Astria vs Pablo: Nano Banana Photoshop Plugin Guide
Pablo and the Astria Nano Banana plugin for Photoshop are the two closest products in this category. Both offer a one-time plugin licence with usage billed separately. Both accept your own Google key. Both send a Photoshop selection to Nano Banana, Seedream or GPT Image and bring the result back as a layer. Astria is the overall winner for a Photoshop professional who wants the panel to stay current without maintenance. New models appear from the Astria platform with no plugin update. Every result lands as a Smart Object at the model's full returned resolution, and the same key runs Astria workspaces and the API for jobs that never open Photoshop. Pablo's genuine strengths are published pass-through rates and agency licensing.
How we evaluated Astria and Pablo
The reader is a retoucher, photographer, designer, agency or ecommerce studio that lives in Photoshop and wants current generative models inside a live document, without a browser round trip. We scored both plugins on six criteria: model roster and update speed; returned resolution, layer type and mask handling; editing workflow; billing flexibility and bring-your-own-key support; maintenance and longevity; and production beyond the document. We reviewed public product, pricing, terms and marketplace pages on 13 September 2026. No output-quality benchmark was run.
Quick verdict: Astria vs Pablo
| Decision factor | Pablo | Astria | Winner |
|---|---|---|---|
| Models in the panel today | Nano Banana, Nano Banana 2, Nano Banana Pro, Seedream 4.0 and 4.5, GPT Image 1.5, 2 and 2.5 (pricing page and Exchange notes) | GPT Image 2 (mask-aware, 1K/2K/4K), Nano Banana 2, Nano Banana Pro, Seedream 5 Pro, Seedream 5, Flux 2 Pro; GPT Image 2.5 Sunburst and Flare and Muse Image already on the platform the panel reads from | Astria for current-generation Seedream, Flux and mask-aware GPT Image |
| New-model lead time | Tied to plugin versions: GPT Image 2.5 arrived in version 0.7.46 on 11 September 2026, three days after OpenAI's release; the user must update | Model list served by the Astria platform; a model added to Astria appears in the panel with no plugin update or re-download | Astria |
| Returned resolution and layer type | 4K generation advertised for Nano Banana Pro; GPT Image priced at 1K; returned pixel size and layer type not publicly verified | Up to 4K per generation; each result is a Smart Object holding the model's full returned resolution | Astria |
| Selection-to-layer workflow | Inpainting and "add and replace objects" advertised; result returns as a layer; selection handling not detailed publicly | Marquee, lasso, brush or colour range; prompt; generate. Mask-aware models read the selection directly; result lands as a new layer | Astria |
| Reference-driven edits | Multi-reference images; one public Exchange review asks for more than four; behaviour with a selection not detailed publicly | References dropped into the panel act as visual instructions, alongside a selection that mask-aware models read directly | Astria for references combined with a mask-aware selection |
| Billing paths | Pablo credits at pass-through rates, or your own Google or Replicate key | Astria key for pay-as-you-go across every model, or your own Google AI Studio key; the same balance drives Astria workspaces and the API | Astria for a balance that is not stranded in one plugin |
| Commercial model | Individual and agency licences, hosted usage with published pass-through rates, or supported provider keys; see Pablo pricing | One-time plugin licence, hosted pay-as-you-go, or a Google key for Gemini models; see Astria pricing | Depends on seat count and model mix |
| Production beyond the document | Plugin plus an account dashboard for licences and credits | Same key runs Astria workspaces, templates and the API for collection-scale work | Astria |
| Maintenance and longevity | On Exchange since 4 December 2025; version 0.7.46 in September 2026; dated blog; support desk | 16,000 sales and 187 ratings (first-party); CCX/UXP rebuild in 2026; dated changelog; one surface of the Astria platform | Astria |
| Best fit | An agency buying seats in bulk that wants published pass-through rates and Nano Banana, Seedream 4.x and GPT Image in Photoshop | A retoucher, photographer or studio that wants every current model as a full-resolution Smart Object, with a production route for the rest of the catalogue | Astria |
| Overall | The closest rival, with public rates and agency licensing | Zero-day models, Smart Objects at original resolution, and a platform behind the panel | Astria |
Reviewed 13 September 2026.
Pablo: a hosted-credit panel with published rates
Pablo is a UXP plugin for Photoshop 26.0.0 or higher, made in Amsterdam and listed on Adobe Exchange on 4 December 2025. Its pricing page lists Nano Banana, Nano Banana 2, Nano Banana Pro, Seedream 4.0 and 4.5, and GPT Image 1.5, 2 and 2.5. The Exchange listing describes multi-reference images, 4K generation and "add and replace objects". Its note for version 0.7.46, updated 11 September 2026, adds GPT Image 2.5 Flare and Sunburst.
Commercially, Pablo publishes what many rivals do not: individual and agency licensing plus hosted rates organized by model and resolution, which it says carry no markup over the provider rate. You can also bring a Google or Replicate key. Check Pablo's pricing page for current licence, usage, refund, and seat terms.
Two things to read before paying. Pablo's terms say prepaid credits expire 24 months after purchase unless otherwise specified at checkout, and one public Exchange review alleges a 90-day window on a balance. Check the expiry shown when you top up. Separately, the public pages do not say what layer type a result lands as, or what pixel dimensions come back when a 4K generation is placed into a small selection.
Astria: the panel is one surface of a platform
The Astria Nano Banana plugin for Photoshop is a CCX/UXP panel for Photoshop 2025 (version 26 and above) on macOS and Windows. The panel carries GPT Image 2 (mask-aware, at 1K, 2K or 4K), Nano Banana 2, Nano Banana Pro, Seedream 5 Pro, Seedream 5, and Flux 2 Pro. Flux Kontext was retired in favour of these. The landing page also lists GPT Image 2.5 Sunburst and Flare and Muse Image as available now on the platform.
That last sentence is the first headline. The panel's model list is served by the Astria platform, so a model added to Astria appears in the panel the day it launches, without a plugin update or re-download.
The workflow is three steps. Select with the marquee, lasso, brush or colour range. Prompt in short, specific language. Generate, and the result lands as a new layer. Mask-aware models read the selection directly. Reference images dropped into the panel act as visual instructions. The model is chosen per task from one dropdown, and keys are switched inside the panel.
The third headline is what that layer is. Each generation is placed as a Smart Object holding the model's full returned resolution. A 4K result placed into a small selection still carries the 4K pixels, so it can be re-transformed, masked and re-edited later without resampling.
The Astria plugin uses a one-time licence with usage billed separately. One Astria key authenticates every hosted model, and Astria abstracts the Google, OpenAI, fal and ByteDance APIs so quota, authentication and retries stay invisible. Or paste a Google AI Studio key and run Gemini models against your own Google account. Current terms are on the Astria pricing page. No GPU is needed.
Public proof, with provenance
Astria's landing page reports 16,000 sales and 187 ratings. It shows seven production reels: fashion edits with references, bridal retouching, sketch and CAD to photo, product design, real estate, hair and clothing swaps, and packaging. All first-party. Pablo's home page carries testimonials drawn from Adobe Exchange and Trustpilot, and its Exchange listing shows 27 ratings as of 13 September 2026. Its blog is dated: Nano Banana Pro on 20 November 2025, Nano Banana 2 on 26 February 2026, GPT Image 2 on 22 April 2026. Vendor-published too. Neither set is independent proof of output quality.
One campaign hero, then forty SKUs
Take a footwear campaign hero shot on location. The art director wants three edits: the shoes swapped for the new colourway from a product reference, the background extended for a 4:5 crop, and the phone in the model's left hand removed. Then the colourway has to appear on the other forty SKUs in the drop.
In Pablo
Drop the product reference in, lasso the shoes, and prompt the swap on Nano Banana Pro. Extend the canvas, select the empty band and prompt the fill. Brush over the phone and prompt its removal. Each result returns as a layer, and the cost of every generation is knowable from the published rate card. What the public material does not say is whether the swap layer holds the full 4K return or only the selection's pixel count, or whether it re-transforms later without loss. Test that before committing the hero.
In the Astria plugin
The same three edits, with a model chosen per task. GPT Image 2 with mask is the natural pick for the shoe swap: it reads the lasso as a mask and is built for clean replacement inside a brushed selection. Seedream 5 handles the 4:5 extension from a longer structural prompt. Nano Banana Pro removes the phone while keeping the hand and sleeve consistent. Each result is a Smart Object at the model's full returned resolution, so when the crop changes next week the swap layer is scaled, re-masked and re-edited without resampling.
And the other forty SKUs
Pablo is a plugin; forty more SKUs are forty more documents and forty more selections. With Astria, the key in the panel is the same key and balance that run Astria workspaces, templates and the API. Photoshop is for the hero and the fix. The workspace runs the rest of the collection from the same product reference. The AI fashion photoshoot guide covers that side of the job.
The differences that matter most
New-model lead time: winner Astria
Pablo is quick. Its Nano Banana Pro how-to went up on Google's launch day, its GPT Image 2 post followed OpenAI by one day, and GPT Image 2.5 arrived three days after OpenAI's 8 September 2026 release. The difference is not speed but mechanism. Pablo's newest model is tied to plugin version 0.7.46. A user who has not updated does not have it, and the roster is bounded by what the plugin build knows about. The Astria panel reads its list from the platform. A model added to Astria appears in the dropdown with no plugin update, which is why GPT Image 2.5 Sunburst and Flare and Muse Image are already listed as available.
Returned resolution and layer type: winner Astria
Both products advertise 4K. Only one says what happens to those pixels afterwards. Astria places every result as a Smart Object holding the model's full returned resolution. Pablo's layer type and returned pixel dimensions are not stated in its public material, so a buyer has to measure them. For a retoucher who re-crops, re-lights and re-masks a hero across a campaign's life, that is the whole reason to care about the layer type.
Cost to start: winner Astria
Both products offer a one-time plugin licence and let a Gemini-heavy user route calls through a Google key. Compare Pablo's pricing page with the Astria pricing page using the models, resolution, seat count, and retry rate you actually expect.
Published rates, agency licensing and a free allowance: winner Pablo
This is Pablo's real advantage and it deserves a plain statement. Pablo publishes pass-through rates, offers centralized agency licensing, and runs both hosted and bring-your-own-key paths, with Replicate as a second key option. An agency buying seats in bulk, whose finance team wants the usage model documented before the first generation, will find Pablo's commercial pages easier to sign off.
Production beyond the document: winner Astria
In the public material, Pablo's account dashboard is for licences and credits, not generation. The Astria key that runs the panel also runs workspaces, templates and the API. The balance you top up for Photoshop produces the forty SKUs, from the same product reference used for the hero. For a studio whose Photoshop work is the visible tenth of a larger production, the panel is the entry point and the ecommerce workspace is the rest.
Maintenance and longevity: winner Astria
Both products are maintained. Pablo has a dated public record, a support desk and a company behind it, which is more than most Exchange listings in this category can say. Astria's panel was rebuilt as a CCX/UXP plugin in 2026, carries a dated changelog and reports 16,000 sales. It is one surface of a platform company rather than a company's only product. If the question is which panel is still current in eighteen months, the one that needs no release to see a new model has the structural edge.
Choose Pablo if...
- Published, no-markup per-image rates are a purchasing requirement.
- You are buying ten to fifty seats and want one licence with centralised billing.
- You prefer a Replicate key to a Google key or need centralized agency licensing.
- You are content to update the plugin when a new model ships and to measure the returned layer yourself.
Choose Astria if...
- You want a new model in the panel the day it reaches the platform, with no plugin update.
- You need mask-aware GPT Image 2, Seedream 5 and 5 Pro, or Flux 2 Pro inside Photoshop.
- Your heroes get re-cropped and re-lit for months, and every generation should be a Smart Object at full returned resolution.
- You want a hosted top-up balance without a recurring subscription.
- The rest of the catalogue should run from the same key in Astria workspaces or through the API.
The bottom line
Pablo is the rival to take seriously. Its rate card, agency licence and free allowance are the clearest commercial pages we reviewed, and its release cadence is measured in days. If the decision turns on those three things, buy it.
For everyone else, the Astria Nano Banana plugin for Photoshop is our recommendation. It can receive the next model without a plugin release, keeps every returned pixel in a Smart Object, and is the front door to a production platform rather than a plugin that ends at the document. How it compares with Photoshop's own model picker is covered in Astria plugin vs Photoshop Generative Fill.
What to test before choosing
Run one PSD through both panels with the same selection, reference and short prompt on Nano Banana Pro at 4K. Read the returned layer's pixel dimensions and layer type, and confirm whether a 4K return placed into a small selection kept its 4K pixels. Ask each vendor which models were added in the last month and whether any needed a plugin update. Price the run from each pricing page at the resolution you ship, including rejected generations. Read the credit terms at checkout, including any expiry, before topping up more than a week's work.
Frequently asked questions
Is the Pablo Photoshop plugin worth buying?
Pablo is a maintained hosted-credit panel with published pass-through rates and agency licensing. Buy it if transparent provider rates and centralized seats decide the purchase. For most Photoshop professionals we recommend the Astria plugin. New models arrive from the platform with no plugin update, and results land as Smart Objects at full returned resolution.
Is Astria better than Pablo?
For a retoucher, photographer or studio that wants the panel to stay current without maintenance, yes. The Astria plugin reads its model list from the Astria platform, places every result as a Smart Object at full returned resolution, and shares its key with Astria workspaces and the API. Pablo wins on published rates and agency licensing.
Do new models need a plugin update in Pablo or Astria?
In Pablo, the public record ties new models to plugin versions. GPT Image 2.5 Flare and Sunburst arrived in version 0.7.46, posted to Adobe Exchange on 11 September 2026, three days after OpenAI's release. In the Astria plugin the model list is served by the Astria platform, so a model added to Astria appears in the panel without a plugin update or re-download.
Can I use a Gemini subscription with Pablo or the Astria plugin?
No. Google only authorises third-party apps through API keys, so a consumer Gemini subscription cannot power any Photoshop plugin. Both products offer hosted pay-as-you-go credits or your own Google AI Studio key billed by Google. Pablo also accepts a Replicate key.
Which costs less, Pablo or Astria?
It depends on the licence, model mix, resolution, and retry rate. Pablo publishes its pass-through rates; Astria publishes current hosted rates on its pricing page. Compare the same representative job on both vendors' live pricing pages.
For a walkthrough of the Astria panel itself, read the Nano Banana Photoshop guide.
Sources and methodology
This article compares public product information reviewed on 13 September 2026. Pablo sources: the product page, pricing page, terms of service, Adobe Exchange listing, and Pablo blog. Pablo's site has no public changelog page; version notes were read on the Exchange listing. OpenAI release dates come from the OpenAI API changelog, and Google's Nano Banana Pro launch date from Adobe's announcement. Current prices and allowances belong on the vendors' live pricing pages rather than in this article.
Astria sources: the plugin landing page (model roster, workflow, billing paths, changelog, sales and ratings, reels, FAQ) and the pricing page.
No competitor product was purchased or run for this article; returned-resolution and layer-type behaviour of Pablo is reported from its public material only. Sales and rating figures from either vendor are first-party. Prices, rosters, plugin versions and credit terms change often; confirm them on the day you buy.