The new AI art product this week is a price table.
On September 11, The Hype Magazine listed ten image generators “powered by or rivaling Nano Banana”. Rank one is “Nano Banana Bingo,” sold on character consistency, conversational edits, and 4K references, starter $9.90 a month for 800 credits. Rank two is Adobe Firefly as a multi-model surface: Gemini 3 / Nano Banana, ChatGPT Image 2, FLUX.2 pro, Veo 3.1 Fast, Kling 3.0, Runway Gen-4.5, Ray3.14 SDR, up to six reference images, Standard $9.99 for 2,000 credits. Kling’s own app sits at three, from $8.80 after an intro. EaseMate, further down, stuffs GPT Image 2, Nano Banana, Wan 2.5, Flux Kontext, and Midjourney into one chat box and takes five reference files.
Two days earlier, Lifehacker ran ChatGPT Images 2.5 against Gemini Nano Banana 2 and gave OpenAI a slight edge, mostly because Sketch and saved prompt templates exist. We already covered Images 2.5 as an edit-history claim on September 12. This is the other half. The models are not multiplying as fast as the storefronts that resell them.
The roundup is a shelf, not a bake-off
Hype Magazine’s table is indicative pricing, not a lab. Take the ranks as merchandising. Take the SKUs as real.
Firefly’s list is the tell. Adobe is not arguing that one in-house look beats Midjourney. It is arguing that your Creative Cloud seat should be the place you pick Gemini, OpenAI, Flux, and three video models without opening five tabs. We already watched Disney park Firefly inside Imagineering. The September 11 list is the consumer version of that move: the model is inventory.
Nano Banana Bingo’s pitch is the part that still belongs to Google’s family: characters that survive a conversation, references at 4K. If that is true, it is a model property. Wrapping it in a $9.90 starter plan does not add a new latent space. It adds a credit meter and a brand.
EaseMate’s five-image reference upload is a workflow feature. So is Firefly’s six. That is the actual competition among wrappers: how many stills you can pin, whether the chat retains the pin, whether the invoice is one line. Older comparison pieces aged into this. The 2024 question was whose pretty pictures won. The 2026 question is who owns the session.
Kling at rank three is an image-plus-video vendor. If you need a still and a clip from the same subject, a combined plan can beat paying Midjourney and Runway separately. If you only need stills, you are buying a video company’s overhead.
Lifehacker is still a taste test
Lifehacker’s author thinks Images 2.5 is slightly ahead of Nano Banana 2 on the day they shot food and a Blade Runner riff. They also think a real forest beats both. Fine. Taste is not a contract.
The operational detail worth keeping: Sketch lives behind the + button, and on mobile you have to find Plugins first. You draw a bad layout, then type a style. That is a layout tool bolted onto a generator. It is not a new generator. Templates and shared prompts are distribution. They help people who teach prompts. They do not lock a face.
Nano Banana 2, in that piece, is the February Google line. The Hype list is already selling “Nano Banana” as a flavor other apps pour. If you are comparing 2.5 to Banana 2 in ChatGPT and Gemini, you are in the vendor apps. If you are comparing them inside Firefly, you are in Adobe’s router. Same weights, different logs, different refusal layers, different bills.
Usage limits on ChatGPT plans are still “vague,” which is Lifehacker’s word. A wrapper that prints credit counts is, in one sense, more honest. A wrapper that hides which model actually ran is not.
Do not treat a Blade Runner side-by-side as a production test. Nobody’s campaign is a still of a wet street. Production is: ten edits later, is it the same person, and can legal see the log.
Virality is the other SKU
Analytics India Magazine asked what happens when you run the 1980s selfie trick on September 12. The Ghibli wave, then action-figure boxes, then Nano Banana’s collectable figurines and younger-self portraits. Google’s numbers in that story: more than 500 million Nano Banana images since launch, and more than 10 million new Gemini users attributed to the feature.
That is user acquisition with a style pack. Miyazaki’s 2016 “insult to life itself” line came back, because it always does. Copyright and consent came back, because the training-data cases did not end. None of that is resolved by a $9.99 Firefly plan.
If your job is pictures, the trend cycle is a warning. A look that pulls 10 million accounts is a look that will be the default sludge in a month. Character sheets and product shots do not benefit from being in the sludge. They benefit from a pin that survives the sludge.
Wrappers love trends because trends burn credits. A starter plan with 800 credits is sized for a weekend of figurines, not for a catalog. Read the meter before you build a pipeline on a rank-one name that exists to sit on a list.
What to actually pay for
If you already live in Photoshop, Firefly’s router is the least stupid place to try Gemini and OpenAI on the same file, as long as you write down which model produced the keeper. Adobe’s $9.99 / 2,000 credits is a trial price. Campaign volume will not stay there.
If you need conversational character lock and you believe Google’s stills more than OpenAI’s, use Gemini and keep the character sheet in the chat. Paying Bingo or EaseMate to do that is only worth it if the vendor app is blocked, or if you need Midjourney in the same thread.
If you need layout control, Images 2.5 Sketch is a reason to stay in ChatGPT for the first pass. Then export. Do not finish a campaign in a plugin menu on a phone.
If you need video from the same subject, price Kling against Runway against Firefly’s video list. Do not assume the still model and the video model share identity because they share a landing page.
If you are a student or a one-person shop, one vendor app plus a folder of references on disk still beats three wrappers. The folder is the part that survives a rebrand.
The session is the product
OpenAI’s September 8 claim was that edit ten still looks like edit one. Google’s claim, spread across a year of Banana posts, is that a character can be talked into new clothes. Adobe’s claim this week, via a listicle, is that you should not leave the suite to check those claims.
Only one of those changes your files. The session memory, the reference slots, the log. The rest is a store.
When a roundup ranks Bingo above Firefly above Kling, it is ranking packaging. Packaging can still waste your week. A six-reference upload that silently drops to two is a bug you will not see in a table. A credit plan that expires mid-batch is a bug you will see on a Friday.
Credit math, not rank math
Hype Magazine’s starter prices sit in a tight band: $8.80 to $19. That band is a consumer app. It is not a studio license. 800 credits at $9.90 is a weekend. 2,000 credits at $9.99 on Firefly looks generous until a 4K batch and a video model share the pool.
If Firefly’s dropdown includes Veo and Kling, your stills budget will die the first time someone on the team “just generates a clip.” Separate the still project from the motion project or you will spend Friday explaining a zero.
Five reference images on EaseMate versus six on Firefly is not a reason to switch suites. It is a reason to test whether reference three still affects edit eight. Wrappers like to advertise the upload count and then silently recap the prompt without the files. You only catch that if you keep the sheet on disk and compare.
Lifehacker’s Sketch path on mobile is a warning about where you draft. A layout drawn on a phone plugin will not be the file you hand to legal. Draft there if you must. Save the PNG. Finish where you have a history list.
Analytics India’s 500 million Banana stills and 10 million new Gemini users are success metrics for Google. They are pollution metrics for you. A style that onboarded 10 million people is a style clients will start rejecting by name. Product work should look like the product, not like the trend that paid for the model’s ads.
Disney already taught the industry that Firefly can live inside a real pipeline. A listicle ranking Bingo first does not undo that. It just proves the SEO layer caught up to the SKU names.
Kittl, ImagineArt, and Higgsfield on the same Hype table are design editors and parallel generators. They matter if your job is a poster with type, not a character who has to survive ten wardrobe changes. Do not buy a template mill to solve a consistency problem. Do not buy a consistency wrapper to set type.
If a vendor will not show which underlying model ran after you hit generate, stop using them for anything you have to defend. Firefly at least names the dropdown. ChatGPT and Gemini are the dropdown. Bingo and EaseMate need to prove they are more than a credit card on top of an API.
C2PA and watermarks still live on the vendor apps, not on every wrapper. If provenance matters, export from ChatGPT or Gemini and keep that file, not a re-encoded JPEG from a rank-four website.
Keep the September 12 piece for the OpenAI latency and watermark story. Keep this one for the invoice. If you can only do one test this week, pick a single character sheet, run it through Gemini, ChatGPT, and Firefly’s copy of both, and count how many edits until the face drifts. Write the number on the sheet. That number is the spec. The rest is a storefront lighting a banana.