The window moved on access, price and rules more than on raw capability
Five conclusions from Nov 3 to Nov 16, 2026, then the dated change log. Rows after Nov 16 are upcoming.
Five conclusions
At the production tier the contest is length and price, and the research frontier moved to real-time worlds
Gemini Omni 1.2 Flash became generally available on Nov 5 with native clips of up to 15 seconds (from 10) and a 30% price cut to $0.07 per second of 720p video. Kling AI said on Nov 4 that Kling 4.0 Flash was open to everyone on its website and API, and World Labs streamed its Atlas world model at up to 15 frames per second as a research preview. OpenAI's Images 2.6 (Nov 4) is 50% faster and adds GIFs; Midjourney V9 did not ship.
For builders, the calendar matters more than the model list
Gemini API requests to gemini-omni-1.1-flash will be rerouted to Omni 1.2 Flash on Dec 3, 2026. Vertex veo-3.1-generate-001 lists a retirement date of "November 17, 2026 or later", a floor, not an announced shutdown. The one in-window open-weight release, MiniMax's H3-Image, carries a licence that excludes the US, EU, UK and South Korea.
Quality is bunched, so price separates the leaders
On Artificial Analysis' text-to-video board the top five sit within 23 Elo points (1175 to 1152) while normalized prices run from $4.80 to $34.12 per minute. GPT Image 2.6 Sunburst leads the next image model by 31 points (1259 against 1228). Even one model varies by seller: Kling 4.0 Flash is $0.094 per second at Runway and "from $0.1174" at Venice.
Rights holders keep litigating; the window's cases are allegations against one music generator
On Nov 6 the Elvis Presley rights holders and Eight Mile Style (Eminem's publisher) each sued Suno in the Northern District of California, pleading just over $202 million and about $47.7 million; Suno denies both. Suno's Nov 3 motion to dismiss a separate class action has a response due Dec 1. No platform-labelling change dated Nov 3 to Nov 16 appears in the material.
Money and distribution concentrate on the layer that owns the creator's workflow
Sand.ai, a Beijing video-model startup, raised $200 million at a reported valuation above $1 billion. At Adobe MAX (Nov 10) Premiere's Generative Media tool offers models from Google, Kling, Runway and Luma, and ElevenMusic output can be published directly to SoundCloud.
The window in one line
Distribution surfaces, not single models, are where creators now choose.
A 15-second clip at $0.07 per second, a dated Google migration calendar, and two lawsuits that test voice and likeness law.
- Flagship video releases in the window: 1 (Gemini Omni 1.2 Flash)
- Major image releases: 1 (OpenAI Images 2.6)
- Open-weight releases: 1 (MiniMax H3-Image)
- Court rulings on training data: none
- Platform rule changes dated in the window: none in the material
What changed at a glance
veo-3.1-generate-001Status as given in the sources. Where two sources give different dates, both are shown.
One flagship video release, one cheaper sibling and one major image release
Specs are vendor claims unless stated. Orange-edged cards are dated inside Nov 3 to Nov 16; grey cards are context from earlier weeks.
- Access
- Gemini API, Google Cloud, Gemini for business, Google Vids
- Max clip
- 15 s (3 to 15 s in 1 s steps)
- Resolution
- Drafts at 360p or 720p, upscale to 1080p or 4K
- Audio
- Dialogue, lip-sync, multi-speaker; audio references allowlisted
- Access
- Kling website, API, MCP, CLI, plug-ins
- Max clip
- 20 s (3 to 20 s)
- Resolution
- 720p
- Audio
- Always generated
- Access
- Browser demo
- Output
- Streamed frames, not clips
- Speed
- Up to 15 frames per second
- Audio
- Not applicable
- Access
- App; API via Runway Dev and others
- Max clip
- 30 s
- Resolution
- 720p, 1080p
- Audio
- Stereo
- Access
- Hosted by third parties, for example OCMaker
- Max clip
- 5 to 15 s on that host
- Resolution
- 480p, 720p, 2K on that host
- Audio
- Yes
What is genuinely new in Omni 1.2 Flash
- Five changes from Omni 1.1 (Google's post): clips up to 15 seconds (from 10), price 30% lower, stronger adherence to reference images and short motion clips of up to three seconds, multi-turn conversational editing, more natural dialogue with clearer lip-sync.
- Production terms (Google Cloud): "production-ready" with Provisioned Throughput and an imperceptible SynthID watermark.
- Cost: $0.07 per second at 720p means a 15-second clip costs $1.05 (15 x $0.07), before any upscale to 1080p or 4K.
- Caveat: named launch customers (Artlist, FLORA, invideo, Krea, Picsart, Photoroom and Runway) are vendor-quoted testimonials, not independent tests.
Kling 4.0 Flash and the world-model contrast
- A cheaper variant, not a new generation: the Venice API docs list it as released Nov 5, 720p only, 3 to 20 seconds, from $0.1174 per second (a 10-second clip is $1.17). Kling's own post is dated Nov 4 and also lists plug-ins for Photoshop, Premiere Pro and After Effects.
- Dates: the one-day gap is probably a time-zone effect, but the material does not say; the vendor's Nov 4 post is followed for the announcement, the host's Nov 5 for the API listing.
- Atlas is not a clip generator: it is an autoregressive world model with camera control and a developer API, Blender integration and an MCP server; there is no clip length to compare, so the two categories should not share a ranking.
- Retention risk: The Information reported on Nov 13 that about 60 people, more than half of the Grok Imagine team, were let go; the product stays online.
- Access
- ChatGPT, Codex, API
- Licence
- Closed
- Capability
- 50% faster; Flare "as little as six seconds" at low quality; GIFs; SVG export; image markup in the API
- Access
- Hailuo AI, MiniMax API, Hugging Face weights
- Licence
- MiniMax's own licence (see Access)
- Capability
- 4096 x 4096 (about 16.8 million pixels) in 10 to 20 seconds, up to 16 reference images; vendor claim
- Access
- Hosted
- Licence
- Not stated
- Capability
- More natural skin, sharper text, transparent backgrounds
- Access
- Not available
- Licence
- Closed
- Capability
- Native full-resolution base model plus new upscaler, in internal testing
OpenAI's own numbers
- Volume: OpenAI says people now create 4 billion images per week across ChatGPT, Codex and the API, "more than half a billion images created per day".
- Safety: the system card dated 2026-11-04 describes the safety stack as unchanged from Images 2.5 plus extra safeguards for GIF creation.
- Cost complaint: a developer thread on the OpenAI community forum says low and medium quality settings bill substantially more output tokens than gpt-image-2.5; that is a user claim, not OpenAI's statement.
Midjourney V9 did not ship
- Status: secondary summaries of the Nov 4 office hours say a functional full-resolution base model exists internally and a release "within approximately 30 days of November 4" was the target.
- Price: Midjourney said V9 images "could cost double".
- Roadmap only: native video with sound, real-time generation and 3D camera controls are roadmap items, not shipped features; nothing shows a release by Nov 16.
- What changed
- Nemotron 3.5 Super VL now in ElevenAgents; 15MW of NVIDIA Blackwell Ultra and Vera Rubin compute added
- Example
- Sovereign voice agent for Poland's national e-Health platform
- What changed
- Experimental meeting-notes and dictation app launched at the company's 11/11 Summit
- Access
- Preview; early-access sign-up
- What changed
- ElevenMusic users publish directly to SoundCloud; SoundCloud Artist Pro users get three months of ElevenMusic Pro
- Type
- Distribution, not a new model
- What changed
- Licensed music clips from two major labels, UMG and WMG, in Adobe's free Premiere mobile app
- Type
- Licensed catalogue
- Window
- No dated 3D model release in the material
- Context
- A secondary guide lists Microsoft's TRELLIS.2 (4B image-to-3D, MIT licence, NVIDIA GPU with at least 24GB VRAM) as the main self-hosted option
- Window
- No new text-to-speech model release and no new Suno or Udio model release dated inside the window is supported by the material
- Note
- The Suno news in this window is legal
What the music rows have in common
Music generation is moving into distribution and editing surfaces that already have audiences, with licensed catalogue (UMG, WMG) on one side and generated catalogue on the other. TechCrunch notes that ElevenLabs says users have created over 25 million tracks with its models (a company claim) and that streaming platforms "seem more inclined toward labeling and helping users filter out AI-generated music".
Adobe MAX, Nov 10: creative tools with partner models
Premiere's new Generative Media tool offers Generate Video, Generate Sound Effects, Generate Soundscape and Generate Music in the timeline, and the user can choose Adobe Firefly or partner models from Google, Kling, Runway and Luma.
Adobe and Google also announced a wider partnership: Google's latest imaging and video models will come to Firefly, Photoshop, Premiere and Express, and Adobe and Google DeepMind will collaborate on image editing starting with Photoshop (a statement of intent, no date given).
What builders can still call: a Google calendar, hosts and open licences
What is switched off, what is rerouted and which weights can be downloaded. Status is compared with the run date, Nov 17, 2026.
Deprecation and retirement calendar
veo-3.1-generate-preview, -fast-, -lite-) in the Gemini APIveo-3.1-generate-001, veo-3.1-fast-generate-001 (Gemini Enterprise Agent Platform)gemini-omni-1.1-flash (Gemini API)"Veo 3.1" now means different things on the two surfaces: the Vertex footnote says models "may remain accessible through the Gemini API" after the listed dates, while the Gemini API's own page says its preview models are already shut down. "November 17, 2026 or later" is not a promise of removal that day. The collected OpenAI changelog and deprecations pages show no image-model retirement dated in the window.

Two surfaces, two clocks
The Gemini API tells developers to use Gemini Omni Flash as the default video model and keeps Veo as the second option. On Google Cloud, Veo 3.1 GA models are still listed, with a retirement floor of Nov 17, 2026.
The API picture behind the 15-second headline
Source: Gemini API release notes. Kling 4.0 Flash, World Labs Atlas and OpenAI's Images 2.6 are not gated beyond ordinary account sign-up in the material, except Google's allowlisting of audio references.
The practical split on open weights
According to MarkTechPost's Nov 13 ranking, which uses Artificial Analysis Elo, the top-ranked text-to-video model (Wan 3.0) is closed, while the open-weight MiniMax H3 sits in the top three.
Open weights and licences
All licence descriptions come from secondary sources that read the licence files; read the licence text before use. No 3D open-weight release dated inside the window appears in the material.
What to do before the next date
Tap an item to tick it off.
Quality is bunched, so price separates the leaders
One source per chart. Artificial Analysis (AA) ranks by Elo from blind pairwise votes; none of its pages shows an as-of date, so every rating is "date not stated on the page; crawled for this issue".
Switchable comparison
How to read the video board
- Omni 1.2 Flash enters at rank 5 with 5,354 votes, the fewest of the top five, and 25 Elo above its predecessor (1152 against 1127). Its rank range of 3 to 9 means the board cannot separate it from FLUX 3 Max or HeyGen Video 1.
- Video editing board: Omni 1.2 Flash is second at 1164, behind Wan 3.0 at 1192.
- Secondary ranking: according to MarkTechPost (Nov 13, built on AA), Seedance 2.5 leads image to video at 1161 and Omni 1.2 Flash scores 1086 there.
- Silent board: AA's text-to-video board scores clips with audio; its silent board has a different leader, HeyGen Video 1 at 1146, with Wan 3.0 fifth at 1136.
How to read the image boards
- Images 2.6 beats 2.5 Sunburst by 32 points on text to image (1259 against 1227), outside the 1244-1274 interval, but by only 9 on editing (1210 against 1201), where all three leaders share ranks 1 to 4.
- Editing board: GPT Image 2.6 Sunburst 1210, Flare 1206, GPT Image 2.5 Sunburst 1201, Nano Banana 2.1 1152, Grok Imagine Image 2.0 1117.
- MiniMax and ByteDance: H3-Image is 13th on editing at 1138, and Seedream 5.5 is 7th at 1159.
- Cost per image: AA lists $210.72 per 1,000 for all four OpenAI max variants against $33.60 for Nano Banana 2.1.
Price per second, Runway Dev list
One reseller listing many makers (1 credit = $0.01); no as-of date, so "crawled for this issue". The Kling 4.0 promotion runs through Nov 30, 2026.
Audio is also a price lever: Veo 3.1 costs $0.40 per second with audio and $0.20 without; Kling 3.0 Pro is $0.17 with and $0.12 without, 42% more. Google's own figure agrees with its resale: $0.07 per second at 720p from Nov 5.
Same model, different sellers
Clip length and audio
Vendor or reseller statements, not test results.
What the comparison does not show
Elo measures blind preference on short prompts, not clip length, consistency or editing control, and a 3-to-9 rank range means ties. AA prices a minute at 1080p, while Google's headline is 720p, so $0.07 and $0.1043 per second are different products. Credit and token billing hide rounding, minimums and failed generations. No creator head-to-heads or arena.ai ratings were collected, so editing workflow, prompt adherence in a real project and brand-safe output remain untested here.
Rights holders keep litigating and no platform rule changed in the window
The legal side is dominated by one defendant, Suno, sued twice on one day. Platform-rule material contains no new rule dated inside the window.
Presley rights holders v. Suno
- Plaintiffs
- Elvis Presley Enterprises, ABG EPE IP (both controlled by Authentic Brands Group), Raleigh Music Publishing, Abilene Music and others
- Claims
- Copyright infringement over 859 compositions; Tennessee ELVIS Act; right of publicity; circumvention of access controls; removal of credits
- Pleaded
- Just over $202 million: $128.85 million statutory plus $2.1475 million and $21.475 million, plus "believed to exceed $50 million" for voice and likeness
- Status
- Filed Nov 6; Suno denies; no ruling
Eight Mile Style v. Suno
- Plaintiffs
- Eight Mile Style and Martin Affiliated (Eminem is not a plaintiff)
- Claims
- Copyright infringement over 269 compositions; altered or removed copyright metadata
- Pleaded
- Just over $47.7 million in total; $40.35 million is the statutory component (269 x $150,000)
- Status
- Filed Nov 6; Suno denies; no ruling
Presley complaint: what was pleaded, $ million
Allegations and ceilings the plaintiffs ask for, not awards. Components as reported by MBW; they sum to 202.47, "just over $202 million".
How the numbers are built
Procedure and defence
- Suno's position: it does not "allow prompts for specific artists' names and songs" and screens uploads with third-party providers.
- Nov 3 (window): Suno asked a Boston judge to dismiss a proposed class action by Jason Isbell and three other musicians; the plaintiffs' response is due Dec 1 (upcoming).
- Context, not window events: the Presley suit against Anthropic (October), Eight Mile Style v. Meta (filed May 2025, allowed to proceed to discovery) and earlier Suno suits by Round Hill, Reservoir, UMG and Sony.
- Unverified: a suit by the American Federation of Musicians over the Universal and Warner licences with Suno and Udio is described only in an Instagram post; court, filing date and status are not in the material.
Deals and training-data policy
- Adobe and Disney announced a multi-year alliance making Adobe Disney's "Official AI Innovation Partner", placed at Adobe MAX. Firefly Foundry builds private models "customized using Disney intellectual property"; Imagineers are only "experimenting" with a sketch-to-image model. No value, length or exclusivity was disclosed.
- Suno (context): Warner Music Group settled last year and licensed Suno, which also has deals with BMG and Believe; a label-backed model released in September drew a further UMG and Sony suit. Suno raised $400 million in June at a $5.4 billion valuation, per the Los Angeles Times.
- Training: no court ruling on training-data liability appears in the window.
Platform labelling, disclosure and provenance
No platform policy change dated Nov 3 to Nov 16 appears in the material. The platforms' own policy pages were not collected, so everything here is "according to" commentators.
What is still unresolved: whether label licences reach session musicians is untested in the material, and on platforms the gap is enforcement, because a platform may not label what it cannot read. This is description, not legal advice.
Money and distribution concentrate on the layer that owns the creator's workflow
One in-window round, and it is Chinese and open-source; the rest of the commercial news is distribution.
Funding and valuations
The Sand.ai figures come from an investor's social post and a sales-intelligence page, not the company or a filing; Sand.ai itself is quoted saying VidMuse passed $10 million in annual recurring revenue within two months of launch (a company claim). Wangsu also disclosed a planned 300 million yuan investment; currencies are not converted. Where two sources disagree on fal's round, neither is followed.
Valuation tracker, $ billion
New Market Pitch tracker, updated Nov 14. It separates 24 valuations backed by statements or round terms from 52 estimates; the values here are estimates, not company statements.
What the money signals

The model maker sells through someone else's surface
The window's distribution moves share one pattern. Adobe acts as an aggregator, Runway Dev resells the official Kling 4.0 API next to its own models, and music reaches listeners through SoundCloud and licensed catalogue in Premiere mobile.
Distribution moves
- Adobe as an aggregator. At Adobe MAX (Nov 10) Premiere's Generative Media tool lets editors choose Firefly or partner models from Google, Kling, Runway and Luma; Google's latest imaging and video models are due in Firefly, Photoshop, Premiere and Express.
- Developer platforms reselling video models. Runway Dev sells the official Kling 4.0 API self-serve, alongside its own models and a Model Router that picks a model by cost, quality or latency; LiteLLM lists Kling 4 on Runway at $0.1256 per second as of Nov 13.
- Google's customer list as distribution. Artlist, FLORA, invideo, Krea, Picsart, Photoroom and Runway are named as builders on Omni 1.2 Flash (vendor-quoted testimonials).
- Music distribution. ElevenMusic output can be published directly to SoundCloud, and Premiere mobile adds licensed UMG and WMG music.
Creator economics
- Plan prices are third-party lists captured in Oct 2026, before the window, so they are directional.
- The $1 million sits on top of nearly $10 million committed in grants and donated products for 2026 (Film & TV Fund).
- A Nov 13 social post says Disney plans a preschool series made with agentic AI tools in December, with artists reviewing every frame; it is not verified.
Price dispersion is the strongest signal and key dates decide what to do next
Capability news was concentrated in two releases, Gemini Omni 1.2 Flash and OpenAI Images 2.6; the larger story is models reaching creators through other companies' surfaces, at prices that differ by seller.
Trends and how strong the evidence is
The strongest signal is price dispersion: the same Kling 4.0 Flash costs $0.094 or $0.1174 per second depending on the seller, and the top-ranked model is not the cheapest. The weakest is the world-model story, which rests on one vendor demo.
What it means by reader
Plans differ by seller: Runway plans work out at $0.096 to $0.230 per Gen-4.5 second; Omni 1.2 Flash is $0.07 per second at 720p.
Compare cost per second, not plan price; keep prompts and references outside any one model.
Runway plan arithmetic, Google's Nov 5 announcement and the Grok Imagine report.
Adobe puts several makers' models in one timeline; the Adobe-Disney alliance uses private models on Disney IP.
Pick a surface that lets you switch models; ask vendors for licence and provenance terms in writing.
Adobe MAX announcements and the Adobe-Disney alliance.
Omni 1.1 reroutes to 1.2 on Dec 3; the Vertex Veo 3.1 retirement floor is Nov 17.
Pin model IDs, test the new price, and read the lifecycle page before cutover.
Gemini API release notes and Google Cloud docs.
Two Suno complaints (Nov 6) test the ELVIS Act and copyright metadata claims; a response is due Dec 1 in the class action.
Dated procedural steps exist, but no ruling on training is scheduled in the material.
MBW's report on the complaints.
These are readings of the evidence, not legal or investment advice.
Key dates
Dates up to Nov 17, 2026 are in effect or carry the matching status word; later ones are upcoming.
veo-3.1-generate-001 retirement date "or later"; MiniMax H3-Image half-price 4K promotion ends (per CellCog)gemini-omni-1.1-flash requests route to Omni 1.2 FlashBottom line
Models ship fast; the dates, prices and licences around them decide what creators can use.
Capability news was concentrated in two releases, while the larger story is models reaching creators through other companies' surfaces, at prices that differ by seller, under licences and lawsuits that decide where they may be used.
