Let me paint a picture and you tell me if it sounds familiar.
You're not someone who generates one image a week for fun. But you're also not running a full creative studio with a dedicated AI pipeline and a spreadsheet tracking cost-per-output. You're somewhere in the middle — maybe you're a marketer who needs a steady stream of visuals for campaigns, or a content creator who uses AI-generated assets as part of a broader workflow. You're on Google Flow regularly. You've figured out the basics. You've started to build habits around it.
And then you start hitting the walls.
That's who this piece is for. If you open Google Flow maybe once a month when inspiration strikes, this probably isn't your article. But if you're in there multiple times a week, generating across different projects and use cases, trying to get real work done — keep reading, because some of what I'm about to describe is going to sound very familiar.
Okay But Before I Start Complaining
Before I get into the frustrations, I want to give credit where it's due, because there's a lot of it.
The free tier gives you 180 credits just for showing up, which is not nothing. And if you're on Google AI Pro, image generation effectively costs you zero credits — you can generate images all day without watching a counter tick down. On top of that, Pro users get 1,000 AI credits for video generation, which is a meaningful budget for a subscription that most people are already paying for other Google stuff anyway.
The async generation is something I don't think people appreciate enough. You fire off a batch, go do something else, come back and your outputs are waiting for you. No babysitting the progress bar. No watching a spinner for 45 seconds and then doing it again. For anyone running multiple generation tasks in a session, this is genuinely a workflow upgrade.
And then there's the model integration, which is where Google Flow really shows its hand. The pipeline between Nano Banana Pro / 2 for image generation and Veo 3.1 for video generation is smooth in a way that feels intentional and considered. You generate a still with Nano Banana, and it flows into Veo 3.1 without the usual export-import-reformat song and dance. If you've tried stitching together AI image and video workflows using separate tools, you know how much friction that normally involves. Google Flow makes it feel like one thing instead of two things awkwardly holding hands.
So yes — genuinely good. Genuinely worth using. And also: genuinely has some rough edges that will start to bother you the more you use it.
The "Good Idea, Mixed Results" Category
These next two aren't bugs exactly. They're design choices that make sense in theory, and sometimes work great in practice, and sometimes make you want to flip a table.
Style Contamination: The Memory That Won't Forget
Here's something that will sneak up on you. You spend a session generating a moody, desaturated editorial look for one campaign. Great results, you're happy. Then you pivot to generating something bright and poppy and commercial for a completely different project. And no matter what you put in your prompt — no matter how many times you write "vibrant," "colorful," "high saturation" — the outputs keep carrying some ghost of that earlier aesthetic. The model has developed a preference, and it's not asking for your opinion.
In some contexts, this is actually a feature. If you're building out a visual world for a single project — consistent character, consistent tone, consistent art direction — the style memory can do a lot of heavy lifting for you. You don't have to keep re-specifying everything from scratch.
But if you're the kind of user who needs to generate across multiple totally different contexts in a single session — different clients, different moods, different use cases — this becomes a real problem. The style bleed is subtle enough that you might not notice it at first, and stubborn enough that prompting your way out of it is often more effort than it should be.
What actually works: Switch projects. Seriously, just switch to a different project for different visual directions. It sounds almost too simple, but starting fresh in a new project context breaks the style memory in a way that reprompting usually doesn't. If you're doing wildly different work in the same session, treat different projects like different workspaces and don't try to do it all in one place.
The Laziness Problem: When Precision Is the Wrong Tool
This one might be specific to Nano Banana 2, but it's worth flagging.
When you ask for a revision — "change the background," "make it warmer," "swap the jacket color" — the model often does exactly and only that. Which sounds like it should be a good thing. Precise, controlled, surgical. And sometimes it is.
But a lot of AI image users have developed workflows built around the opposite behavior. The "throw spaghetti at the wall" approach — you give a vague direction, and the model uses it as a jumping-off point to generate something genuinely different and surprising. That randomness, that creative reinterpretation, is part of the value proposition for a certain kind of user. It's how you find results you wouldn't have thought to ask for directly.
With Nano Banana 2 in Google Flow, that's often not what you get. Ask for a revision and you'll get a very tidy, very literal version of that revision, with everything else frozen in place. For users who were hoping the model would take the wheel a little, this feels like pushing a boulder uphill — you have to be very deliberate and specific about every single thing you want changed, which is more work, not less.
What actually works: Put more into your prompts. If you want genuine creative variation, you have to ask for it explicitly rather than hoping the model will improvise. And again — switching to a new project seems to reset the model's disposition in a way that makes it more generative and less literal. It's the blunt instrument solution, but it tends to work.
The Actual Limitations
Okay, now we're past the nuanced stuff and into the things that are just... limitations. No both-sides framing here.
The Size Selection Problem
Google Flow supports a reasonable set of aspect ratios and dimensions for image generation. "Reasonable" being the key word. For casual use, you'll probably never notice the gap. For people actually using AI generation as part of a production workflow, you will absolutely eventually need a size that isn't on the list.
Weird crop ratios for specific platform placements. Ultra-wide formats for web banners. Portrait dimensions that don't match any of the standard options. This stuff comes up constantly in real marketing and content work.
And before you ask — no, prompting your way around it doesn't work either. Telling it to "generate the image at X dimensions" or "keep the subject centered within an X by X frame and leave room for cropping" will get you absolutely nowhere. Nano Banana will just... do something else entirely. Change the background color. Rearrange elements that didn't need rearranging. Something confidently wrong that has nothing to do with what you asked. It's not ignoring you so much as it's interpreting your size request as a vibe and acting accordingly, which is somehow worse.
What actually works: For occasional odd-size needs, Gemini can handle some of what Google Flow can't — but you're trading away the async speed advantage and accepting slower generation times, with a real risk of outright failures during peak hours. Not ideal.
For higher volume needs or more unusual dimensions, APIPASS's Nano Banana 2 API is worth knowing about. The default parameter set includes aspect ratios like 1:8, 8:1, and 21:9 — the kind of dimensions that actually show up in real production work — and it handles them natively without workarounds. If size flexibility is a recurring issue for you rather than an occasional one, the API path is probably the cleaner long-term solution.
The Cost Reality: Google Flow Can Get Pricey
This is where we need to have an honest conversation, because the pricing picture is more complicated than it first appears — and understanding it will change how you think about when to use Google Flow and when to look elsewhere.
For users who are hitting Google Flow's limitations on size flexibility or generation behavior, the API route is often where people end up anyway. And once you're already looking at APIs, it's worth running the actual numbers — because depending on how you work, APIPASS can come out meaningfully cheaper, and the billing model alone makes it worth a serious look. Unlike Google One's monthly subscription, APIPASS credits don't expire, which matters a lot if your generation volume isn't perfectly consistent month to month.
Google Flow's pricing starts with the subscription tier you're on:
| Plan | Monthly Price | Included Credits |
|---|---|---|
| Google AI Pro | $19.99/month | 1,000 credits |
| Google AI Ultra | $249.99/month | 25,000 credits |
This subscription structure matters because it affects the real cost of using Google Flow. If you use enough images or videos each month, the monthly plan can be cost-effective; if your usage is light, unused credits expire before you fully benefit from them.
Video: Pricing Comparison
For video generation, it's worth taking a close look at APIPASS's Veo 3.1 API before defaulting to Google Flow. The price difference on fast-tier video is significant, and if you're generating at any real volume, it adds up fast.
| Model / Plan | 8-Second Video Price (USD) | Billing Model |
|---|---|---|
| APIPASS Veo 3 Lite | $0.1364 | Pay-as-you-go credits |
| APIPASS Veo 3.1 Fast | $0.1364 | Pay-as-you-go credits |
| APIPASS Veo 3.1 Quality | $1.1818 | Pay-as-you-go credits |
| Google AI Pro / Flow Fast | $0.3998 | Monthly subscription |
| Google AI Ultra / Flow Fast | $0.2000 | Monthly subscription |
| Google AI Pro / Flow Quality | $1.9990 | Monthly subscription |
| Google AI Ultra / Flow Quality | $1.0000 | Monthly subscription |
For fast video generation, APIPASS is clearly cheaper. Its Veo 3 Lite and Veo 3.1 Fast both cost $0.1364 per 8-second video — well below Google Flow's $0.2000 on Ultra, and far below $0.3998 on Pro. If fast-tier video is the bulk of what you're generating, APIPASS Veo 3.1 Fast is a straightforward win on price.
For quality video generation, the picture changes. APIPASS Veo 3.1 Quality costs $1.1818 per 8-second video, while Google Flow on Ultra comes in at $1.0000. Google's Ultra plan is slightly cheaper for this tier, although APIPASS still beats Google Pro by a wide margin.
The key difference that matters more than any individual price point: APIPASS credits don't expire. Google Flow runs on a monthly subscription, and whatever you don't use in a given month is gone. If your video generation needs are lumpy — heavy one month, light the next — you're going to feel that expiry in a way that adds up over time.
Image: Pricing Comparison
For images, APIPASS's Nano Banana 2 API and Nano Banana Pro API are both worth knowing about — especially if you've already been pushed toward the API route by the size flexibility issue. You get the same underlying models as Google Flow, native support for non-standard dimensions, and a straightforward pay-per-image structure with no expiry clock running in the background.
| Model | APIPASS Image Price | Billing Model |
|---|---|---|
| Nano Banana 2 1K | $0.0455 per image | Pay-as-you-go credits |
| Nano Banana 2 2K | $0.0682 per image | Pay-as-you-go credits |
| Nano Banana 2 4K | $0.1000 per image | Pay-as-you-go credits |
| Nano Banana Pro 1K | $0.0864 per image | Pay-as-you-go credits |
| Nano Banana Pro 2K | $0.0909 per image | Pay-as-you-go credits |
| Nano Banana Pro 4K | $0.1727 per image | Pay-as-you-go credits |
| Google Flow / Google AI | $0 marginal cost per image if image generation uses 0 credits | Monthly subscription |
For images, APIPASS gives you a simple fixed price per generation — $0.0455 to $0.1727 per image depending on model and resolution. Predictable, easy to budget, no expiry to worry about.
Google Flow is a different calculation entirely. If image generation costs zero credits on Pro, then the marginal cost of each additional image is $0 after you've already paid for the subscription. That makes Google extremely cost-effective for high-volume image generation — but only if you're already generating enough to justify the monthly fee in the first place.
The real question isn't "what is cheaper per image." It's how many images you generate every month. Irregular or low-volume user? APIPASS is probably the better deal — you only pay for what you use. Generating at high volume consistently every month? Google Flow starts to win on image economics because the subscription cost gets spread across more and more outputs.
Which One Fits You?
Choose APIPASS if:
- You generate content occasionally, not every month
- You want credits that never expire
- You prefer a pay-as-you-go model with no monthly commitment
- You want a clearer direct price per video or image
- You care about avoiding wasted subscription value
Choose Google Flow if:
- You generate a lot of content every month
- You are comfortable with a monthly subscription
- You expect to use enough images to amortize the subscription cost
- You want potentially lower cost on quality video under the Ultra tier
- You prefer a bundled AI plan rather than purchasing credits separately
That's the Whole Picture
Google Flow is good. For the right kind of user, doing the right kind of work, it might be close to ideal. Free image generation on Pro alone is genuinely hard to argue with if you're generating at volume. The Nano Banana to Veo 3.1 pipeline is smooth in a way that other multi-tool setups aren't. The async generation is a real quality-of-life win.
But the style contamination will get you. The precision-over-creativity behavior will get you. The size limitations will get you at some point. And once you start generating video with any regularity, the economics get more complicated than the "it's basically free" impression might have given you.
None of that means don't use it. It means use it with clear eyes, keep the workarounds in your back pocket, and know when it makes sense to reach for something else.
And on that note — if you want to explore what the API side looks like for more flexible sizing, non-expiring credits, or better fast-video pricing, we've got Nano Banana 2, Veo 3.1, and a bunch of other models available through APIPASS. Worth a look if any of the limitations above are hitting you regularly.
More soon.
— Julian