LiblibAI

1.1 Identity, product stack, and operating scope

LiblibAI should be understood as a platform company rather than a single image generator. The current public record ties the image community, the LibTV video workspace, and the Xingliu design agent together under the same Beijing operator, Beijing Evoken Technology, through updated user-agreement and privacy-policy language. That matters because it means the company is already trying to unify account, content, and API surfaces across creator, team, and developer workflows rather than running isolated point products. The product evidence is consistent with that framing. LiblibAI markets an image-creation community, a model and LoRA ecosystem, creator points and memberships, and a commercial API. LibTV extends the stack into professional video creation, while Xingliu positions itself as a design agent. The strategic through-line is workflow aggregation: image assets, model training, video production, and design delivery all live inside one ecosystem. This gives later chapters a clean ground truth: Evoken is building creative-AI infrastructure with community distribution at the top and monetized tools beneath it. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CO001, CO002, CO003, CO004, CO005, CO006]

1.2 Founder anchor, investor base, and public scale signals

Chen Mian is the clearest founder anchor in the public record. Multiple independent reports connect him to ByteDance, especially the Jianying/CapCut commercialization stack, and present May 2023 as the practical founding window for LiblibAI/Evoken. That founder profile matters because the investor roster is unusually strong for a two-to-three-year-old application company. The 2025 Series B was framed as the largest single AI-application financing in China that year, while the June 2026 B+ round pushed post-money valuation above $2 billion with Granite Asia, Tencent, and Shunwei co-leading. Public scale claims are also large enough to matter: more than 30 million cumulative LiblibAI users, more than 500,000 original models, more than 10 million Xingliu users, and nearly 1,000 professional teams using LibTV. These are still mostly company-reported numbers rather than audited operating KPIs, but they collectively support the thesis that LiblibAI has already crossed from niche tool into a broad creator platform with meaningful commercial reach. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CO011, CO012, CO013, CO014, CO015, CO016]

1.3 Milestones, governance opacity, and early frictions

The milestone path is unusually compressed. Public sources describe angel financing only months after formation, a 2025 Series B, a 2026 B+ unicorn round, a 2.0 product upgrade, the launch of Xingliu, and the March 2026 launch of LibTV. That speed is a strength, but it also creates diligence asymmetry. The company publishes operational policies and product surfaces, yet the public record still does not reveal a reconciled board roster, verified headcount, detailed ownership structure, or audited financial package. In addition, the risk picture is not hypothetical. China’s AI-labeling rules now apply to image and video platforms, and critical reporting has already highlighted moderation gaps and the possibility that LiblibAI’s aggregation-led model remains structurally vulnerable if upstream models reduce price or offer better native workflows. The company overview therefore supports a balanced conclusion: LiblibAI is real, large, and fast-growing, but still public-data-light relative to the valuation and strategic ambition it now carries. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CO021, CO022, CO023, CO024, CO025, CO026]

1.4 Exhibits

2.1 Market boundary, included spend, and buyer map

The narrowest useful market definition for LiblibAI is not “all generative AI” but the spending tied to visual-content creation workflows that can be monetized through community, software, or API access. That includes creator image generation, professional AI video production, AI-assisted design, model training or sharing, and adjacent developer usage where teams embed creative generation into downstream products. The market should exclude pure foundation-model training economics and most horizontal office AI spend. On the demand side, the company serves at least four visible buyer clusters. Individual creators use image and model tools for ideation and production. Professional video teams and short-drama studios use LibTV for higher-output workflows. Designers and agencies use Xingliu or Lovart-style design agents. Developers and automation teams consume image APIs and model assets programmatically. This matters because budget ownership, retention logic, and willingness to pay differ sharply across those groups. LiblibAI’s market is therefore multi-segment and workflow-bound, not one uniform creator-software bucket. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CM001, CM002, CM003, CM004, CM005, CM006]

2.2 Sizing lenses and demand signals

The broadest third-party lens comes from Research and Markets, which sized generative AI in creative industries at $5.38 billion in 2026 and forecast $14.03 billion by 2030. That is a real category signal, but still too broad to treat as LiblibAI’s practical near-term TAM. More grounded signals come from the sub-markets that already show monetization. Sensor Tower said global short-drama app downloads exceeded 850 million in Q1 2026 with roughly $750 million of IAP revenue, while Business of Apps described a 2025 ecosystem with over 700 monthly active micro-drama app advertisers and creative volume per advertiser up 144.9% year over year. ThinkChina’s Caixin-backed deep dive adds a further enterprise lens: AI video is one of the few generative applications already producing visible revenue across advertising, e-commerce, and entertainment, and Douyin estimated the enterprise AI video application market could reach $36 billion by 2030. For LiblibAI, the most defensible interpretation is that the core commercial wedge is the production side of image, design, and video workflows, where marketing and content budgets already exist. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CM012, CM013, CM014, CM015, CM016, CM017]

2.3 Adoption drivers, switching frictions, and constraints

Market growth alone will not guarantee capture. The strongest adoption drivers are creative-efficiency gains, cheaper iteration relative to traditional studio pipelines, and the ability to compress previously fragmented tools into one workflow. LiblibAI’s own product surfaces reinforce that thesis with points-based API access, memberships, shared account systems, and creator-community discovery. The constraints are equally visible. China’s labeling rules and broader regulatory expectations force ongoing moderation and metadata work on every image and video platform. Copyright and safety issues are already shaping the sector, especially in AI video. Competitive pressure is also high because buyers can multi-home. Communities like Civitai compete on models and discovery, while Adobe Firefly, Canva, Runway, Kling, and upstream model vendors compete on workflow quality, brand trust, or direct generation. The practical market takeaway is that LiblibAI has access to a large and expanding demand pool, but the reachable near-term market is the subset where workflow integration, Chinese creator density, and price-performance are strong enough to outweigh regulatory load and easy switching. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CM023, CM024, CM025, CM026, CM027, CM028]

2.4 Exhibits

3.1 Direct, incumbent, and adjacent rivals

LiblibAI’s direct rivalry starts with creative communities and creator tools rather than with every frontier model company. Civitai is the clearest global community analogue because it organizes models, images, videos, and creators in one place. Midjourney competes on image quality and creator mindshare even though its collaboration model is different. Adobe Firefly and Canva compete from the opposite direction: they begin with workflow trust, installed design behavior, and brand relationships, then fold in AI generation. Runway and Kling matter because they set the pace in AI video, where LibTV is trying to move from demo appeal into true production use. This landscape matters because LiblibAI is trying to win across multiple battlefields at once: creator community, image tools, video workflow, and design-agent experience. That breadth is a strategic opportunity, but it also means the company rarely faces one weak or fragmented rival set. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CP001, CP002, CP003, CP004, CP005, CP006]

3.2 Pricing, distribution, and switching economics

Competitive dynamics here are shaped by low switching costs. Runway publishes multi-tier creator pricing and positions itself as an AI video brand with broad creator and enterprise reach. OpenAI sells business seats and enterprise plans, making it a budget alternative for some teams that can tolerate horizontal tools. Adobe and Canva defend design budgets through familiar workflows, distribution, and brand trust. Civitai competes on community and discovery rather than enterprise polish. LiblibAI’s advantage is that it can blend community, model inventory, and workflow utility inside one product family. Its weakness is that buyers can multi-home across many of these tools, especially when upstream model vendors improve quickly. 36Kr’s “AI middleman” critique captures the key risk: if the best model quality and the best UX converge elsewhere, aggregation alone may not remain enough. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CP013, CP014, CP015, CP016, CP017, CP018]

3.3 Moat durability and competitive verdict

The strongest moat candidate is not proprietary model ownership; it is ecosystem density. LiblibAI has a Chinese creator base, a large model library, a shared-account structure across products, and evidence that it is extending into team and API usage. That combination can create real distribution power if creators, agencies, and video studios start treating the platform as a default workspace rather than a cheap alternative. But the moat is still conditional. Competitors such as Adobe and Canva own workflow trust; Runway owns global AI-video brand equity; Civitai owns a strong model-community identity; and Chinese video leaders are moving fast under intense commercialization pressure. LiblibAI therefore looks differentiated but not yet insulated. The competitive underwriting stance should be that the company has a plausible path to platform status, but it still operates in a category where many users can switch tools whenever price, quality, or queue times move against them. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CP025, CP026, CP027, CP028, CP029, CP030]

3.4 Exhibits

4.1 Revenue model, monetization surfaces, and pricing logic

The public record is clear that LiblibAI is no longer a pure traffic story. Its official API page shows direct monetization through point-based plans, commercial rights, and custom quotas. The VIP page adds recurring consumer-like subscription logic through memberships and bundled points. LibTV and Xingliu broaden that into workflow spending, where teams are paying not just for generation but for a production environment. Media coverage of the B and B+ rounds reinforces the same point: investors are underwriting a product family that monetizes creators, professional teams, and developers simultaneously. That matters because a multi-surface revenue stack is usually higher quality than one-off consumer novelty revenue. The main open question is mix. Public sources still do not disclose how much of ARR comes from membership, API, video teams, design-agent usage, or large custom deals, so the quality of the top line remains only partially visible. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CI001, CI002, CI003, CI004, CI005, CI006]

4.2 Unit economics, delivery costs, and cost-structure inference

The core economic debate is whether LiblibAI is building software-like margins or simply repackaging expensive upstream model capacity. The adverse 36Kr analysis argues that AI aggregators can be squeezed on price, queue time, and model quality if upstream vendors improve or cut prices. That is the right skepticism to apply. At the same time, the company’s business model is broader than raw generation resale. Community distribution, model libraries, workflow orchestration, and team collaboration can all support software-style value capture if they meaningfully improve throughput or retention. Benchmark filings from Adobe, Autodesk, Duolingo, and C3 AI do not make LiblibAI directly comparable, but they do show what investors reward: durable revenue growth paired with visible gross-margin structure, operating leverage, and enough differentiation that compute or content costs do not consume the business. Public evidence today proves strong demand and monetization, but not yet a verified margin profile. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CI013, CI014, CI015, CI016, CI017, CI018]

4.3 Capital adequacy, funding dependency, and underwriting gaps

The June 2026 B+ round reduces immediate financing pressure, but it does not remove the need for hard diligence on burn and runway. A company scaling image, video, agent, and API products at once will almost certainly carry meaningful compute, model-access, moderation, and go-to-market costs. Public sources say ARR exceeded $300 million as of May 2026 and growth surpassed 3000% year over year, which is impressive, yet those figures are not accompanied by cash balance, gross margin, operating loss, deferred revenue, or capex disclosure. The most defensible financial verdict is therefore balanced. LiblibAI looks commercially real and unusually fast-growing for its age; however, the public record is still too thin to determine whether the company is compounding efficiently or merely spending aggressively into a hot category. The next round should not be viewed as inevitable, but neither can current capital adequacy be underwritten from public evidence alone. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CI025, CI026, CI027, CI028, CI029, CI030]

4.4 Exhibits

5.1 Product surface, SKUs, and workflow definition

LiblibAI’s product scope is broader than the company’s name suggests. The reviewed official and semi-official materials show at least five meaningful product surfaces: the core creator/model community, VIP subscriptions, API access, LibTV for AI video production, and Xingliu for design-agent workflows. Additional pages around digital humans, brand LoRA use cases, model upload, and pretraining show that the company is not just offering a consumer interface but is also trying to organize the supply side of creative assets and reusable model components. The important product takeaway is that LiblibAI is selling a workflow environment in which creators can discover assets, generate outputs, train or upload models, commercialize via API, and then move into richer design or video use cases. That breadth makes the platform more interesting than a single prompt box, even if the breadth also raises complexity. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CE001, CE002, CE003, CE004, CE005, CE006]

5.2 Architecture, dependencies, and delivery model

The technical architecture described in public sources is a workflow-and-orchestration layer. LibTV’s clearest differentiator is its infinite canvas plus node-based workflow, which turns scriptwriting, shot design, model calls, and editing into a structured production graph instead of a chat interaction. That is a genuine product-architecture choice, and it aligns with the needs of teams producing repeat video output. The same pattern appears elsewhere in the product family: API pages emphasize custom model access and commercial rights, model-upload and pretraining pages emphasize creator supply, and Lovart/Xingliu materials emphasize agent-assisted design delivery. The dependency profile is equally important. Public sources repeatedly suggest that the company integrates multiple upstream models rather than owning every core generation engine itself. That speeds time to market and broadens capability coverage, but it also means the product stack must constantly defend its value above the model layer. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CE013, CE014, CE015, CE016, CE017, CE018]

5.3 Differentiation, trust, and roadmap

LiblibAI’s strongest product differentiation is not secret model IP; it is the combination of creator community, Chinese-language workflow design, asset reuse, and multi-product extension from image into design and video. That can be durable if users start treating the platform as a default operating layer. But the trust surface is still mixed. The privacy policy and user agreement show unified accounts, policy coverage, and moderation obligations, while China’s labeling regime means image and video outputs need explicit compliance handling. Yet public evidence still does not provide enterprise-grade uptime reporting, model eval benchmarks, or detailed security architecture. The roadmap signal is nonetheless strong: LibTV added team collaboration and additional workflow features within months, and the broader product family keeps expanding into new creation surfaces. The underwriting conclusion is that product velocity is clearly high, but technical defensibility remains more architectural and ecosystem-driven than model-proprietary. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CE025, CE026, CE027, CE028, CE029, CE030]

5.4 Exhibits

6.1 Customer segmentation and adoption trajectory

Public evidence suggests that LiblibAI serves several distinct customer layers rather than one monolithic user base. The broadest layer is the creator community tied to image generation, prompt/model discovery, and memberships. A second layer is developers who integrate creative generation through the API. A third layer is design users, where Xingliu and brand-style workflows imply team or commercial usage. The most commercially important layer may now be LibTV’s professional buyers—short-drama studios, film teams, ad agencies, and brand customers—because that cohort appears closer to workflow budgets than hobby experimentation. Adoption signals are unusually strong in public for such a young company: 30 million cumulative users, 500,000-plus original models, 10 million-plus Xingliu users, more than 100,000 LibTV visits on launch day, nearly 1,000 professional teams served, and more than 300 Team Edition business customers. These are not perfect customer-quality metrics, but they clearly indicate real demand across multiple cohorts. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CU001, CU002, CU003, CU004, CU005, CU006]

6.2 Named customer proof and usage quality

Customer proof is strongest around LibTV because that product is attached to professional production outcomes. BaiduWiki and related coverage say LibTV Team Edition quickly signed more than 300 short-drama companies and film studios after launch, while Firecat says the broader platform served nearly 1,000 short-drama teams, film studios, advertising companies, and brand customers. The same body of material describes team features such as shared canvases, asset libraries, permission management, and project handoff—signals more typical of repeat production workflows than casual consumer play. Another important proof point is the AI short drama “The Laid-Off Girl,” which BaiduWiki says was produced entirely using LibTV. That does not equal a broad case-study library, but it does show at least one real production outcome tied to the platform. The rest of the customer picture is more diffuse. Creator and design adoption are supported by user counts, model inventory, and brand-style workflow pages, yet public sources still stop short of naming many large recurring customers or quantifying contract value. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CU013, CU014, CU015, CU016, CU017, CU018]

6.3 Durability, expansion, and concentration

The central customer-diligence question is durability. Public evidence strongly supports acquisition and breadth, but much less clearly supports retention and net expansion. The community model should help low-end acquisition because creators can discover models and examples before paying. The API and team products create plausible expansion paths from trial usage into embedded or collaborative workflows. LibTV’s rapid addition of Team Edition, asset libraries, and production features suggests management is intentionally moving toward higher-value, repeat-use accounts. Yet none of the reviewed sources disclose GRR, NRR, churn, contract length, or top-customer concentration. That matters because AI creator tools often look strong on traffic while remaining weak on durable paid behavior. Multi-homing is also easy, especially in image generation and early-stage video workflows. The practical customer verdict is therefore that LiblibAI has unusually strong public proof of adoption for its age, but only moderate public proof of retention quality and concentration safety. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CU025, CU026, CU027, CU028, CU029, CU030]

6.4 Exhibits

7.1 Regulatory and legal risk ranking

The first risk bucket is regulatory and legal because LiblibAI operates directly in synthetic image and video distribution. China’s AI-generated-content labeling rules and the deeper deep-synthesis framework make labeling, provenance, moderation, privacy, and platform governance product-level obligations. This is not a box-checking exercise: a company distributing creative assets and AI video at scale can face reputational damage, enforcement exposure, or customer trust erosion if labels are missing, content controls are weak, or copyrighted material is mishandled. The reviewed policy and legal-analysis sources are directionally consistent that both generators and distributors have obligations. Public company policies show that LiblibAI is aware of platform governance, but awareness is not the same as proven control maturity. Legal risk is therefore manageable in theory, but only if operational controls scale with product breadth and usage growth. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CR001, CR002, CR003, CR004, CR005, CR006]

7.2 Operational, quality, and dependency risks

The second risk bucket is operational and dependency risk. LiblibAI’s differentiation depends on community density, orchestration, and workflow quality, yet many underlying capabilities appear to rely on external or fast-evolving model supply. 36Kr’s “AI middleman” critique is important precisely because it frames the business as vulnerable to upstream price cuts, queue-time competition, or native product improvements by model owners. The move into AI video raises the risk further: video workflows are more compute intensive, more safety sensitive, and harder to support reliably than simple image generation. Team features, asset libraries, and collaborative canvases expand customer value but also increase the consequences of poor reliability, weak permissions, or moderation failure. Operational risk is therefore not just outage risk; it is the compound risk that supplier dependence, workflow complexity, and content governance all fail at the same moment. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CR015, CR016, CR017, CR018, CR019, CR020]

7.3 Financial/model risk, execution risk, and mitigations

The final risk bucket is financial-model and execution risk. The public record shows extraordinary ARR and growth, but very limited burn, gross-margin, or retention disclosure. That leaves real uncertainty around whether the company is compounding with software-like economics or spending heavily to sustain growth in a crowded market. Customer concentration, cross-sell success, and enterprise support burden are also underdisclosed. Execution risk is higher because management is simultaneously scaling creator community, API usage, AI video, design agents, and compliance work. The mitigating factors are also real: fresh capital from the B+ round, very strong adoption signals, and evidence that the company can ship new products quickly. The appropriate underwriting view is that LiblibAI can be investable with risk controls, but only if diligence converts the current public narrative into a quantified view of margin, governance, retention, and dependency concentration. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CR029, CR030, CR031, CR032, CR033, CR034]

7.4 Exhibits

8.1 Investment thesis, anti-thesis, and financing context

The investment thesis starts with reality, not possibility. LiblibAI already appears to have crossed into genuine commercial scale through a combination of creator distribution, workflow breadth, and unusually rapid monetization. The company has public signals for 30 million users, 500,000 models, roughly $300 million ARR, and a B+ round at a valuation above $2 billion. That is a much stronger starting point than most AI-application businesses. The anti-thesis is equally important: public data still does not prove retention quality, gross margin durability, or insulation from upstream model competition. If the company is primarily an aggregator with shallow switching costs, a premium private multiple may be hard to defend through market cycles. Valuation therefore hinges less on whether LiblibAI is real and more on whether its workflow and ecosystem advantages are deep enough to justify paying above broad software medians. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CV001, CV002, CV003, CV004, CV005, CV006]

8.2 Comparable set, scenario framing, and range building

The best valuation lens is blended. Public creative-software and AI-application comps show that the market rewards growth, workflow stickiness, and visible margin structure, but penalizes commoditized or low-trust software quickly. Multiples.vc’s August 2026 view is especially useful because it shows design and engineering software around 4.2x NTM revenue, AI around 4.0x, productivity around 3.4x, and a much lower broad median around 2.2x. Against that backdrop, LiblibAI’s implied valuation of about 6.7x ARR is full relative to broad software, yet not outrageous if the company’s growth, retention, and product breadth are materially better than median. The scenario framework should therefore ask what kind of company LiblibAI is becoming: a durable workflow platform, a fast-growing but lower-margin AI utility, or a hype-rich business with weak retention beneath the traffic story. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CV014, CV015, CV016, CV017, CV018, CV019]

8.3 Recommendation, confidence, and final diligence asks

The correct recommendation is a qualified positive rather than an uncritical yes. Public evidence supports serious diligence and can justify interest at the current scale, but it does not justify underwriting solely from headlines. Confidence should therefore be medium, not high. Risk rating should be high relative to mature software because regulatory, dependency, and retention questions remain unresolved. Valuation stance should be described as fair-to-slightly-full on public evidence alone, with upside only if private diligence confirms strong net retention, healthy gross margins, disciplined burn, and genuine workflow stickiness in LibTV and adjacent products. The final diligence asks are correspondingly practical: prove retention, prove margin, prove control maturity, and prove that the most valuable accounts are not trivially multi-homing. If those tests fail, the $2B+ round begins to look aspirational rather than well-supported. From a diligence perspective, this public evidence is directional rather than fully institutional-grade. It shows how the company is positioning itself and where demand is visible, but it still leaves material unanswered questions around conversion quality, retention, governance, unit economics, and the durability of any apparent moat once private diligence data is introduced. The underwriting implication is that each public signal still needs private validation against cohort, margin, and control data.[CV027, CV028, CV029, CV030, CV031, CV032]

8.4 Exhibits