Version history

Release notes.

The version history of every Bielfor product.

Oct 13, 2026

Version 2.3.0: Nano Banana 2.1 joins the panel.

4 changesHide details

Added: Nano Banana 2.1 from Google, with 1K, 2K and 4K output, reference images, Google Search and Thinking Level controls.
Nano Banana 2.1 is the default Google model for a new installation or when switching to Google. A saved model choice is preserved. Nano Banana 2 remains available.
The lineup now includes eleven image models across Google, OpenAI and xAI. Provider API usage is billed separately on your own keys.
Published on aescripts + aeplugins on 13 October 2026. Adobe Exchange lists version 2.3.0 as updated on 14 October 2026. Both public listings were verified on 16 October 2026.

Sep 24, 2026

Version 2.2.0: two new OpenAI models, and costs you can see before you pay.

6 changesHide details

Two new OpenAI models: GPT Image 2.5 Sunburst for precise edits, and GPT Image 2.5 Flare for speed.
Both add Extra High and Max quality for the most detailed results.
See the cost before you pay: the status line shows an estimate while generating, then the real cost next to it.
Google costs now include thinking tokens and are read from each response, so Usage & Spend matches your Google bill.
Nano Banana 2 input price updated to Google's current rate.
Released: published on aescripts + aeplugins on 24 September 2026. At that publication date, the Adobe Exchange submission was in review.

Aug 27, 2026

Version 2.1.0: Grok Imagine 2.0, and three seconds to change your mind.

5 changesHide details

Added: Grok Imagine Image 2.0, xAI's current image model and the new default when you pick xAI. It generates and edits, outputs 1K or 2K, and takes your selection plus up to two reference images. The lineup is now eight models across three providers.
Added: a Quality control for Grok Imagine Image 2.0, Low for cheap drafts and Medium for final work. It is also the first Grok model whose price moves with both quality and resolution, from $0.04 to $0.08 per image, and the first that bills $0.01 for each reference image. The pre-submit cost hint follows your choice and the tier is recorded in Usage & Spend. The two earlier Grok models are unchanged: still flat per image, still free on references.
Added: Send Delay (Settings, Generation), off by default. Providers bill the moment a request is sent, so cancelling after that costs you anyway. Switch it on and every generation waits three seconds first, counting down with Cancel live. Cancel inside that window and nothing is sent, generated, or billed. Leave it off and generations start immediately, as before.
Changed: Submit with nothing selected now selects the whole canvas instead of stopping with an error. If the document is too far from square to send whole, the message says so and asks you to select an area.
Released: published on aescripts + aeplugins on 27 August 2026.

Jul 2026

Version 2.0.0: three providers, seven models, one panel.

18 changesHide details

Renamed: Banana MAX is now Creator MAX, under the new plugin ID bielfor.creator-max. Photoshop treats the new ID as a separate installation, so 2.0.0 appears alongside an existing 1.x panel rather than upgrading it in place; remove the old version once you've moved over.
Added: OpenAI and xAI as providers. GPT Image 2 (1K / 2K / 4K, where 4K is the model's ~8-megapixel ceiling) and Grok Imagine Image in Quality and Standard tiers (1K / 2K) join the Google lineup. Provider pills and a per-provider model picker replace the old two-model toggle; every model shows its cost range in the picker.
Added: two further Google models: Nano Banana 2 Lite (1K only, roughly half the price of Nano Banana 2) and the original Nano Banana. The lineup is now seven models across three providers.
Added: Manage Models. Enable or disable whole providers and individual models; only enabled ones appear in the picker. Out of the box only Google is shown. At least one provider and one model per provider always remain active.
Added: Manage API Keys. Separate keys for Google, OpenAI and xAI, each validated live against the provider and removable individually. One key is sufficient to run the plugin. Calls continue to go directly from Photoshop to the provider, with no intermediary and no markup.
Added: Options. Request one to four variants per submit; they generate in parallel and paste back as individual layers grouped under 'Creator MAX options'. Partial failures are reported and the successful variants are kept.
Added: multi-document concurrency. Each open document can run its own generation; results route back to the file that requested them and are applied silently on return. A shared cap of four in-flight requests protects the providers, and the panel indicates when other files are generating.
Added: Quality control for GPT Image 2 (low / medium / high) with per-quality cost and time estimates. Actual OpenAI cost is computed from the token counts the API returns, with the cached-input discount credited when present.
Added: aspect-ratio control for Grok (Auto plus seven fixed ratios; Auto follows your selection).
Added: Usage & Spend. A persistent 'Today' spend chip in the top bar, backed by an on-disk monthly ledger, opens a dashboard with Today / 7-day / 30-day / all-time totals, a per-model breakdown, recent generations, the full rate table, and CSV or Markdown export. Every entry is marked by cost basis: measured, table rate, or estimate.
Changed: Enhance Prompt is provider-aware. The rewrite model is selectable per provider from a curated set of fast vision-capable text models; flagship tiers are deliberately excluded. Enhancement cost is now recorded with the generation.
Changed: reference limits are per model: 14 on Google, 15 on OpenAI, 2 on Grok. References beyond the active model's cap grey out with an explanatory note instead of failing, and are restored when you switch back to a higher-cap model.
Changed: panel controls follow the active model. System Prompt appears on Google models only, Thinking on Nano Banana 2 and 2 Lite, Google Search on Nano Banana 2 and Pro. A hidden control can no longer submit an unsupported parameter.
Changed: Prompt History reworked: search, provider filters and pagination; reference and result images displayed side by side at full card width; relative timestamps and model / feature / cost badges; click any prompt to copy it. Reuse restores the exact model, resolution, provider and the original source selection. Clear All requires confirmation, and deleting an entry no longer removes other entries with identical prompt text.
Changed: History opens instantly. Cards are pre-built and thumbnails pre-decoded at startup; the underlying storage decode was rewritten, removing the multi-second stall on large histories.
Fixed: provider errors are mapped to plain language across all three providers: invalid key, billing, rate limits, content policy, overload and timeouts each state the cause and the next step, with the provider's own detail preserved beneath.
Fixed: costs under $0.10 display with three decimals instead of rounding up. Generation timeouts were raised (five minutes on Google, seven on OpenAI), and Show Files now opens the image folder directly.
Released: published on the Adobe Creative Cloud Marketplace on 14 July 2026.

Apr 17, 2026

Faster default model, a visual history you can rerun, and an end to the 500 errors.

11 changesHide details

Nano Banana 2: Google's faster Flash-tier model, now the default. It's quicker and cheaper for everyday edits. Nano Banana Pro stays one click away for the shots where you want the extra quality.
Prompt History became a real gallery. Your last 50 generations show their actual reference and result images, the original prompt and the enhanced version, and what each one cost. Reuse restores the exact setup in one click, and you can export the whole thing as JSON, CSV or Markdown.
Enhance Prompt: one click turns a rough line into a fuller, production-ready prompt, so you spend less time wording the instruction and more time judging the result.
Thinking Level on Nano Banana 2 (None, Minimal, High) to trade speed for quality when a render is worth the extra reasoning.
Google Search grounding: let a generation pull live web context for current events, trends or fresh references, instead of relying only on the model's training.
Photoshop Action recording: every generation is captured to the Actions panel automatically, so you can replay or batch a setup you liked.
System prompt presets: save a few hidden style instructions and switch between them in one click, without rewriting them into every prompt.
Fix: large reference images no longer fail. Big PNGs used to make Google's image models return a bare 500. References are now compressed before they're sent, which killed those errors and made every generation a bit lighter.
Fix: when Google's servers are overloaded, the panel now explains it in plain English and suggests switching models, instead of showing a raw API code.
Fix: inpainting pastes back aspect-perfect. The model sometimes returns a result at a slightly different ratio than your selection; the panel now crops and fits it to the exact selection so it lines up cleanly every time, even at 4K.
Fix: Prompt History opens instantly. It used to freeze for a second or two loading every thumbnail from disk; thumbnails are now cached so the window appears with no wait.

Feb 27, 2026

The first release: AI image generation and inpainting, native in Photoshop.

4 changesHide details

The first public version, on the Adobe Marketplace. A native UXP panel that lives inside Photoshop, paid for once, no subscription.
Generate images and edit a selection in place with Google's top Gemini image model, with the result pasted straight back as a layer. No round trip to a separate app, no losing your place in the document.
Bring your own Gemini key: calls go directly from Photoshop to Google. I collect nothing. Your key, your prompt history, and your billing all stay yours.
Prompt presets and a saved prompt history from day one, because a tool you use daily should remember what you did yesterday.