a going-concern basis, stating that they expected adequate resources for the foreseeable future. They also identified the parent company as the primary source of funds and said it had initiated another funding round when the accounts were signed. The strategic report identified changes in customer demand and continued parent funding as key risks.^[2]

Stable Diffusion

Research provenance

Stable Diffusion grew out of research on latent diffusion models by Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bjorn Ommer. Their method first compresses images into a lower-dimensional latent representation, then trains a diffusion process in that space. This reduces the computing cost of image synthesis compared with running the entire denoising process at full pixel resolution.^[4]

The original Stable Diffusion repository describes the project as a collaboration among CompVis, Stability AI, and Runway. It says Stability AI donated computing resources and LAION provided support, and that the model was trained on subsets of LAION-5B.^[5] Stability AI's own August 2022 announcement similarly credited Esser and Rombach as project leads and named CompVis, Runway, LAION, EleutherAI, and the company's generative AI team as contributors.^[3]

Public release and licensing

Stability AI announced the public release on August 22, 2022. Model weights and code were made downloadable under the CreativeML OpenRAIL-M license, which permitted commercial and non-commercial use but included use-based restrictions. The original repository also warned that the weights had known limitations and biases and recommended additional safety mechanisms for products and services.^[5][12]

The reference implementation included a safety checker in its default generation pipeline.^[5] Because the weights and code could be run and modified outside Stability AI's infrastructure, that component was not equivalent to a mandatory platform-wide control. The release therefore combined broad access and local adaptability with fewer centralized enforcement mechanisms than a hosted-only service.

Later image-model generations

Stable Diffusion XL, or SDXL, expanded the architecture with a UNet about three times larger than earlier Stable Diffusion versions, a second text encoder, additional conditioning, and an optional refinement stage. The accompanying paper described a two-stage base-and-refiner pipeline and evaluated the model primarily through human preference studies.^[13]

The Stable Diffusion 3 research introduced a multimodal diffusion transformer architecture and a rectified-flow formulation. The paper studied scaling up to an 8-billion-parameter model and focused on prompt following, typography, and text-to-image quality.^[14]

Stability AI released Stable Diffusion 3.5 Large and Large Turbo in October 2024, followed by Medium later that month. The company described Large as an 8.1-billion-parameter base model, Large Turbo as a four-step distilled version, and Medium as a 2.5-billion-parameter model intended for consumer hardware.^[15] These models use the Stability AI Community License rather than the original CreativeML OpenRAIL-M terms.^[15][7]

Model portfolio

Stability AI's May 2026 Core Models list groups its licensed portfolio into media, language and code, 3D, and audio. The list is a current licensing record rather than a permanent product catalog, and the company expressly reserves the right to modify it.^[6]

Video, audio, language, and 3D

The Stable Video Diffusion research adapted latent video diffusion models to text-to-video and image-to-video generation. Its released image-to-video variants generated 14 or 25 frames, and the work also explored multi-view generation for 3D applications.^[16]

Stable Audio Open, released in 2024, is a text-to-audio model documented as having been trained only on Creative Commons licensed audio. Its paper describes generation of stereo audio at 44.1 kHz for clips up to 47 seconds and presents it mainly as a model for sound effects and short production elements rather than complete songs.^[17]

The May 2026 Stable Audio 3 technical report describes small, medium, and large latent diffusion models for variable-length music and sound-effect generation, editing, inpainting, and continuation. It says the models were trained on licensed and Creative Commons data. The small and medium weights were released, while the product announcement made the Large model available through the API and enterprise self-hosting. Medium and Large support outputs of up to 6 minutes 20 seconds.^[18][19]

Stable LM began in April 2023 with 3-billion and 7-billion parameter base models. The base releases used a CC BY-SA 4.0 license, while the first instruction-tuned research versions used a non-commercial license, an important distinction within the family.^[20] The company's later Core Models list includes Stable LM 2 and Stable Code variants under its current licensing program.^[6]

Stable Fast 3D reconstructs a textured, UV-unwrapped mesh from a single image. Its paper reports approximately 0.5 seconds of generation time on an Nvidia H100 GPU and describes direct mesh prediction, material estimation, and fast UV unwrapping.^[21]

Research published in 2026

The company's research group continued to publish on video, 3D, and audio generation. In October 2026 its research page added the following three papers.^[58]

Access and licensing

The filed accounts describe four principal routes to market: models hosted by Stability AI, self-hosted annual licenses that may include custom training and support, distribution partners operating under revenue-share or listing arrangements, and applications sold by subscription. Hosted access can be sold through web interfaces, APIs, credits, subscriptions, or enterprise seats.^[2]

For current Core Models, the Community License permits research, non-commercial use, and commercial use by people or organizations with less than $1 million in annual revenue. Commercial users above that threshold must obtain an Enterprise License. The license also incorporates an Acceptable Use Policy and restricts using Core Models or their outputs to create a competing foundational model, although it permits fine-tunes and LoRAs.^[7][22]

These terms allow inspection and local deployment of selected weights but are not the same as an unconditional open-source grant. Individual releases can also carry different licenses, so the terms for a specific model and version must be checked rather than inferred from the Stability AI brand.^[6][7]

Business and partnerships

The group's reported business model combines hosted inference, self-hosted model licenses, custom model training, applications, and partner distribution.^[2] Amazon Web Services added Stable Image Ultra, Stable Diffusion 3 Large, and Stable Image Core to Amazon Bedrock in September 2024, giving customers a managed route to those models through AWS.^[23]

Stability AI and WPP announced an investment and partnership in March 2025. Their announcement described joint research and development and integration of Stability AI workflows into WPP Open. Stability AI Ltd's accounts independently list WPP among the group's strategic partnerships.^[2][24]

Arm reported in March 2025 that the companies had optimized a distilled Stable Audio Open model to run locally on Arm-based smartphones. Arm said the collaboration used KleidiAI, XNNPack, and ExecuTorch and reduced generation time from minutes to seconds in its demonstration.^[25]

Electronic Arts announced a partnership with Stability AI in October 2025 to co-develop generative tools and workflows for game artists and developers.^[26] Universal Music Group announced a separate alliance that month to develop professional music-creation tools, with an emphasis on artist involvement and licensed music data.^[27] Warner Music Group followed in November 2025 with a collaboration to build professional tools that let artists, songwriters, and producers compose and produce with what the companies called ethically trained models.^[47] These announcements describe intended collaborations; they are not evidence by themselves of revenue or broad adoption.

In April 2026, Stability AI launched Brand Studio, a managed creative-production platform for professional and enterprise teams. The product combines custom brand models, campaign references, editing tools, and model routing that may use both Stability AI and third-party models.^[28] The launch marked a further shift toward managed applications and enterprise workflows alongside downloadable model weights.

The company also expanded its audio products. In September 2025 the company released Stable Audio 2.5, which it described as the first audio generation model designed for enterprise sound production, citing inference in under two seconds on a GPU and support for audio inpainting.^[49] In August 2026 it released a beta Stable Audio plugin that runs as an instrument inside digital audio workstations such as Logic Pro and Ableton Live (as a macOS AU or VST3 plugin), together with an expanded StableAudio.com editor for iterative direction, audio-to-audio variation, and multitrack mixing. The company said users own their outputs and may distribute them freely.^[48]

Leadership and governance

Mostaque resigned as chief executive and from the board in March 2024. The company said he left to pursue decentralized AI, while Wong and Laforte served as interim co-chief executives.^[10] Prem Akkaraju became chief executive in June 2024, and Sean Parker joined as executive chairman.^[11] Akkaraju was previously chief executive of the visual-effects company Weta Digital.^[43]

Filmmaker James Cameron joined the board in September 2024. Variety reported at the time that former Fox chief technology officer Hanno Basse had been named Stability AI's CTO the previous month.^[45] In March 2025 the company hired visual-effects veteran Robert Legato, whose credits include Avatar and Titanic, as chief pipeline architect, a role meant to align its applied research with the needs of visual-effects production.^[46]

In October 2026, Stability AI's board page listed Sean Parker, Prem Akkaraju, James Cameron, Dana Settle, and Thomas Laffont.^[29] The August 2026 Series B announcement said that Laffont, a co-founder of Coatue, had joined the board, although an archived copy of the board page shows him listed by December 2025.^[38][59] The company's current leadership structure therefore differs substantially from the founder-led organization that released the first Stable Diffusion weights.

Data, safety, and litigation

Getty Images litigation

Getty Images brought proceedings in both the United Kingdom and the United States over Stable Diffusion. The outcomes must be separated because the cases involve different law, allegations, and procedural stages.

In the UK case, Getty abandoned its training-and-development claim after acknowledging that there was no evidence the relevant training occurred in the United Kingdom. It also abandoned its output-copyright claim after Stability AI blocked the prompts at issue, and its database-right claim could no longer proceed.^[30]

The November 2025 High Court judgment rejected Getty's secondary copyright theory. The court held that the final model did not store or reproduce the copyrighted works and therefore was not an "infringing copy" for the relevant secondary-infringement provisions. Getty did succeed on limited trademark claims involving a few historic outputs from Stable Diffusion 1.x and 2.x that resembled Getty or iStock watermarks. The judge described those findings as "historic and extremely limited in scope" and found no corresponding UK-user evidence for SDXL or version 1.6.^[30]

That judgment did not decide whether training Stable Diffusion in another country infringed copyright there. The court explicitly confined itself to the diminished UK case before it, so it should not be read as a general ruling that Stable Diffusion training or outputs are lawful everywhere.^[30]

At a consequentials hearing on December 16, 2025, the judge granted Getty permission to appeal the dismissal of its secondary-infringement claim. She called the meaning of "infringing copy" in the context of an AI model a novel and important point of law that no court had previously considered. She refused Stability AI permission to appeal the trademark findings, holding that its grounds had no real prospect of success.^[50] The Getty Images v. Stability AI article covers both proceedings in more detail.

Getty's US litigation did not end when it left Delaware. Getty voluntarily dismissed the Delaware action without prejudice on August 14, 2025 and filed a new complaint in the Northern District of California the same day.^[31] In April 2026, the California court considered a motion directed at counts II through VII. It dismissed the false copyright-management-information count without prejudice and allowed Getty to amend it, but declined to dismiss the challenged federal trademark, false-designation, dilution, and California unfair-competition claims. The copyright-infringement count was not part of that motion.^[31] Stability AI Ltd, Stability AI, Inc., and Stability AI US Services Corporation answered the complaint on May 21, 2026, listing fair use as their first affirmative defense alongside license, statute-of-limitations, and other defenses.^[51]

Artist class action and founder dispute

The artist lawsuit led by Sarah Andersen also remained active. A June 17, 2026 discovery order required the plaintiffs to provide some additional financial information and discussed the unsettled application of fair use to generative AI. The order resolved a discovery dispute, not the merits of the copyright claims.^[32]

The Hodes founder litigation likewise remained in discovery as of June 2026. As noted above, the surviving fraud, negligent-misrepresentation, and fiduciary-duty claims had not been adjudicated on the merits.^[8]

Lawsuits filed in 2026

On July 2, 2026, the automotive photography company EVOX Productions sued Stability AI, Inc., Stability AI Ltd, Stability AI US Services Corporation, Runway, Hugging Face, and DeviantArt in the US District Court for the Central District of California (case 2:26-cv-07201). The complaint alleges that the LAION-2B-en subset of LAION-5B links to at least 103,957 of EVOX's copyrighted vehicle photographs and that Stable Diffusion was therefore trained on them without a license. It brings claims for direct copyright infringement over training, outputs, and distribution, contributory infringement, and removal or alteration of copyright management information under the Digital Millennium Copyright Act.^[52] Stability AI, Runway, and DeviantArt filed separate motions to dismiss on October 12, 2026.^[53]

Also in July 2026, Stability AI was added as a defendant in a class action that three Tennessee teenagers had filed earlier that year against Elon Musk's AI company xAI, which by then used the SpaceXAI brand. NPR reported that the amended complaint alleges perpetrators used the companies' AI models to turn photographs taken when the plaintiffs were minors into child sexual abuse material, and that the application used against the Tennessee plaintiffs "relied on Stability AI's image-producing tools." The suit also alleges that Stability AI could have restricted its models' ability to generate such material in more recent releases but chose not to. Stability AI told NPR that "Any suggestion that safety is not a top priority for us is categorically wrong" and said it had put safeguards in place. The allegations had not been tested in court.^[54]

LAION-5B and child-safety concerns

The first Stable Diffusion release was trained on subsets of LAION-5B, a dataset maintained by the independent nonprofit LAION rather than by Stability AI.^[3][5] LAION-5B distributed metadata, including image URLs, text descriptions, and embeddings, rather than a bundle of the referenced image files.^[33]

A 2023 Stanford Internet Observatory study identified 3,226 LAION-5B entries as suspected child sexual abuse material, or CSAM. Of those, 1,008 were manually validated by the Canadian Centre for Child Protection as CSAM or likely CSAM. The report stressed that many URLs were dead or could not be conclusively classified and that the dataset entries were links and metadata, not stored copies of the images.^[33]

LAION took LAION-5B and its derivatives offline in December 2023. It released Re-LAION-5B in August 2024 after working with the Internet Watch Foundation, the Canadian Centre for Child Protection, and the Stanford researchers. LAION said it removed 2,236 links that matched partner-provided image or URL hashes, including the 1,008 validated links from the Stanford report. Its broader filtering for the research release removed about 1.121 percent of the original samples, or approximately 65 million entries from the 5.8-billion-entry source.^[34]

The revised metadata did not retroactively change model weights that had already been trained and released. It did, however, provide removal lists that dataset users could apply to derivatives of LAION-5B.^[34]

Company-stated safeguards and their limits

Stability AI says that, for model versions it develops exclusively, it filters training data to remove unsafe images. It also says its hosted applications and APIs filter prompts and outputs, use red teaming and safety fine-tuning, compare uploads against known-CSAM hashes, and add C2PA provenance metadata to API-generated media.^[35][36]

The company-wide Acceptable Use Policy applies contractually to hosted access, third-party platforms, and self-hosted code or weights.^[22] Technical enforcement differs by deployment route, however. The prompt and output filters described by Stability AI operate in its hosted applications and APIs; independently run weights do not pass through that hosted moderation layer. This distinction is one of the central tradeoffs of open-weight distribution.^[35][22]

In February 2026, Stability AI announced that it had joined the Tech Coalition after participating in the coalition's Pathways program since 2025.^[37] Membership is evidence of industry participation, while the effectiveness of the company's safeguards still depends on implementation, coverage, and the deployment environment.

Developments through October 2026

By late October 2026, Stability AI was pursuing two connected strategies. It continued to publish selected weights, including the small and medium Stable Audio 3 models, while selling enterprise licenses, hosted APIs, custom training, and managed applications.^[2][7][19]

Its 2026 releases and announcements included Tech Coalition membership in February, Brand Studio in April, Stable Audio 3 in May, the Stable Audio plugin beta in August, and the $76 million Series B later that month.^{[37][19][28][48][38]} Together with the 2025 partnerships in advertising, games, mobile hardware, and music, these developments indicate a stronger focus on professional creative workflows, music in particular, and licensing.^[44] They do not erase the company's continuing dependence on parent financing or the unresolved US litigation described in its filed accounts and court records.^{[2][31][32][52]}

See also

• Stable Diffusion
• Getty Images v. Stability AI
• Emad Mostaque
• Stable Audio
• Black Forest Labs
• List of neolabs
• Neolab

References

1. ^1 ^2 ^3Companies House: Stability AI Ltd overview
2. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12Stability AI Ltd: accounts for the year ended December 31, 2024
3. ^1 ^2 ^3Stability AI: Stable Diffusion announcement
4. ^1 ^2High-Resolution Image Synthesis with Latent Diffusion Models
5. ^1 ^2 ^3 ^4 ^5CompVis Stable Diffusion repository
6. ^1 ^2 ^3 ^4Stability AI Core Models list
7. ^1 ^2 ^3 ^4 ^5Stability AI licensing
8. ^1 ^2 ^3Delaware Court of Chancery opinion in Hodes v. Mostaque
9. ^1 ^2Stability AI announces $101 million in funding
10. ^1 ^2Stability AI announcement on Emad Mostaque's resignation
11. ^1 ^2 ^3 ^4Stability AI announces new investment and appoints Prem Akkaraju
12. ^Stability AI: Stable Diffusion public release
13. ^SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis
14. ^Scaling Rectified Flow Transformers for High-Resolution Image Synthesis
15. ^1 ^2Stability AI: introducing Stable Diffusion 3.5
16. ^Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets
17. ^Stable Audio Open
18. ^Stable Audio 3 technical report
19. ^1 ^2 ^3Stability AI: Stable Audio 3 release
20. ^Stability AI launches Stable LM
21. ^SF3D: Stable Fast 3D Mesh Reconstruction
22. ^1 ^2 ^3Stability AI Acceptable Use Policy
23. ^AWS: Stability AI models in Amazon Bedrock
24. ^Stability AI and WPP partnership
25. ^Arm and Stability AI on-device audio collaboration
26. ^Electronic Arts partnership with Stability AI
27. ^Universal Music Group and Stability AI alliance
28. ^1 ^2Stability AI: introducing Brand Studio
29. ^Stability AI board of directors
30. ^1 ^2 ^3Getty Images v. Stability AI: UK High Court judgment
31. ^1 ^2 ^3Getty Images v. Stability AI: April 2026 US order
32. ^1 ^2Andersen v. Stability AI: June 2026 discovery order
33. ^1 ^2Stanford Internet Observatory: Identifying and Eliminating CSAM in Generative ML Training Data and Models
34. ^1 ^2LAION: Re-LAION-5B safety revision
35. ^1 ^2Stability AI safety practices
36. ^Stability AI annual integrity transparency report
37. ^1 ^2Stability AI joins the Tech Coalition
38. ^1 ^2 ^3 ^4 ^5 ^6 ^7Stability AI (Louisa Marshall): The Entertainment Industry's Biggest Names Back Stability AI in Latest Funding Round, August 25, 2026
39. ^1 ^2TechCrunch (Lucas Ropek): Stability AI, maker of image generator Stable Diffusion, raises $76 million in fresh funding, August 25, 2026
40. ^1 ^2Music Business Worldwide (Tim Ingham): Universal, Sony, Warner join $76M funding round in Stability AI, August 25, 2026
41. ^Tech.eu (John Reynolds): Stability AI raises $76M, with Warner and Sony investing, August 25, 2026
42. ^1 ^2TechCrunch (Kyle Wiggers): Stability AI, the startup behind Stable Diffusion, raises $101M, October 17, 2022
43. ^1 ^2 ^3TechCrunch (Kyle Wiggers): Stability AI lands a lifeline from Sean Parker, Greycroft, June 25, 2024
44. ^1 ^2TechCrunch (Connie Loizos): Sean Parker is rebuilding Stability AI around music, October 2, 2026
45. ^Variety (Carolyn Giardina): James Cameron Joins Stability AI Board of Directors, September 24, 2024
46. ^Stability AI: Introducing Our New Chief Pipeline Architect, Robert Legato, March 19, 2025
47. ^Stability AI: Warner Music Group and Stability AI Join Forces To Build The Next Generation Of Responsible AI Tools For Music Creation, November 19, 2025
48. ^1 ^2Stability AI (Louisa Marshall): Sharing a new way to work with Stable Audio, August 18, 2026
49. ^Stability AI: Stability AI Introduces Stable Audio 2.5, the First Audio Model Built for Enterprise Sound Production at Scale, September 10, 2025
50. ^High Court of England and Wales: Getty Images (US), Inc & Ors v Stability AI Ltd, neutral citation 2025 EWHC 3343 (Ch), December 16, 2025
51. ^US District Court, N.D. Cal.: Defendants' answer to complaint, Getty Images (US), Inc. v. Stability AI Ltd., No. 3:25-cv-06891-TLT, May 21, 2026
52. ^1 ^2US District Court, C.D. Cal.: Complaint, EVOX Productions, LLC v. Stability AI, Inc., No. 2:26-cv-07201, July 2, 2026
53. ^ChatGPT Is Eating the World: Stability AI, DeviantArt, Runway AI each file motions to dismiss EVOX Productions' claims, October 16, 2026
54. ^NPR (Huo Jingnan), via St. Louis Public Radio: Class action suit against AI makers over deepfake child sexual abuse material expands, July 9, 2026
55. ^Cao et al.: 4Director: Controlling Video World Models with Rigid 3D Geometry, arXiv, October 1, 2026
56. ^Dereviannykh et al.: SemanTok: Predictable Semantic Tokens for Efficient Autoregressive Video Generation, arXiv, September 30, 2026
57. ^Dihlmann et al.: Arbor: Explicit Geometric Conditioning for Controllable 3D Asset Generation, arXiv, June 22, 2026
58. ^Stability AI: Research, accessed October 27, 2026
59. ^Stability AI: Board of directors (archived copy), December 17, 2025