I set out to review ElevenLabs Flows and ran into something bigger: how ElevenLabs prices AI image and video generation, and a credit cliff steep enough to triple your bill between one image and the next.
Summary: ElevenLabs is a world-class audio company trying to sell a plain, half-baked image canvas by forcing you to rent a $299 monthly subscription. While their per-character cost is reasonable, the credit system hides a steep plan cliff. If a client project needs 500 Nano Banana Pro generations, the Pro plan ($99) empties, and you must pay $299 just to finish the job. For a $600 gig, spending half your invoice on a single tool's infrastructure is business suicide.
I do real estate visualization. It is the kind of work where a client needs a dozen finished images, and getting to those twelve means pressing generate 500-1000 times. On average, eight hundred. You generate, you look, you throw most of it out, you change three words in the prompt, you generate again. Anyone who does this work knows the ratio, and nobody is surprised by it.
So I tried to run one of those projects through ElevenLabs Flows, the node-based canvas they launched in March. The pitch is the one every node tool makes: build the pipeline once, run it forever, chain your image, video, and audio models into one place.
I never ran that whole project through Flows. I did test the canvas itself, all of it, which did not take long. Then I went looking at the price.
Why I never finished
Two things stopped me, and neither was the one I expected.
The first was the content filter. I generate pictures of living rooms. Kitchens, hallways, a sofa under a window. A run of those ordinary prompts came back flagged as policy violations. What makes that strange is that the same prompts, feeding the same model, Nano Banana Pro, generate without a complaint on every other platform I use. The same words and the same model, and only ElevenLabs called it a violation. That points at ElevenLabs' own monitoring rather than anything Google built into the model. I never found out what the filter thought a kitchen was.
The second was simpler. I am on the Creator plan, twenty-two dollars a month for 121,000 credits. For Nano Banana Pro that is sixty-six images at 4K, ninety-nine at 1K. Either way, fewer than a hundred. My projects need five hundred to one thousand, so the credits were gone almost before I started.
So I stopped trying to force a real project through it. I run Scopeful, where I track what every creative AI tool actually costs, so I did that instead. I worked out what it would take to run one full project through Flows. That is where the cliff is.
The cliff
Flows runs on ElevenLabs credits, the same credits behind their text-to-speech. That matters more than it sounds. It means almost nothing below is specific to Flows. The credit price of an image is an ElevenLabs price, the same whether you generate it inside Flows or anywhere else they let you generate. Nano Banana Pro, Google's top image model, costs 1,818 credits per image at 4K. ElevenLabs' Pro plan is ninety-nine dollars a month for 600,000 credits. Divide one number into the other and you get 330. Not "around 330." Exactly 330. The Pro plan makes 330 images at 4K, and then it is empty.
Image 331 does not cost a little more than image 330. Image 331 is on a different plan. To generate it you move up to the Scale tier, and ninety-nine dollars a month becomes two hundred and ninety-nine.
What a real project would cost
To finish a project in Flows, not test it, finish it, I would need a plan that holds five hundred to one thousand generations. The Pro plan does not, at any resolution. The cheapest tier that survives a single project is Scale, two hundred and ninety-nine dollars a month. That is the cost of walking in the door, before I have one image worth showing a client.
Here is the same project bought another way. Those five hundred-plus generations, run through Google's own API for the very same Nano Banana Pro model, cost somewhere between seventy and a hundred and ninety dollars, depending on how much of the work you do at full 4K. Google sells you its own model by the image, and you pay for what you generate. ElevenLabs makes you rent the tier.
Two hundred and ninety-nine dollars on one side. Under two hundred on the other. That looks like a markup. It is not one, quite, and the difference is the actual story. Or, if you want a browser interface instead of raw code, you can run the exact same project on Magnific's $40 subscription. Same model, same images, same job.
The strange part
Ninety-nine dollars buys 600,000 credits, which makes one credit worth about a sixtieth of a cent. You spend 1,818 of them on a single 4K image without blinking, because who stops to count something that small. Run the conversion anyway and that image costs you almost exactly thirty cents. Google charges twenty-four cents for the same image through its API. Per picture, ElevenLabs is barely above the rate of the company that built the model.
So they are not overcharging you per image. Hold onto that, because it is the whole point.
The three hundred dollars does not come from the price of an image. It comes from the fact that you cannot buy an image. You buy a plan. Five hundred generations at 1K is 609,000 credits, and the Pro plan holds 600,000. You miss the cheaper tier by nine thousand credits. That is about seven images. Seven images over, and the bill triples.
The credit system is the product
This is the part I keep coming back to.
If ElevenLabs put a dollar sign on the generate button, you would feel this. "Thirty cents," sitting next to a button you are about to press four hundred times in an afternoon, would make you stop and do arithmetic. You would see the cliff at image 330 coming, because you would watch the running total climb toward it.
"1,818 credits" does none of that. Credits do not read as money. They read as a resource, something you spend down, like a phone battery. Nobody audits a battery. You use it until a warning appears, and by the time the warning appears you have already crossed onto the next plan.
The credit system is not a pricing model. It is a way of not showing you the price.
I do not think anyone drew this on a whiteboard as a trick. It does not have to be a trick. Abstract units do this on their own. A credit is a layer between you and the cost, and it runs in one direction: easy to spend, hard to read. The cliff is steep, but the reason it stays hidden is not that it is complicated. It is that the units are built so you never run the conversion.
What the premium buys you
I tested all of Flows. That did not take long, because there is not much to it. It is a small set of node types, and once you have wired a few together you have seen the whole tool. Simple is not automatically bad. But you should know that the premium price is buying you a small, plain canvas, not a deep one.
The clearest example is the LLM node. The LLM node is the brain of an automated pipeline. It writes the prompts and decides what the pipeline does next. In Flows it offered me exactly two models: Gemini 2.5 Flash and Gemini 2.5 Flash Lite. That is the whole menu. Both are a generation old, and both are the fast, inexpensive tier of that generation. Whatever your pipeline thinks with, it thinks with one of those two.
Even if you ignore the pricing, the canvas itself lacks the basic tools a real workflow needs. The LLM node has no option to easily export a list, which makes it useless if you want to generate a series of prompt ideas or write image metadata. And if you need to overlay copy onto your renders, you are out of luck. There is no way to compose images and text together because Flows has no compositor node. It is a node tool that does not let you do actual node work.
But ignoring the price is a luxury you do not have when you run a business.
Let's do the actual math for a real-estate job. I charge a client $600 for a dozen finished visualization images. If I run that project through ElevenLabs Flows, the $299 Scale subscription eats exactly half of my gross invoice. Paying fifty percent of your revenue to a single infrastructure provider for one project is not just expensive. It is business suicide.
I do not understand why companies insist on shipping half-baked image and video features when they are this expensive and this limited. If you cannot build a canvas that actually solves a practitioner's problems, it is better not to do it at all. Focus on what you are actually good at and leave the nodes to someone else.
One thing I will say plainly, because it is true. ElevenLabs, the audio company, is excellent. Their voices and their voice cloning are among the best available, and I would still send anyone there for sound. This is a review of their image and video generation, and of how it is priced and gated. Both things are true at once.
I built Scopeful, where I review every creative AI tool. I add the price per model that the tools offer and do the math to show the actual price per generation. But you do not need Scopeful to run it. You need a few minutes and the discipline to spend them before you subscribe instead of after.
I have tested all of Flows now, and run none of my real work through it. The canvas was quick to learn. The price is the part that should take you longer.
Want to check your own bill before you subscribe? The calculator at Scopeful runs the credit math on every major creative AI tool.
Scopeful tracks where creative-AI prices move, credit systems converted into plain dollars, every week. If you would rather see the cliff before you are standing on it, that is what the newsletter is for.