Mad Brain Games went from 14 people to 6, with only one artist left. With deadlines for the next batch of games pressing, vice president of creative Maor Sabag took on all art production, and the team had no dedicated animator. A customer story page[1] records this division of labor—not an industry sample, just one studio’s situation at the time.
His tasks were very concrete: Find the Dogs needed several thousand distinct dogs. Maor trained a model on a small number of initial dog illustrations and then generated them in batches. Backgrounds were even more direct: he used Scenario’s API to make requests in batches, generating hundreds at a time, ultimately serving more than 2,000 levels, then manually selecting and upscaling them to 4K. With no animator, marketing trailers and in-game animation were also handed over to video generation. The page says that “concept to final trailer,” which used to take a full creative team two weeks, now takes a day; an entire animation sequence can also be completed within a day.
These durations are the customer story page’s own account, with no independent timing. But the problem it points to is real: what small teams lack is not inspiration, but the person willing to draw the same style two thousand times over.
Scenario chose this position from the start. In 2021, Emmanuel de Maistre, who had spent several years in the 3D modeling industry, and Hervé Nivon, who had a data science background, co-founded the company. In 2023, de Maistre told TechCrunch[2] that they were inspired by DALL-E 2 but thought the raw output was not consistent enough, so they chose to build game art tools that let users train generators on their own data and keep consistency in their own hands. The terms require users to use only data they own, are authorized to use, or that is open source; the company does not claim ownership of models or images generated by customers; he advised consulting IP professionals for commercial projects, with compliance responsibility resting with users.
This trade-off is not hard to understand. What makes art valuable lies in “consistency”: the same character from ten angles, the same set of environments spread across one hundred rooms. A general-purpose model struggles to provide this consistently; give it a batch of one’s own images, and only then might it learn.
The product form grew out of this as well. At Mad Brain, Maor was the one person training models and calling the API; at Future Run, creative director Sami Timonen built a tool himself that wrote prompts and made requests one at a time for him. The official website[3] says that apps, agents, and a team’s own pipeline share the same models, workflows, and assets: a workflow can string models and tools together as nodes on the same canvas and be published as an app; other people on the team can open a simple form and run it, seeing how much this run will cost in CU before running; hand a brief to an agent, and it can also be turned into a node graph and run. Building and using are thus separated—one person connects the nodes, and others open a form and run it. The same things can be used in the browser or connected to a team’s own pipeline through the API.
Image: source Scenario actual image source[4], retrieved 2026-11-10, original interface retained.
CU is the unit of pricing here, and also one of the most candid parts of this company. According to the usage documentation[5], almost every generation, training run, and tool call deducts CU based on model, resolution, steps, batch, and similar factors; the button shows how much this run will cost in CU before generation, so you do not have to wait for the bill.
The pricing page[6] sets the thresholds this way: Free offers 50 CU per day, one seat, and one project; free output is limited to personal and evaluation use; upgrading does not retroactively turn old outputs into commercial licenses, and to use them you must regenerate during a paid period. Paid plans start at Starter for $15 per month; the API is available on Starter; custom model training and Canvas require Pro; multi-user workspaces start at Max and cost an additional $30 per seat per month.
Worth singling out is the team layer: CU is a shared pool, but seats are individual licenses and cannot be shared; caps set by member, project, and API key only restrict how the same pool is used and do not add quota; only Enterprise has this kind of granular cap. So an organization buys a shared CU pool, with one seat per person—“who is spending” and “who is clicking the button” can be separated. After quota is exhausted, Free and Starter stop; on the web, Pro and above can still wait, but API requests stop at the limit; the cancellation documentation[7] says that cancellation takes effect at the end of the current period, with no refund for that period, the organization reverts to Free, the API key stops working, and assets and models are retained for 30 days and then deleted; resubscribing within 30 days keeps them.
This billing logic explains one thing: what Scenario charges for is not “how much per image,” but how many times a team can try each month.
The Future Run story[8] lays out this “how many times you can try” most clearly. Creative director Sami Timonen built a tool himself, with Scenario as the generation layer: with rough drawings and reference images as input, the tool writes prompts and makes repeated requests, results go through Blender into Unity; 3D models are handed to models on the platform to generate, then another model reviews them one by one against the concept art, revising until they resemble the concept art, and after entering Unity, humanoid setup and rigging are done automatically. For an eighties high-school first-person horror prototype, he stayed up all night batch-generating environment props; another time, he gave the tool only a few reference images, and when he booted up the next day, the entire game had been changed to a gangster-themed Las Vegas look. He said the studio burns points quickly because of heavy use, but does not consider this a problem. He calls AI a double-edged sword: it may take away the joy of manual animation, but it also lets independent creators show bigger ideas to investors and studios.
Boundaries are visible here too. In terms of entry point, Rosebud[9] covers “generate and publish a playable web game from one sentence,” while Scenario covers team production pipelines and the API; the two are adjacent; there is no data on customer switching, price overlap, or revenue overlap, so it is not enough to say who replaces whom. Scenario’s entry point is not here: a person who does not want to learn a pipeline and just wants to immediately get a playable game wants a different kind of product.
Public sources do not include Scenario’s revenue, number of paying customers, or the dollar price per CU, so profit cannot be judged; those “two weeks to one day” claims can only be read as accounts from a customer and a vendor, and cannot be extrapolated into revenue.
What Scenario really sells is not a particular image, but turning the craft of “style consistency” into quota that can be shared, measured, expires, and must be renewed. Value and risk sit in the same pocket: budget, permissions, and cloud asset access all follow subscription status; the commercial rights to content generated during a paid period remain after cancellation.
[1] How Mad Brain Games Scales Art Production with AI Art Generation
[2] Scenario lands $6M for its AI platform that generates game art assets | TechCrunch
[3] Scenario | AI for Creative Production
[4] Scenario actual image source
[5] Usage limits and Compute Units (CU)
[6] Pricing & Plans
[7] Upgrade, Downgrade or Cancel your Plan
[8] How Future Run Automated Game Asset Production with Scenario