Model

Luma AI cut resold Kling and Seedance rates by about half, and Adobe put models in Premiere

The rate card on Luma's Model Marketplace fell by about half on Nov. 7, with the catch buried in the billing basis; three days later Adobe put model choice, Luma included, inside Premiere.

Vincent JiangVincent Jiangthree-minute read
Share
A production crew films a model on a white studio set, a camera operator on a tripod and assistants holding a large black panel
1 / 6Slide 1 of 6
A crew shoots on a studio set, the kind of footage Luma's marketplace now prices per second for resold AI video models.

On Nov. 7 the rate card that prices a second of AI video for Dream Machine subscribers quietly fell by about half, according to the pricing tracker PricingSaaS: ByteDance's Seedance 2.0 at 720p dropped from 64 to 32 credits a second, MiniMax's H3 from 56 to 28, Kuaishou's Kling 4.0 Omni at 1080p from 98 to 47 1. The same week Luma added two cheap Kling models, Flash and Flash Omni, at 32 credits a second 1. On the Plus plan a credit is about $0.003, so a 10-second Kling Flash clip now runs about $0.96 1.

The fine print halves the halving

The cut came with a meter change. Luma~$1.1B — Luma, private, latest valuation ~$1.1B now bills video-to-video on generated plus reference seconds, where the old card charged generated seconds alone 1. Feed in a short reference against a long clip and the halving is real money. Feed in a reference as long as the output and the two changes largely cancel 1.

Video-to-video rates on Luma's card fell by half in a week

Every resold model's edit rate fell by half in one week

  • Nov 2
  • Nov 9
0 credits/sec20 credits/sec40 credits/sec60 credits/sec80 credits/sec100 credits/secSeedance 2.0 720p32 credits/sec64 credits/sec-50%MiniMax H3 768p28 credits/sec56 credits/sec-50%Kling 4.0 Omni 1080p47 credits/sec98 credits/sec-52%
Data
Nov 2Nov 9Change
Seedance 2.0 720p64 credits/sec32 credits/sec-50.0%
MiniMax H3 768p56 credits/sec28 credits/sec-50.0%
Kling 4.0 Omni 1080p98 credits/sec47 credits/sec-52.0%
Video-to-video rates on Luma's card fell by half in a week1

The house model is now the expensive option

The part nobody wrote up is what the cut did to Luma's own shelf. Its live rate card lists the flagship Ray3.2 at 72 credits a second for video-to-video at 720p, against 32 for resold Kling 4.0 Omni and Seedance 2.0 and 21 for Kling Flash Omni 2. On the editing path, the cheapest model in Luma's neutral shop is now somebody else's.

On Luma's own card, the house model costs the most to edit

Luma's own flagship edits at more than twice the cheapest resold model

0 credits/sec20 credits/sec40 credits/sec60 credits/sec80 credits/secRay3.2 (Luma's own)72 credits/secKling 4.0 Omni32 credits/secSeedance 2.032 credits/secKling 4.0 Flash Omni21 credits/sec
Data
Value
Ray3.2 (Luma's own)72 credits/sec
Kling 4.0 Omni32 credits/sec
Seedance 2.032 credits/sec
Kling 4.0 Flash Omni21 credits/sec
On Luma's own card, the house model costs the most to edit2

Adobe+0.24% — Adobe, up 0.24 percent today is bidding for the same seconds from the other end. On Sep. 8 it embedded Firefly, Google Veo, Runway, Kling and Luma directly inside Premiere's timeline, with partner credit costs disclosed only in the taskbar and credit revenue flowing through Adobe whichever model runs 3. At MAX on Nov. 10 it added a Model Selector to Firefly and put MCP agent integration into private beta for early 2027 4. Adobe charges per generation; Luma charges per second.

Whose money the second feeds

Three balance sheets sit on this rate card. Luma is private, valued near $4 billion, with one in five employees a working creative embedded in client teams from Coca-Cola to Lufthansa 5; its compute anchor is AWS−1.20% — AWS, down 1.20 percent today, set when it announced Ray 2 on SageMaker HyperPod, Trainium and Inferentia 6. Kuaishou (1024.HK) holds two-thirds of Beijing Kling, which raised $2.98 billion on Oct 78. 30 and plans an IPO, after Kling revenue grew over 200 percent year over year 78. Adobe collects the credit either way 3.

What settles it is whether the storefronts that resell the same models match the cut, and whether Adobe ever publishes a partner-model rate card. Until then, the second of video is getting cheaper everywhere except the file where it is spent.

Get Model by email, free

The AI labs, their models, and the business behind them. In our daily email, every weekday.

Loading ADBE

Reader comments

0 comments

    Model

    OpenAI settles for $1.25 trillion, below the $1.405 trillion Anthropic asks public buyers Nov. 18

    The private ceiling for AI now sits below what Anthropic is asking public buyers to pay on November 18, and venture firms are already pushing founders to cut their marks. The two labs were on pace to absorb more than a third of all startup capital.

    Model

    Musk says Grok 4.8 trails Claude Opus 5.5; Anthropic will pay SpaceX $3.1 billion a month from December

    The Nov. 10 concession is a dated promise from inside SpaceX, whose biggest compute tenant, Anthropic, will pay about $3.1 billion a month from December. At 44 times forward earnings the stock is priced on that rent; Grok is the part not yet paid for.

    Model

    Google opens its strongest Gemini model at $2 per million tokens ahead of Anthropic's IPO pricing Nov. 18

    Google's strongest model ships at the previous model's introductory price, a day after Anthropic cut its flagship 20 percent ahead of an IPO banks may push past $1.5 trillion. The contested ground is what a frontier token costs, and Google just named its price.

    Model

    OpenAI says its models go dark in Cursor on Nov. 12, and SpaceX's $7.56 billion segment shows no split

    The OpenAI model contract inside the $60 billion Cursor acquisition lapses on November 12, and SpaceX's third-quarter report is the only public window into what that asset earns, folded into a $7.56 billion segment that discloses no split.

    Model

    Anthropic hired a UK lobby firm the week it priced its $1.4 trillion IPO

    The same week Anthropic set its $70-to-78 IPO range, the UK lobbyist registry showed it paying for political access in three countries. Goldman's $685 billion revenue case runs on data centers that voters and regulators must permit, and buyers fund that insurance without seeing its cost.

    Model

    Mistral built its enterprise stack in eight days and priced none of it

    Between October 29 and November 5 the Paris lab published its biggest model's weights, expanded its agent platform and bought a fifth startup, all funded by a September mark of €21 billion. Not one piece of the stack carries a published price, while free Chinese open-weight models still outscore it.

    Get the newsletter

    The day's AI and market news, in your inbox every weekday.