1. Key takeaways from the two weeks
Between Oct 23 and Nov 2, 2026 the multimodal market shipped little that was new and changed a lot about what can be called, what it costs and what rules apply to the output. Five conclusions follow from Chapters 2 to 7.

1.1 Five conclusions
1. The window held few launches: the clip market has converged on 30-second clips with audio, and the genuinely new item is an interactive world model. Only two video events fall inside the window: Genie 4, a world model announced as a research preview on Oct 27 for trusted testers with a waitlist, and Kling 4.0's API, available on fal by Oct 23 with single takes up to 30 s, double Kling 3.0's 15 s. Seedance 2.5 and Wan 3.0 also claim 30 s, but both launched earlier (context). Outside video, Hi3D V1.0 open weights (Oct 23) is the only dated downloadable release; Midjourney V9 is not out. [Chapter 2] [Chapter 3]
2. For builders, access moved more than capability. Google's Gemini API Veo 3.1 preview IDs were reported shut down on Oct 22, one day before the window, with Gemini Omni Flash, itself still a preview, as the replacement. Google Cloud's veo-3.1-generate-001 retires on Nov 17, 2026 or later, 14 days after Nov 3, which is a floor, not a commitment. Kling 4.0 is reachable through hosts such as fal, and no Kling-owned API page is in the material. [Chapter 4]
3. Price, not rank, separates the leading video models. On Artificial Analysis the top five text-to-video models sit within 18 Elo (1174 to 1156), yet the four with a public API list at $2.40 to $34.12 per minute, a 14.2x spread, and the snapshot carries no as-of date. On Runway's rate card a 10-second 1080p clip costs $2.00 on Wan 3.0 and $6.80 on Seedance 2.5. Sources disagree on Seedance 2.5 at 1080p: $0.68 per second at Runway, $0.90 at WaveSpeed. [Chapter 5]
4. Rights holders are suing and licensing in parallel, and no court has ruled on training. The Elvis Presley catalogue owners filed against Anthropic on Oct 23 over 274 songs, with statutory damages named at $48.635 million, which is an allegation, not an award. In the Suno cases a magistrate judge allowed a deposition on Oct 26, and Apple's subpoena fight has a hearing on Dec 1. Platform rules stay at labels and provenance, such as the EU AI Act marking duty in effect since Aug 2. [Chapter 6]
5. Money is landing on the distribution layer, and every growth figure is self-reported. Higgsfield added Nvidia and TCV to its Series B extension on Oct 23 with amount and valuation undisclosed, while claiming 40M+ users and $1B+ annualized revenue. Utopai Studios is reported at $60M on a $1B valuation. Runway both resells models and supplies plugins to Adobe's editors. [Chapter 7]
The window moved less on capability than on access, price and rules: model IDs, price lists and labels are dated facts to track, and distribution and licensing, not clip length, are where the contest is shifting.
1.2 What changed at a glance
| Model / product | Change | Date | Status | Source |
|---|---|---|---|---|
| Gemini API Veo 3.1 previews | Shut down; Omni Flash named as replacement | Oct 22 (context, day before window) | Reported in effect | AI Foresight 360; ReviewRamble |
| Kling 4.0 API | Opened to outside developers on fal | Oct 23 (sources say Oct 22 to 28) | API available via hosts | fal on X |
| Hi3D V1.0 | Code and weights released | Oct 23 | Open weights; vendor claims unverified | PR Newswire; OrcaRouter |
| Higgsfield | Nvidia and TCV join Series B extension | Oct 23 | Company statement | PR Newswire via Yahoo Finance |
| Elvis rights holders v. Anthropic | Copyright complaint, 274 songs | Oct 23 | Allegation; pending | CellCog |
| ElevenLabs v4 | clone_strength control, fast LLM tier |
Oct 26 | Shipping | Releasebot |
| UMG and Sony v. Suno | Deposition of former Menlo Ventures principal allowed | Oct 26 | Order; case pending | AI Musicpreneur |
| Genie 4 | World model announced, trusted testers | Oct 27 | Research preview; waitlist | Google DeepMind |
Google Cloud veo-3.1-generate-001 |
Retirement | Nov 17, 2026 or later | Upcoming | Google Cloud docs |
2. New video and world models
Chapter 1 summarises the window; on the video side only two events fall inside Oct 23 to Nov 2, and everything else below is context.
2.1 Release timeline
Two window events, not ten: Genie 4 was announced as a research preview on Oct 27, and Kling 4.0's API opened to outside developers on Oct 23 by the earliest dated statement (sources range from Oct 22 to Oct 28). Seedance 2.5, Hailuo H3, Wan 3.0 and Kling 4.0's app launch all predate Oct 23.
Specs are vendor claims unless stated; rankings and prices are in Chapter 5.
| Model | Maker | Date | Access | Max clip | Resolution | Native audio | Status |
|---|---|---|---|---|---|---|---|
| Genie 4 (world model) | Google DeepMind | Oct 27 (window) | Trusted testers; waitlist | Not stated | Not stated | Yes | Research preview |
| Kling 4.0 API | Kuaishou | Oct 23 (window; sources say Oct 22 to 28) | API via fal and resellers | 30 s | 720p, 1080p live | Stereo | API available via fal, Oct 23 |
| Kling 4.0 | Kuaishou | Oct 15 (context) | App | 30 s | 4K / HDR not yet live | Stereo | Context |
| Gemini Omni Flash | Jun 30 (context) | API preview | 3 to 10 s | 720p at preview | Not stated | Preview | |
| Seedance 2.5 | ByteDance | Jul 31 (context) | Jimeng; hosts | 30 s | 1080p; 4K tier | Yes | Context |
| Wan 3.0 | Alibaba | Aug 6 beta (context) | Hosted only | 30 s (some listings 15 s) | 480p to 1080p | Yes | Context |
| MiniMax H3 | MiniMax | Jul 31 (context) | App, API | 15 s | 2K | Stereo | Context |
Kling 4.0 timing conflicts. Kuaishou's own page is not in the material, so every date is second-hand. A reseller's Oct 16 comparison called the model "fully released" on Oct 15 but listed Kling's own developer API as "coming soon". A paid SelfyzAI press release (vendor claim) says the API "officially launched on October 22" and was opening gradually to enterprises and developers. fal posted on Oct 23 at 3:00 PM that Kling 4.0 was "publicly available" with API access "open to all customers". CellCog says the API opened on Oct 28. This chapter follows fal's timestamped Oct 23 post as the earliest dated public statement; none of the three is Kuaishou's own. A review site updated Oct 23 still called API pricing unpublished; Atlas Cloud lists $0.11 per second (30% off $0.157), a Chapter 5 topic.
Not found as in-window releases: Runway's changelog (as crawled) tops out at Oct 2 with Seedance 2.5 Draft Mode in Runway Agent. Luma Ray 3.2, Grok Imagine 1.5 and Midjourney's image-to-video are listed by aggregators or a review with no release date stated. Midjourney's V2 video model is not out: a creator recap posted Nov 1 says the team is "starting to think about" it and that video becomes a bigger focus after V9, which that recap says is expected before year-end.
2.2 What is genuinely new versus incremental
New: Genie 4. DeepMind describes it as a world model with "unprecedented realism", natively multimodal so that music, ambience and interactive dialogue are generated with the world. Google's Korean blog carries the same announcement on Oct 28. It is a research preview for trusted testers; wider access starts with Google AI Ultra subscribers in Project Genie and early-access partners of the Gemini API and Google AI Studio, with a waitlist open. No clip length, resolution, price or date for general release is stated. Kling 4.0 is a step, not a category change. Its headline jump is single-take length from 15 s (Kling 3.0) to 30 s, plus up to 10 keyframes and up to 15 combined references. Seedance 2.5 and Wan 3.0 also claim 30 s; the distinctive part is keyframe control, which WaveSpeed says they lack on image-to-video. The September "4K and 10-bit HDR" headline was still "coming soon" on Oct 16, and a price tracker lists no 4K tier for Kling 4.0. As of Nov 3 the material shows no confirmation that 4K or HDR is live.
Incremental or older:
- Seedance 2.5 (Jul 31, context). Pixo reports a 180-second ultra-long mode on Jimeng, against 30 s standard; other sources cite 30 s. Treat 180 s as a platform vendor claim.
- Wan 3.0 (Aug 6 beta, wider rollout announced Aug 24, context). Weights are closed; Wan 2.2 remains the last open Wan release, under Apache 2.0, according to Kenerate (as of Oct 30). Clip length is disputed: 30 s versus 2 to 15 s on Versely.
- Gemini Omni Flash. It entered public preview on Jun 30 for 3 to 10 s clips at 720p via the Interactions API. Its role in the window is as the recommended replacement after the Veo 3.1 preview model IDs were shut down on Oct 22 (context, one day before the window), according to a third-party review citing Google's documentation; see Chapter 4.
- Sora (context). According to shattered.io, quoting OpenAI documentation, the Sora web and app ended Apr 26 and the API on Sep 24; see Chapter 4.
The read: the clip market is converging on 30 s clips with audio, not producing a new leader this fortnight.
2.3 World models versus clip generators
A clip generator turns a prompt into a fixed video (Kling 4.0: 3 to 30 s; Gemini Omni Flash: 3 to 10 s). A world model like Genie 4 is steered while it runs: users move and aim the camera with standard keyboard controls and change the weather, camera or characters with text prompts, and the model keeps the scene coherent between inputs.
Compared on Google's own description:
| Aspect | Clip generators (Kling 4.0) | Genie 4 (vendor description) |
|---|---|---|
| Interaction | Prompt, then wait for a clip | Real-time steering by keyboard and text |
| Coupling | Standalone video model | Built on Gemini Omni plus other Gemini models for reasoning |
| Stated use | Ads, storyboards, reference-driven scenes | Training environments for robots and agents, interactive experiences |
Genie 4 is the only world model in the material, so no head-to-head is possible. With waitlist-only access, builders cannot yet depend on it; clip generators remain the usable product.
3. New image, voice, music and 3D models
Chapter 2 covered video and world models; outside video, the window (Oct 23 to Nov 2, 2026) was quieter, because most big image, voice and music launches landed in the two weeks before it.
Only one non-video release in this material is both dated inside the window and downloadable: Hi3D V1.0 open weights (Oct 23). Everything else is context from Sept 8 to Oct 22, an alpha test, or an incremental changelog entry.
3.1 Image models
Rows dated before Oct 23 are context. Access and quality claims are vendor claims unless stated.
| Model (maker) | Date | Access / licence | Headline | Status | Source |
|---|---|---|---|---|---|
| Canva Craft (Canva Research) | Oct 22 (newsroom); Oct 23 (research-blog index) | In Canva and public API; weights not mentioned | 2K, 30+ aspect ratios; plans layout, then renders | GA; context, day before window | Canva |
| Canva Brush (Canva Research) | "a week after" Craft, about Oct 29; exact date not stated | Canva AI, Pro plans first | Design-editing model, not an image generator | Shipping for Pro | Canva |
| Midjourney Thinking mode, draft upscales, shared folders | Oct 2026 | Alpha website only | Rerun "thinking harder"; draft to full HD | Alpha | Woollyfern (YouTube) |
| Midjourney V9 | Expected before end of 2026 | Not released | Better coherence and prompt understanding | Upcoming (creator report) | — |
| Nano Banana 2.1 (Google) | Oct 6 | Gemini API, closed | Reported $0.0336 per 1K image | Context | shattered.io |
| FLUX 3 Image Klein (Black Forest Labs) | Oct 15 | Open weights, 5.6B | Free under $5M revenue | Context | shattered.io |
Canva Craft is the most documented release. Canva's newsroom says it is available now in Canva and through a public API, and calls it the company's first image model built from scratch. The architecture splits a planning model, which writes out composition, scale and text, from a rendering model. Canva says it debuted fourth on DesignArena, ranked #1 in People & Portraits (of 87 models) and Landscapes & Nature (of 85), and so made Canva Research the "second most advanced" lab. These are Canva's statements about a third-party arena; DesignArena itself is not in our material. Two cheaper variants, Craft-fast and Craft-ultrafast, are upcoming with no date. API pricing and licence terms are not stated. The newsroom page says Oct 22, while Canva's research index lists Oct 23. We follow the newsroom page, so Craft counts as context.
Canva Brush followed "just a week after" Craft, giving about Oct 29. Canva claims up to 69% faster edits and 46% lower cost per edit than its previous design agent in internal testing. It started from an open-weights model, and Canva does not name it.
Midjourney did not ship V9. A creator's monthly recap (posted Nov 1) says V9 training is going well and release is expected before year end, that Niji 8 is "any day", and that video V2 comes after V9. What shipped is alpha-site testing, and the creator advises against using the alpha site for everyday prompts. Unite.AI's review (Oct 23) tested V8.2 and found it renders short label text correctly but warps small text in video. According to shattered.io, V8.2 has been the default since July 24 and has no official API.
Not found in the material: Microsoft MAI-Image-2.7 and any in-window GPT Image or FLUX update.
3.2 Voice and speech
| Model (maker) | Date | Access / licence | Headline | Status | Source |
|---|---|---|---|---|---|
| Eleven Realtime v1 Mini / Max (ElevenLabs) | Oct 19 | Gradual rollout, closed | Whole voice-agent config as one model | Rollout; context | Releasebot |
| Eleven Realtime Translation (ElevenLabs) | Oct 19 (changelog); Oct 21 (announcement) | Enterprise workspaces, alpha, closed | Live speech-to-speech, 90+ languages | Alpha; context | — |
| Voxtral 3, four models (Mistral) | Oct 20 (changelog); Oct 22 (launch page) | Public Preview, hosted API, no public weights | Transcription, realtime, TTS | Preview; context | digitalapplied |
| ElevenLabs v4 clone_strength, fast LLM tier | Oct 26 | API, closed | Control over v4 voice clone adherence | Shipping | Releasebot |
Both headline voice launches predate the window. ElevenLabs released Eleven Realtime, a family of voice-agent models, with a Mini default and a Max variant that reasons before answering. It also released Realtime Translation, an enterprise WebSocket API. Voice Over Herald dates the announcement Oct 21 and calls it an enterprise alpha with 90+ languages. The changelog's Oct 19 date is earlier; we cite both.
Mistral Voxtral 3 is, per digitalapplied citing Mistral's changelog, four Public Preview models: two offline transcription, one realtime, one TTS. Shattered.io agrees on four products. Techjacksolutions describes five models, which we do not follow because the changelog-based list is more specific. Prices read Oct 26: Mini Transcribe 3 at $0.0035/min, Small Transcribe 3 at $0.004/min, Realtime 3 at $0.008/min, and TTS 3 at $13 per million characters against $16 for the previous Voxtral TTS. The Small model costs 0.004 / 0.0035 = 1.14x Mini, about 14% more, and the TTS cut is 3 / 16 = 18.75%. Voice agents cost $0.04 to $0.08 a minute, so a five-minute call at the low tier is 5 × $0.04 = $0.20 before telephony. Mistral's benchmark charts are its own, with no printed figures. Cross-vendor prices are in Chapter 5.
In-window voice news is an ElevenLabs changelog entry dated Oct 26. It adds clone_strength (0 to 1) for v4 Professional Voice Clones, a fast LLM service tier for ElevenAgents, and service-account API keys that expire in 15 minutes to 90 days. Consent is the open issue. Voice Over Herald notes Mistral's "Pro Voice" claims 30 minutes of audio is enough, and it found no published consent framework, while ElevenLabs requires a spoken verification for Professional clones.
3.3 Music
No music model in the material has a release date inside the window. The dated items are context:
| Product (maker) | Date | Access / licence | Headline | Status | Source |
|---|---|---|---|---|---|
| Suno V6 | Sept 9 | Paid app; downloads capped | Replaced v4.5 to v5.5 | Context | Undetectr |
| ElevenLabs Video to Music API | Oct 19 | API, closed | Scores up to 10 videos (200 MB combined) | API available; context | Releasebot |
| Stable Audio 3.0 (Stability AI) | Date not stated | Plugin and API | Tracks up to 6 minutes | Shipping | Undetectr; TechieHub |
Suno V6 launched on Sept 9 with three models, and Suno says it is moving "entirely onto the v6 generation". Undetectr, an aggregator, says Pro gets 20 downloads a month and Premier 60, with extra downloads at $2.99. TechieHub says Udio has blocked downloads since October 2025. Both rest on third-party summaries, not Suno or Udio pages. The ElevenLabs Video to Music API lets you attach a text description and up to 10 style tags, and can return music and sound effects as separate files. Court filings and label deals around Suno, Udio and ElevenLabs Music are in Chapter 6. For builders, music is the least changed modality in this window.
3.4 3D and world generation
| Model (maker) | Date | Access / licence | Headline (vendor claim) | Status | Source |
|---|---|---|---|---|---|
| Hi3D V1.0, formerly Sparc3D (Hi3D) | Oct 23 | Open weights, hi3d-open-license-agreement-1.0 per model card |
Sparse-grid resolution up to 1536³ | Released, in window | PR Newswire (paid release); OrcaRouter |
| Hi3D V3.0 | Date not stated | Commercial | 2048³ voxel resolution | Claimed available | PR Newswire |
Hi3D V1.0 is the window's one dated open-weights release. Hi3D's press release, a paid vendor statement, dated Oct 23, 2026, says source code and weights are online and the web service is free. Its claims: training data grew from about 500,000 to 2 million 3D assets, a 90% slicing success rate, and the "highest" multi-view fidelity in its own evaluation, with no figures printed. Hi3D also announced a credit cut of over 50%. OrcaRouter, which inspected the repository, says the release is inference code plus eight runtime files, takes an image plus a coarse mesh and outputs geometry only, and was measured on a 96 GB H20 with no CPU path. It also says no benchmark table and no independent benchmark exists. The press release frames it as a 3D generator and OrcaRouter as geometry refinement; we follow the repository description. Treat the quality claims as unverified.
Not found in the material: Tripo H3.1 and the L3G-World research paper, so neither is assessed here.
4. Access, retirements and open weights
Chapter 3 covered the image, voice, music and 3D models that shipped; this chapter covers what builders can still call, what has been switched off, and which weights they can download.
The video model Google steered Gemini API developers to in June was itself retired: the Veo 3.1 preview identifiers are reported shut down on Oct 22, 2026 (context, one day before the window), with Gemini Omni Flash, still a preview, as replacement. Google Cloud's
veo-3.1-generate-001retires on Nov 17, 2026 or later, 14 days after Nov 3.
No retirement dated inside the window appears in our material; the window's access news is Kling 4.0's API and Hi3D's open-source release.
4.1 Deprecation and retirement calendar
| Product | Action | Date | Replacement | Status on Nov 3 |
|---|---|---|---|---|
| Gemini API Veo 3.1, Fast, Lite preview IDs | Shutdown | Oct 22, 2026 | Gemini Omni Flash | Context, in effect (secondary sources) |
Google Cloud veo-3.1-generate-001, -fast-generate-001 |
Retirement | Nov 17, 2026 or later | None named | Upcoming |
Gemini API gemini-3.1-flash-tts-preview |
Shutdown | Nov 17, 2026 | gemini-3.8-flash-tts or -lite-tts |
Upcoming |
Gemini API veo-3.0-generate-001, -fast-generate-001 |
Shutdown | Jun 30, 2026 | Veo 3.1 preview IDs or Google Cloud GA | Context |
OpenAI Videos API, sora-2 aliases |
Removal | Sep 24, 2026 | None named | Context |
| Sora web and mobile apps | Shutdown | Apr 26, 2026 | None | Context |
Google's own page is missing from our evidence. Our snapshots of the Gemini API deprecations page are truncated or old; one still shows the Veo 3.0 shutdown as its latest Veo row. The Oct 22 date rests on BenchLM's tracker (via shattered.io, which says "scheduled"), ReviewRamble ("retired this week", Oct 23) and AI Foresight 360, which says Google's Veo documentation states it. We treat it as reported. Google says listed shutdown dates are the earliest possible, with exact dates given on notice.
The replacement is a preview. gemini-omni-flash-preview has been in public preview since Jun 30, 2026, via the Interactions API, making 3 to 10 second 720p videos with conversational editing. Veo 3.1 makes 4, 6 or 8 second clips. Google named the Veo 3.1 previews as the Veo 3.0 replacement on Jun 15, so that target lasted 129 days (Jun 15 to Oct 22).

Google Cloud. Veo 3.1 GA dates from Nov 17, 2025; the page lists "Nov 17, 2026 or later" as retirement of support on the Gemini Enterprise Agent Platform, adding that models may stay reachable through the Gemini API. "Or later" is a floor, not a commitment. The page shows pay-as-you-go unsupported, provisioned throughput and fixed quota supported, us-central1 only; the Lite row is cut off.
Sora (context). According to Dead Canary, quoting OpenAI's deprecations page, developers were notified Mar 24, 2026 and removal on Sep 24 named no replacement. Spheron says the apps ended Apr 26 and API calls now return 410 Gone; Make deprecated its OpenAI video modules for the same date. OpenAI's own page is not in our material.
4.2 API versus app availability
| Model | App | API | Status |
|---|---|---|---|
| Kling 4.0 | Kling app, released Oct 15 (context) | fal.ai, open to all customers, Oct 23 | API via third parties |
| Gemini Omni Flash | Gemini app (per HeyGen) | Gemini API | Preview |
| Wan 3.0 | Runway paid plans, Aug 24 (context) | Alibaba cloud | API only |
Kling 4.0. A SelfyzAI press release (Oct 23, paid distribution) says the API "officially launched" Oct 22 and is "gradually opening". fal.ai's dated post (Oct 23, 15:00) is the earliest verifiable in-window event, so we follow it. No Kling-owned API page is in our material, and ReviewRamble says first-party pricing is unpublished. Treat access as "API available through hosts" (prices: Chapter 5).
Google's lineup. Sources disagree on where Veo 3.1 survives: HeyGen (Oct 29) says Omni Flash replaced it inside the Gemini app; AI Foresight 360 says Google's Veo page still lists it in the Gemini app and Flow. Neither is a Google page, so we leave it open.
Seedance 2.5. Pixo said it launched on Jimeng (Dreamina) only on Jul 31; by Oct 23 ReviewRamble lists ModelArk API pricing.
4.3 Open weights and licences
| Model | Date | Weights | Licence as stated | Source |
|---|---|---|---|---|
| Hi3D V1.0 | Oct 23, 2026 (window) | GitHub, Hugging Face | Hi3D Open License Agreement 1.0 | — |
| LTX-2.5 | Aug 11 (context) | Open | LTX-2.x Community License | — |
| MiniMax H3 | Context | Hugging Face-tagged | MiniMax H3 Community License | — |
| Wan 3.0 | Aug 2026 (context) | Closed | None | — |
Hi3D V1.0. Hi3D (Beijing) announced code and weights for V1.0, formerly Sparc3D, on Oct 23 in a paid release, so its claims are vendor claims: about two million training assets, a 1536³ grid, a 90% slicing success rate. Thaki Cloud says "around October 27" and the licence is dated Oct 22; we follow the dated Oct 23 release. Hosted services and fine-tunes are allowed, and generated meshes are not "Model Derivatives". Thaki reports notice within 60 days above 200,000 monthly users or USD 1,000,000 monthly revenue, plus a "Powered by Hi3D 1.0" label; our licence copy is a third-party mirror cut off before those clauses, so treat them as unverified. It needs Linux and an NVIDIA GPU (compute capability 8.0 or newer), has no CPU path, validated its top stage only on a 96 GB H20, and outputs untextured geometry.
LTX-2.5. Per Vivimo, free commercially under $10M annual revenue, paid above, "not OSI open source", with machine-generated disclosure required. A reviewer confirms the threshold and notes add-ons carry their own licences, some research-only.
Wan and H3. Wan 2.2 (Jul 2025, Apache 2.0) is the last open Wan; later versions are hosted. ReviewRamble says H3's weights are off limits in the US, EU, UK and Korea.
4.4 What to do before the next date
- Nov 17 (upcoming): inventory Google Cloud
veo-3.1-*-generate-001andgemini-3.1-flash-tts-previewcalls; treat the Veo date as a floor and ask Google for a firm one. - Oct 22 (in effect): a
veo-3.1-*-previewcall is now an outage; move to Omni Flash (preview) or Google Cloud GA, and plan for 10-second clips. - Sora (in effect): confirm no
sora-2calls remain. - Kling 4.0: pin the host; first-party terms are unconfirmed.
- Weights: log each licence's revenue threshold (LTX $10M; Hi3D unverified) before deployment.
5. Quality and price comparison
Chapter 4 listed what builders can still call after the Veo 3.1 preview shutdown; this chapter asks what those models score in blind votes and what they cost.
The top five text-to-video models sit within 18 Elo (1174 to 1156) on Artificial Analysis, but the four with a public API list at $2.40 to $34.12 per minute, a 14.2x spread (34.12 / 2.40); price, not rank, decides most choices.
5.1 Quality leaderboard snapshot (Artificial Analysis)
As-of date: not stated; the pages are undated crawls, and they list models released in October 2026. Elo comes from blind pairwise votes fitted with a Bradley-Terry model; video boards anchor 1000 at Kling 3.0 1080p (Pro).
| Rank | Text-to-video with audio | Elo (95% interval) | AA price per minute | Source |
|---|---|---|---|---|
| 1 | Wan 3.0 (Aug 2026) | 1174 (1160-1188) | $12.00 | AA |
| 2 | Utopai X (Sep 2026) | 1168 (1153-1183) | no API | AA |
| 3 | MiniMax H3, 768p (Jul 2026) | 1158 (1144-1172) | $4.80 | AA |
| 4 | HeyGen Video 1 (Oct 2026) | 1157 (1142-1172) | $2.40 | AA |
| 5 | Dreamina Seedance 2.5 (Jul 2026) | 1156 (1142-1170) | $34.12 | AA |
| 10 | Gemini Omni Flash 1.1 (Aug 2026) | 1128 (1114-1142) | $9.12 | AA |
| Rank | Text-to-image | Elo | AA price per 1k images | Source |
|---|---|---|---|---|
| 1 | GPT Image 2.5 Sunburst (max) | 1222 | $210.7 | AA |
| 2 | GPT Image 2.5 Flare (max) | 1207 | $210.7 | AA |
| 3 | GPT Image 2 (high) | 1186 | $211.0 | AA |
| 4 | Nano Banana 2.1 (Oct 2026) | 1176 | $33.6 | AA |
| 6 | Grok Imagine Image 2.0 | 1166 | $60.0 | AA |
Reading the video board. The board's own rank ranges (Wan 3.0 1-3; HeyGen and Seedance 2.5 2-7) overlap, so places 1 to 5 are not cleanly separated. Leadership also flips by task: Seedance 2.5 leads image-to-video (1157) while Wan 3.0 is ninth (1111), and Wan 3.0 leads video editing (1196 against 1152). HeyGen Video 1 leads the silent text-to-video pool (1143). MiniMax H3 (768p) leads open-weights video (1158). Kling 4.0 (context: app launch Oct 15) is absent from the retrieved boards, which show only Kling 3.0.
Reading the image and voice boards. GPT Image 2.5 Sunburst also leads editing (1199). Nano Banana 2.1 trails it by 46 Elo (1222 - 1176) at 6.3x lower price ($210.7 / $33.6 = 6.27). FLUX 3 Image Klein leads open-weights images (1043, $9.0 per 1k). For speech, Eleven v4 and v4 Turbo tie at 1333 ($80 and $40 per 1M characters) and Mistral's Voxtral 3 TTS scores 1221 at $13.0.
5.2 Price table
Runway's developer rate card is the only page in the material that prices many makers' video models in one unit (1 credit = $0.01). It is Runway's resale price, undated in the material; the makers' own pages for Seedance, Wan and Kling are not in the material.
| Model (Runway Dev, per second) | Price | Source |
|---|---|---|
| Gen-4.5 | $0.12 | Runway Dev |
| Wan 3.0 | $0.05 / $0.10 / $0.20 at 480p / 720p / 1080p | Runway Dev |
| Seedance 2.5 | $0.20 / $0.30 / $0.68 at 480p / 720p / 1080p; minimum $0.80 per generation | Runway Dev |
| Kling 4.0 | list $0.157 (720p), $0.235 (1080p); 20% promo to 2026-11-30: $0.1256, $0.188 | Runway Dev |
| Gemini Omni Flash 1.1 | $0.10 (720p), $0.15 (1080p), $0.30 (4K) | Runway Dev |
| Veo 3.1 | $0.40 with audio, $0.20 without, same at 720p and 1080p | Runway Dev |
A 10-second 1080p clip therefore costs $2.00 on Wan 3.0 (10 x $0.20), $6.80 on Seedance 2.5 (10 x $0.68), $2.35 on Kling 4.0 at list (10 x $0.235) and $4.00 on Veo 3.1 with audio (10 x $0.40).
Where sources disagree. Seedance 2.5 at 1080p per second: Runway $0.68, WaveSpeed $0.90, fal about $1.164 (a per-token estimate), and AA $34.12 per minute, or $0.569 (34.12 / 60). I follow Runway for cross-model comparison because it is one rate card, and AA inside its own boards. At 720p, an August 20 snapshot (context) listed six resellers from $0.134 to $0.473 per second, a 3.5x spread. Wan 3.0 agrees everywhere: $0.20 per second at 1080p (AA $12.00 / 60). For Kling 4.0, WaveSpeed and fal list $0.157 and $0.235; according to UnificAlly, $0.126 and $0.189 included a 10% launch sale that ended Oct 29. I follow the list rate, which three sources share.
Subscriptions (Runway app, credits separate from API credits). Gen-4.5 costs 12 credits per second.
| Plan | Price per month (annual / monthly billing) | Credits | Gen-4.5 seconds | Annual $ per second |
|---|---|---|---|---|
| Standard | $12 / $15 | 625 | 52 (625 / 12) | $0.23 |
| Pro | $28 / $35 | 2,250 | 187.5 | $0.149 |
| Max | $76 / $95 | 9,500 | 791.7 | $0.096 |
All three rows: Runway. Max undercuts the API's $0.12, but the credit pools cannot be swapped.
Adobe, music and image. Adobe Firefly's native video model costs 100 credits per second at 1080p, per an Adobe community manager (Sep 15, 2026). The "unlimited" promotion ran to Jan 15, 2026 (context). Adobe's own credit FAQ and partner-model pages carry no rates in the material. According to techiehub, re-verified Oct 3 (context), Suno Pro is $10 a month for 2,500 credits and Premier $30 for 10,000; another list gives Premier at about $24, and I follow the first. According to Shattered, Midjourney starts at $10 a month with no official API.
5.3 Clip length and audio capability
These are vendor specifications from WaveSpeed's reseller listing, not test results.
| Model | Clip length | Resolution | Native audio | Source |
|---|---|---|---|---|
| Seedance 2.5 | 4-30 s | 480p-4K | yes, default on | WaveSpeed |
| Wan 3.0 | 2-30 s | 480p-1080p | yes, default on | WaveSpeed |
| Kling 4.0 | 3-30 s | 720p, 1080p | yes, stereo | WaveSpeed |
fal lists Seedance 2.5 only to 1080p. According to kie.ai (Oct 16), Kling's 4K and 10-bit HDR were announced but not live. Runway's Gen-4.5 is limited to 2-10 seconds at 720p.
Audio changes cost differently by vendor on Runway Dev: Kling 3.0 Pro $0.12 without audio and $0.17 with (+$0.05, +42%); Veo 3.1 $0.20 and $0.40 (2x); Veo 3.1 Fast $0.10 and $0.15. Seedance 2.5 on fal costs the same either way. AA ranks audio and silent output in separate pools.
5.4 What the comparison does not show
Elo is relative within one pool, so it cannot be compared across text-to-video, image-to-video, editing, or audio versus silent boards. AA prices "1080p where offered", yet its top H3 entry is 768p. Creator tests are thin: CellCog's ten one-take Seedance 2.5 versus Kling 4.0 tests went 10-0 on its picks and 7-3 on a strict review (408 against 401 of 700, a 1.7% gap), and CellCog says it builds on Seedance 2.5. WaveSpeed's comparison is spec-level only, and kie.ai says no standard benchmark exists for the pair. Nothing here measures cost per usable clip: one Adobe user (Dec 2025, anecdotal) reported 18 to 22 retries at 500 credits each, about 10,000 credits per usable five-second clip.
6. Copyright and platform policy
Chapter 5 compared what the models cost and how they rank; this chapter covers the legal and platform conditions attached to using their output.
Core claim: The window's one new complaint came from the owners of a single artist's catalogue: Elvis Presley rights holders sued Anthropic on 2026-10-23 over 274 songs. The Suno and Udio cases moved only on procedure, none has a final judgment, and platform rules stay at labels and disclosure, not bans.
6.1 Litigation and settlements
A complaint is an allegation. The material holds no docket pages, so statuses rest on press reports and law-firm trackers.
| Case | Court | Claim | Filing or event date | Status (2026-11-03) | Source |
|---|---|---|---|---|---|
| ABG EPE IP LLC and co-plaintiffs v. Anthropic, 5:26-cv-12473 | N.D. Cal. | Copyright, DMCA credit-removal and circumvention; 274 Presley compositions (lyrics, not masters) | Filed 2026-10-23 (window) | Pending; Anthropic says it will defend | CellCog; Law Commentary; NBC |
| Reservoir Media v. Suno and v. Anthropic | N.D. Cal. (Anthropic suit); Suno court not stated | 1,050 compositions against Suno | Filed 2026-10-21 (context) | Pending | CellCog; AI Musicpreneur video |
| UMG and Sony v. Suno, second suit, 1:26-cv-14275 | D. Mass. | 60,202 recordings; model v6 | Filed 2026-09-18 (context); deposition order 2026-10-26 (window) | Pending | Mishcon; AI Musicpreneur |
| Sony v. Udio, second suit | New York court | 30,117 recordings | Filed 2026-07-20 (context) | Ongoing | Boddominator |
| Disney, Universal and WBD v. Midjourney, 2:25-cv-05275 | C.D. Cal. | Visual and film case; claim detail not in material | Dispositive motions due by 2026-11-23 | Upcoming (tracker as of Oct 4) | Axis |
| Apple motion to quash Suno's subpoena (inside the UMG and Sony case) | Not stated | Testimony on Apple Music's AI deals | Suno opposed 2026-10-27 (window); hearing 2026-12-01 | Upcoming | Mubert |
Elvis v. Anthropic. The complaint says licensed lyrics were scraped from Common Crawl, that Anthropic removed credit information in data cleaning, and that Claude's lyric filters are "readily circumvented by rephrasing the prompt". Damages are statutory maxima: 274 works × $150,000 = $41.1 million (willful infringement); 274 × $25,000 = $6.85 million (credit removal); 274 × $2,500 = $685,000 (circumvention); total $48.635 million. It also seeks an injunction, destruction of copies and a jury trial. Anthropic: "We disagree with the claims and intend to defend this case fully". Pitchfork calls it one of the first high-profile suits by a single artist's catalogue owner. It follows publisher suits, including Sony Music Publishing and Warner Chappell's complaint of 2026-08-28 (context). Anthropic's $1.5 billion book settlement, finally approved 2026-07-20, is also context.
Reading across both complaints, the claims share one three-count structure (copying, credit removal, circumvention), so the stated exposure scales with the number of works listed.
Suno. The Reservoir figures reconcile: 1,050 × $150,000 = $157.5 million on the main claim; adding $25,000 and $2,500 per work gives $186.375 million, matching the "up to $186.4 million" in the video. For UMG and Sony: 60,202 × $150,000 ≈ $9.03 billion, a theoretical ceiling set against a $5.4 billion Suno valuation in June, according to that video. On 2026-10-20 Suno told the Boston court it "has not deleted all" training recordings; on 2026-10-26 a magistrate judge let the labels depose a former Menlo Ventures principal (window). GEMA won in Munich on 2026-07-31 (context); Suno can appeal. The material has no filing or docket event for the Disney-led studios' suit against MiniMax (Hailuo), so none is reported.
6.2 Licensing deals and training-data policy
No licence deal dated inside the window appears in the material; the deal map below is context, and deals are announced terms, not rulings.
- Udio: Universal settled in October 2025 and Warner in November 2025. Udio is a "walled garden" where nothing leaves the platform; downloads have been off since October 2025, and Warner, Merlin and Kobalt deals are signed, according to TechieHub, with restored downloads only hoped for later in 2026.
- Suno: Warner settled on 2025-11-25. Suno committed to retire its current models, train only on licensed works from 2026 and move downloads behind paid tiers. Suno v6 (launched 2026-09-09) was built with Warner, BMG and Believe, and UMG and Sony call it "the fruit of the same poisoned tree". Universal talks stalled in April 2026; Sony has settled with neither company.
- ElevenLabs Music: TechieHub and BuildMVPFast describe it as licensed-data, with Merlin and Kobalt deals. No ElevenLabs release is in the material, so this is "according to".
- Others: Klay signed all three majors on 2025-11-20. LANDR's Fair Trade AI pays artists 25% of net licensing revenue plus a $1 million advance fund (about $5 per track, recoupable), expanded July 2026.
On training-data policy, Suno's terms give it a perpetual, royalty-free licence to train on user creations with no opt-out, and the labels argue v6 launders its older models' alleged infringement. The window's contribution is fair-use evidence: Suno wants Apple's AI-clause contracts for the market-harm factor. Suno and Udio both argue fair use. The only US appellate ruling in the material, the Third Circuit's 2026-09-29 Thomson Reuters v. Ross, rejected fair use for a non-generative tool and reserved the generative question (context).
6.3 Platform labelling and disclosure rules
| Platform | Rule | Effective date | What creators must do | Source |
|---|---|---|---|---|
| YouTube | Disclose realistic altered or synthetic content; YouTube auto-labels content from its GenAI tools, with C2PA metadata, or detected by its systems, and those labels cannot be edited | Date not stated; July 2026 monetization clarification is context | Answer the Studio AI-disclosure survey; unrealistic animation needs no label | YouTube Help Community; Kenerate |
| Google Ads, Merchant Center | AI label setting; visible overlays for campaigns targeting the EU, India and New York; labels cannot be overwritten | Date not stated | Choose "Label this asset as created or edited with AI" per asset | Google Help |
| Google Play | Developer Program Policy version | In effect 2026-10-29 | Extract shows no AI-specific rule | Google Play Help |
| EU AI Act, Article 50(2) | Machine-readable marking of AI output | In effect 2026-08-02; systems already on the market by 2026-12-02 (upcoming) | Mark outputs | NHI Mgmt Group |
| TikTok | Label AI content that could mislead | Date not stated | Label it | Fliki |
The material has no TikTok or Meta policy page, so TikTok rests on a vendor blog and no Meta labelling rule is in the material.
On provenance, Google says it embeds SynthID and C2PA markup in all images and videos from its AI tools. According to HyprForge, the SynthID Detector portal opened on 2026-10-07 (context) and reads OpenAI, NVIDIA and Kakao marks; Gemini promoted its "Is this AI generated?" check on 2026-10-27 (window), limited to one file up to 100 MB. C2PA's index pointed to version 2.4 on 2026-10-02, and a credential proves neither truth nor cleared rights. NIST reports that most watermarks are removable. A creator reported on 2026-10-25 an uneditable AI label on a podcast with only noise reduction.
Likeness harms sit mostly outside model-maker policy in this material. Resemble AI logged a 2026-10-20 election deepfake and a 2026-10-22 Korean civil award of ₩17.7 million (both context).
6.4 What is still unresolved
- Fair use for generative training: no US appellate ruling; Suno and Udio rulings are pending.
- Elvis case: Anthropic had not answered the complaint as of 2026-10-25.
- Dated next steps (upcoming): Apple hearing 2026-12-01; Midjourney dispositive motions by 2026-11-23; EU marking deadline 2026-12-02.
- Gaps: MiniMax docket, TikTok and Meta primary policy, and Play's AI section are absent from the material.
Taken together, the window shows rights holders and platforms converging on licensing and provenance, while courts have not yet said whether training on copyrighted works is lawful. The dated, in-effect items are labels and disclosure; training-data exposure has no ruling date in the material, and nothing here is legal advice.
7. Commercialization and the creator ecosystem
Chapter 6 covered the rules and lawsuits around generated media; this chapter follows the money and the channels through which creators actually reach the models.
7.1 Funding and valuations
The only new financing in the window is Higgsfield's Series B extension, announced on Oct 23, 2026. Its size and valuation are undisclosed. Everything else in the table is either a report or context from before Oct 23.
| Company | Round | Amount | Valuation | Lead / participants | Date and status | Source |
|---|---|---|---|---|---|---|
| Higgsfield | Series B extension | Not disclosed | Not disclosed | New: Nvidia, TCV. Returning: Goldman Sachs Alternatives, Tribe Capital, AI Capital Partners, Smash Capital, GFT Ventures, Valor Capital | Oct 23, 2026; company statement (paid press release) | PR Newswire via Yahoo Finance |
| Higgsfield | Series B | $400M | $5.4B | DST Global lead | Aug 2026 (day not stated); context | MarketScale, citing Unite.AI |
| Higgsfield | Series A | $80M (according to Creatorflow, citing Sacra and Tracxn) | $1.3B | Not stated | Date not stated; context | MarketScale; Creatorflow |
| Utopai Studios | Growth equity | $60M | $1B | Investors not named | Reported; Dealroom page undated, earliest dated repost Oct 24 | The Information via Dealroom; Instagram |
| Kling AI (Beijing Kling, Kuaishou) | Subscription round | Limit RMB 20.4471B ($3B); press says $2.8B or "nearly $3B" | $15B pre-transaction | Kuaishou HKEX filing, Jul 2 | Limit fully used per Aug 31 update; context | gradually.ai citing Kuaishou; Briefs |
| Runway | Series E | $315M | $5.3B | Nvidia participated | Feb 10, 2026; context | gradually.ai; TMC Insight |
The Kling figures differ by source. Kuaishou's filing, as quoted by gradually.ai, gives a $3B subscription limit. Press reports give $2.8B or "nearly $3B". We follow the filing's limit as a ceiling and the press figures as the likely amount actually raised. The $15B is a pre-transaction valuation, not a post-money one.
Higgsfield's claims. All figures below are company statements that "have not been independently verified".
| Metric | Earlier | Oct 23, 2026 | Source |
|---|---|---|---|
| Registered users | 30M (at Aug Series B) | 40M+ | Higgsfield Instagram |
| Annualized revenue | $700M (Aug) | $1B+ (crossed in Sep) | MarketScale |
| Revenue under contract, business customers | Base: June | 10x | PR Newswire via Yahoo Finance |
| Countries and territories | n/a | 238 | PR Newswire via Yahoo Finance |
| Generations per month | n/a | 150M | MarketScale |
Run-rate growth is (1,000 − 700) ÷ 700 ≈ 43% between the August round and September, a run-rate rather than recognized revenue. Two disagreements matter. TMC Insight, via Bloomberg, says revenue is "approaching" $1B, while the company says it crossed $1B. We follow the company's wording but label it a vendor claim. Briefs says Higgsfield "added 10 million paying customers", whereas Higgsfield's own post says users grew from 30M to 40M+. We follow the company post: the 10M is registered users, not payers. Higgsfield's enterprise claim also appears in two forms: "contracts up 71%" since August and "revenue under contract up tenfold since June". The baselines differ, so they cannot be reconciled.
Utopai Studios is a different bet: it makes films rather than selling tools. Per the reports, it has more than 20 projects, including two TV series and three features planned for 2027 in about 3,000 North American theaters. It claims AI cuts scene-planning and CGI costs by 80% to 90%. It also licenses its technology and built its model on a customized MiniMax H3. Term sheets at a $2B to $3B valuation are reported but not completed. Artificial Analysis lists Utopai X second on its text-to-video board (Elo 1168, no API available; see Chapter 5).
7.2 Distribution and aggregator platforms
No new reseller deal is dated inside the window. The map below is therefore mostly context, but it shows who sells whose model.
| Platform | Whose models it carries | Status and date | Source |
|---|---|---|---|
| Higgsfield | About 35 third-party video models plus its own; Kling 3.0, Veo 3.1, Seedance, MiniMax Hailuo and Wan named | Shipping, Oct 2026 (aggregator account) | Wall St Engine; AI Foresight360 |
| Runway | Seedance 2.5 Draft mode (Oct 2), Wan 3.0 (Aug 24), Grok Imagine 1.5 (Aug 13), Kling models via MCP (Sep 18), ElevenLabs V4 (Sep 29) | In effect; all context | Runway changelog |
| Runway as supplier | Plugins in Adobe Premiere Pro and After Effects (Sep 8), DaVinci Resolve (Sep 23); Runway MCP in Cursor (Sep 23), Grok Bot (Sep 17), OpenAI's Dots (Oct 1) | In effect; context | Runway changelog |
| Adobe Firefly | Partner models GPT Image, Gemini (Nano Banana), FLUX, Runway, Kling | In effect, Sep 15, 2026 | Adobe community manager |
Three patterns stand out. First, Runway is both a buyer and a seller of models. Its Aleph model sits inside Firefly's Prompt to Edit (Adobe blog, Dec 16, 2025 per its URL, context), and Runway itself runs inside Adobe's editors. Second, aggregators outlive first-party retirements. According to AI Foresight360, Google's Gemini API shut its Veo 3.1 previews on Oct 22, yet Higgsfield still offers Veo 3.1 (see Chapter 4). Third, Nvidia's tie to Higgsfield is also infrastructure: it scaled on Nvidia GPUs from L40 and A40 through Hopper and Blackwell and is moving to Vera Rubin, running custom inference on Nvidia Dynamo.

7.3 Creator economics: credits, prices and monetization
Reseller pricing is credit-based, and the unit changes by platform. According to AI Foresight360, Higgsfield individual plans cost $9, $29 and $79 a month (monthly billing) for 120, 600 and 1,800 credits. That is $9 ÷ 120 = $0.075, $29 ÷ 600 ≈ $0.048 and $79 ÷ 1,800 ≈ $0.044 per credit. Creatorflow instead lists Starter, Plus and Ultra tiers, says the older Basic and Creator tiers are retired, and quotes no price because the page loads by script (checked July 2026). Plan names and prices therefore conflict; no vendor pricing page is in our material, so treat $9, $29 and $79 as unconfirmed.
On the same aggregator, an 8-second Pro Kling 3.0 clip cost 20 credits and Veo 3.1 cost 80 (test run Oct 22, 2026). At the rates above, Kling is 20 × $0.044 to 20 × $0.075 = $0.88 to $1.50 per clip ($0.11 to $0.19 per second), and Veo is $3.51 to $6.00 ($0.44 to $0.75 per second). Model prices are in Chapter 5. On Runway, according to Creatorflow citing runway.com/pricing (Oct 2026), the $15 per month Standard plan includes Kling 3.0 and Gen-4.5. Runway's enterprise offer of two free months, for contracts signed by "10/31" (year not stated), has now passed.
Monetization rules did not change inside the window in our material. According to Kenerate AI (updated Oct 31, 2026), YouTube pays AI videos unless the channel is "inauthentic" (mass-produced or template-based). The wording dates from Jul 15, 2025 and was clarified on Jul 16, 2026, both context. TikTok Creator Rewards requires 10,000 followers, 100,000 views in 30 days and videos of at least one minute. TikTok publishes no rate; reported cases give $362 to $1,035 per million qualified views, or $0.36 to $1.04 per 1,000 (Digiday, Sep 1, 2026, via Mubert). A $29 plan therefore needs about 29 ÷ 1.035 ≈ 28,000 to 29 ÷ 0.362 ≈ 80,000 qualified views a month to be covered, a sample-based estimate that excludes taxes and ineligible views.
7.4 What the money signals
Capital is landing on the packaging layer (Higgsfield) and on a studio that borrows open models (Utopai), while every growth number is self-reported.
A mixed-date check shows the caution needed. Higgsfield's August valuation over its September run-rate is $5.4B ÷ $1B = 5.4x, and Runway's February valuation over its September ARR is $5.3B ÷ $200M = 26.5x (Bloomberg, Sep 8, via gradually.ai). Each pairs figures from different months and the Higgsfield extension valuation is undisclosed, so these are rough ratios, not valuations. Nvidia appears on both Higgsfield and Runway, which TMC Insight reads as spreading bets across the field; that is an interpretation, not a company statement.
Market-size estimates disagree. Grand View Research puts AI video generators at $946.4M for 2026 and Meticulous Research puts video generation and editing software at $3.67B, while a LinkedIn market report (Oct 27, 2026) says $3.0B for 2026. The definitions differ, so we use none as a base. For builders, the signal is that distribution (resellers, plugins, agents) is where model makers now compete.
8. Trend assessment and what to watch next
Chapters 2 to 7 covered models, access, price, rules and money; this chapter reads them together and lists what to watch in the next 30 days.
The window shows a market that is shifting from model launches to control points: who hosts a model, what it costs per second, and what rules apply to its output. Capability news was thin, and the dated risks ahead are retirements, court deadlines and labelling duties.
8.1 Trends and how strong the evidence is
| Trend | Evidence in this window | Confidence |
|---|---|---|
| Audio-native generation is standard | Kling 4.0 ships stereo audio; Seedance 2.5 and Wan 3.0 audio on by default; Genie 4 generates music and dialogue with the world; AA ranks audio and silent output in separate pools | Strong |
| Clips to interactive world models | Genie 4 is the only world model in the material, waitlist-only, with no clip length, price or release date | Weak |
| Aggregators alongside model makers | Higgsfield carries about 35 third-party models; Runway resells and supplies; Higgsfield still offers Veo 3.1 after Google's shutdown | Strong |
| Licensing-first music | Suno v6 was built with Warner, BMG and Believe, but UMG and Sony still sue; no licence deal is dated inside the window | Mixed |
| Open weights | Hi3D V1.0 opened Oct 23; Wan 3.0 is closed and Wan 2.2 is the last open Wan; LTX-2.5 is free only under $10M revenue | Mixed |
The strongest signal is price dispersion. The same model costs very different amounts by seller, and the cheapest is not the top-ranked. The weakest is the world-model story, which rests on one vendor post.
8.2 What it means by reader
| Reader | What changed | Implication | Evidence |
|---|---|---|---|
| Solo creator | Credits differ by platform: Runway Max works out at about $0.096 per Gen-4.5 second; Higgsfield plans are unconfirmed | Compare cost per second, not plan price; Kling 4.0's 20% Runway promo ends Nov 30 | — |
| Studio or agency | Higgsfield and Utopai attract capital; growth numbers are company claims; platforms label AI output | Treat vendor revenue and user figures as unverified; budget for labelling and disclosure | — |
| Developer integrating APIs | Veo 3.1 previews gone since Oct 22; Cloud retirement Nov 17 or later; Kling 4.0 only via hosts | Inventory model IDs, move to Omni Flash (preview) or Cloud GA, and pin the Kling host | — |
| Rights holder | Elvis suit filed Oct 23; Suno deposition order Oct 26; Apple hearing Dec 1 | Dated procedural steps exist, but no ruling on training is scheduled in the material | — |
These are readings of the evidence, not legal or investment advice.
8.3 Key dates
Dates before Nov 3, 2026 are in effect or the matching status word; later ones are upcoming.
| Date | Event | State on Nov 3, 2026 | Source |
|---|---|---|---|
| Apr 26, 2026 | Sora web and mobile apps ended (context) | In effect | Spheron |
| Aug 2, 2026 | EU AI Act Article 50(2) marking of AI output | In effect | NHI Mgmt Group |
| Sep 24, 2026 | OpenAI Videos API removal (sora-2) (context) | In effect | Dead Canary |
| Oct 15, 2026 | Kling 4.0 app launch (context) | In effect | Atlas Cloud, ReviewRamble |
| Oct 22, 2026 | Gemini API Veo 3.1 preview IDs shut down (reported) | In effect | AI Foresight 360 |
| Oct 23, 2026 | Kling 4.0 API on fal; Hi3D V1.0 weights; Elvis v. Anthropic filed | API available; open weights; pending | — |
| Oct 26, 2026 | Deposition order in UMG and Sony v. Suno | In effect | AI Musicpreneur |
| Oct 27, 2026 | Genie 4 research preview announced | Preview; waitlist | Google DeepMind |
| Oct 29, 2026 | Kling 4.0 launch sale at UnificAlly ended | Ended | UnificAlly |
| Nov 17, 2026 | Google Cloud veo-3.1-generate-001 retirement ("or later"); gemini-3.1-flash-tts-preview shutdown |
Upcoming | — |
| Nov 23, 2026 | Dispositive motions due, Disney, Universal and WBD v. Midjourney | Upcoming | Axis |
| Nov 30, 2026 | Runway Dev 20% Kling 4.0 promo ends | Upcoming | Runway Dev |
| Dec 1, 2026 | Hearing on Apple's motion to quash Suno's subpoena | Upcoming | Mubert |
| Dec 2, 2026 | EU AI Act marking deadline for systems already on the market | Upcoming | NHI Mgmt Group |
| Before end of 2026 | Midjourney V9 (creator report, unconfirmed) | Upcoming | Woollyfern |
9. References
This document draws on the following 103 sources, listed in the order they are first cited.
- Genie 4: A new generation of world simulation — deepmind.google
- fal on X: "Kling 4.0 is now publicly available on fal, with API access open to all customers. The latest model from @Kling_ai brings: • Up to 30 seconds of native video generation with audio • Up to 15 image, video, and element references • Precise control with up to 10 keyframes • Improved motio… / X — x.com
- Kling 4.0: Release Date, Features, 30s Video, 4K HDR & Keyframes — aiidelist.com
- Seedance 2.5 vs WAN 3.0 vs Kling 4.0: Specs, References and Price Compared | WaveSpeed Blog — wavespeed.ai
- Hi3D V1.0 Goes Open Source: The First Production-Validated Open-Source AI 3D Model — prnewswire.com
- Midjourney Update | October 2026: V9, THINKING Mode, Draft Upscales & Shared Folders — youtube.com
- Best AI Video Generators: Avoid 4 Costly Budget Mistakes — aiforesight360.com
- Gemini API changelog - Google AI for Developers — ai.google.dev
- Veo 3.1 | Gemini Enterprise Agent Platform | Google Cloud Documentation — docs.cloud.google.com
- Best AI video models (2026) | 10 reviewed and ranked — reviewramble.com
- AA-Video-T2V v2.1 Leaderboard - Top AI Video Models | Artificial Analysis — artificialanalysis.ai
- API Pricing & Costs | Runway Dev — docs.dev.runwayml.com
- AI Video, Image and Audio API Pricing | Runway Dev — dev.runwayml.com
- Elvis Rights Holders Sue Anthropic Over 274 Songs | CellCog — cellcog.ai
- Suno's Legal Situation Just Got WORSE — youtube.com
- Suno Apple Subpoena Fight: What We Know (Oct 2026) — mubert.com
- AI Model Provenance and Watermarking Guide — nhimg.org
- Higgsfield Extends Series B With New Investment From TCV and NVIDIA — finance.yahoo.com
- Instagram — instagram.com
- Nvidia, TCV join Higgsfield Series B extension — marketscale.com
- Utopai Studios hits $1B valuation with $60M raise as video AI heats up | Dealroom News — dealroom.co
- Product Updates & Changelog | Runway AI — runwayml.com
- Hi3D V1.0 Open Source: What the Repo Proves and Doesn't — orcarouter.ai
- Eleven Labs Release Notes - October 2026 Latest Updates - Releasebot — releasebot.io
- SelfyzAI Among First to Integrate Kling 4.0 API: New Model Available in Workspace - Kitsap Sun — kitsapsun.com
- Seedance 2.5 vs Kling 4.0: We Ran 10 Head-to-Head Tests | CellCog — cellcog.ai
- Kling 4.0 vs Seedance 2.5: Specs, Pricing, Verdict — kie.ai
- Kling 4.0 API Pricing & Cost: From $0.126 per second (2026) — unifically.com
- MiniMax H3 vs Seedance 2.5: Which Model Is Better? | Pixo — pixo.video
- Best Video Models 2026: HeyGen Video vs Veo, Seedance, Runway | HeyGen — heygen.com
- Is Wan 3.0 Free? Free Options and Real Costs (2026) | Kenerate AI — kenerateai.com
- AI models: video, image, voice, music | Versely — versely.studio
- Kling 4 API Now Live: $0.11/sec, 30s Video | Atlas Cloud — atlascloud.ai
- Midjourney Review: Can Its AI Videos Get Text Right? — unite.ai
- 지니 4(Genie 4): 월드 시뮬레이션의 새로운 세대 — blog.google
- Best AI Video Generators: 3 Tested, Veo Wins 2026 — shattered.io
- AI Video Model Compare: Seedance, Wan, Veo, Kling — Credits & Quality | SunoMV — suno.bi
- Introducing Canva Craft — canva.com
- Canva Research — AI That Empowers the World to Design — canva.com
- Introducing Canva Brush, a vision-language model for design — canva.com
- Best AI Image Generator 2026: GPT Image 2.5 vs 3 Rivals — shattered.io
- ElevenLabs and Mistral Make Voices Speak Any Language - Voice Over Talent Industry News — voiceoverherald.com
- Voxtral 3: Speech Models, Prices and Preview Limits — digitalapplied.com
- Mistral Voxtral 3: Speech AI Hits $0.0035/Min 2026 — shattered.io
- What Is Voxtral 3? Mistral Speech Models and Voice Agents — techjacksolutions.com
- Suno Alternatives After V6: 13 AI Music Generators Compared (2026) — Undetectr — undetectr.com
- Best AI Music Generator in 2026: 6 Tools Compared — techiehub.blog
- Hi3D V1.0 Goes Open Source: The First Production-Validated Open-Source AI 3D Model — prnewswire.com
- Gemini API. A "**deprecation - Google AI for Developers — ai.google.dev
- Gemini deprecations — ai.google.dev
- OpenAI Videos API / Sora 2 removed from API shutdown date · Dead Canary — deadcanary.io
- Sora 2 API Shutdown: Self-Hosted Video Alternatives (2026) | Spheron — spheron.network
- OpenAI: Sora and Video Modules Deprecation by September 24, 2026 - Help Center — help.make.com
- 15 Best Veo 3 Alternatives With Native Audio (2026) | HeyGen — heygen.com
- Hi3D V1.0 Goes Open Source: The First Production-Validated Open-Source AI 3D Model — prnewswire.com
- LICENSE · OzzyGT/hi3d_v1 at main — huggingface.co
- LTX-2.5 Guide (2026): Specs, VRAM, Price, How to Run Lightricks' Open Video Model — vivimoai.com
- Hi3D 1.0: a production-validated image-to-3D model, now open source - Thaki Cloud Tech Blog ThakiCloud 다키클라우드 기술 블로그 — thakicloud.com
- Local AI Video Is Here LTX 2.5 and It's Free — youtube.com
- Video Generation Benchmarking Methodology | Artificial Analysis — artificialanalysis.ai
- AA-Image-T2I v2.1 Leaderboard - Top AI Image Models | Artificial Analysis — artificialanalysis.ai
- AA-Video-I2V v2.1 Leaderboard - Top AI Video Models | Artificial Analysis — artificialanalysis.ai
- AA-Video-Editing v1.1 Leaderboard - Top AI Video Models | Artificial Analysis — artificialanalysis.ai
- AA-Image-Editing v2.1 Leaderboard - Top AI Image Models | Artificial Analysis — artificialanalysis.ai
- AA-Image-T2I v2.1 Open Weights Leaderboard - Top AI Image Models | Artificial Analysis — artificialanalysis.ai
- Text to Speech Leaderboard - Top AI Speech Models | Artificial Analysis — artificialanalysis.ai
- Wan 3 vs. Seedance 2.5: How Do These AI Video Generators Compare? — fal.ai
- Seedance 2.5 API pricing vs fal.ai, WaveSpeed, Atlas Cloud & more – Apiframe — apiframe.ai
- How do credits work? – Runway — help.runwayml.com
- AI Image and Video Pricing from $12/month | Runway AI — runwayml.com
- Help understanding which models are unlimited. | Community — community.adobe.com
- Adobe Firefly improves AI video creation with new tools, new models and unlimited generations | Adobe Blog — blog.adobe.com
- Generative credits FAQ | Creative Cloud — helpx.adobe.com
- Partner models in Adobe products | Creative Cloud — helpx.adobe.com
- Best AI Music Generators September 2026: Suno, Udio, ElevenLabs, Stable Audio — buildmvpfast.com
- Creating with Gen-4.5 – Runway — help.runwayml.com
- Idea: Reassessing Generative Credit Costs for Firefly Video to Drive Adoption | Community — community.adobe.com
- Anthropic Faces $41 Million Copyright Lawsuit Over Elvis Presley Songs | Law Commentary — lawcommentary.com
- Elvis Presley's rights holders sue Anthropic saying it trained AI on his music — nbcnewyork.com
- Generative AI – IP cases and policy tracker | Mishcon de Reya — mishcon.com
- AI music and the major labels: who has settled, who is suing again | Boddominator — boddominator.com
- AI Copyright Lawsuits Tracker 2026: Live Case Status — axis-intelligence.com
- Elvis Presley Catalog Owners Sue AI Company Anthropic for Copyright Infringement — pitchfork.com
- AI Music Licensing in 2026: The Indie Artist's Guide — pophits.co
- AI labels are being applied inconsistently to my fictional anime Shorts - YouTube Community — support.google.com
- Are AI Videos Monetized on YouTube? 2026 Rules | Kenerate AI — kenerateai.com
- Use AI content label settings and disclosures - Google Ads Help — support.google.com
- Use the AI content label settings and disclosures in Merchant Center - Google Merchant Center Help — support.google.com
- Developer Program Policy - Play Console Help — support.google.com
- Do AI-Generated Videos Get Monetized in 2026? - Fliki — fliki.ai
- How Gemini Detects AI Media with SynthID | HyprForge — hyprforge.com
- Content Credentials and C2PA for AI video: practical guide — magichour.ai
- AI label in error - YouTube Community — support.google.com
- The Deepfake Watchlist: Week of October 16–22, 2026 — resemble.ai
- Higgsfield AI Alternatives: 10 Best Tools Compared (2026) — creatorflow.so
- [Distribution Show on Instagram: "Utopai Studios has raised $60 million at a $1 billion valuation, becoming the latest AI video company to reach unicorn status. But instead of just selling AI video tools, the Mountain View-based startup is building its own film studio and plans to bring AI-assisted movies to theaters.
Founded in 2022 by Cecilia Shen, a former Google X AI researcher, Utopai has more than 20 projects in development, including two TV series and three feature films planned for theatrical release in 2027. The company is targeting around 3,000 theaters across North America.
Utopai says its AI tools can reduce the costs of scene planning and CGI by 80% to 90%. It still works with human directors and actors, while using AI to make production less expensive. The company also licenses its technology to other filmmakers.
Rather than building everything from scratch, Utopai has customized the open-source MiniMax H3 video model to develop its own technology. Its strategy is to use AI in actual productions while improving the tools through the filmmaking process.
The funding comes as investors pour money into AI video companies. Higgsfield reached a $5.4 billion valuation, while Kling AI reportedly raised nearly $3 billion at a $15 billion valuation. Utopai has also received term sheets for a potential future round valuing it at $2 billion to $3 billion, though that deal has not been completed.
Source: Dealroom, The Information
Follow @distributionshow for more!
#UtopaiStudios #ArtificialIntelligence #AIFilmmaking #EntertainmentTech #AIStartups"](https://www.instagram.com/p/De4aRpHPGwL) — instagram.com 97. AI Video Statistics 2026: Key Numbers, Data & Facts — gradually.ai 98. Nvidia, TCV Back Higgsfield in AI Video Push — briefs.co 99. Higgsfield Adds Nvidia, TCV Funding for AI Video | TMC Insight — insight.tmcnet.com 100. Entrelligence — facebook.com 101. Wall St Engine on X: "$NVDA INVESTS IN AI VIDEO STARTUP HIGGSFIELD Nvidia and TCV have joined an extension of Higgsfield's Series B funding round, following its $400 million raise at a $5.4 billion valuation in August. The AI video and image generation startup, founded by former Snap executive Al… / X — x.com 102. TikTok Creator Rewards: Pay per 1,000 Views in 2026 — mubert.com 103. AI Video Generator/Editor Market: US, Japan, Korea, Europe - Strategic Insights & Forecasts 2026-2034 — linkedin.com