-
Notifications
You must be signed in to change notification settings - Fork 0
NanoBanana2_User_Guide_en
zxwzxw edited this page Mar 10, 2026
·
2 revisions
NanoBanana2 is a UE5 Gemini image generation plugin with both Runtime and Editor modules:
-
NanoBanana2(Runtime): async Blueprint nodes, generation requests/results, history/prompt library helpers, image split tools. -
NanoBanana2Editor(Editor): full control panel UI (CN/EN switch, reference image workflow, queue, history, prompt library, split preview).
Core capabilities:
- API key verification
- Text-to-image and image-guided generation (multi-reference images)
- SSE streaming generation (text/image chunks)
- Batch queue generation
- Prompt rewriting with retries and fallback model
- History and prompt library management
- Image splitting with custom ratios
- Put the plugin under your project:
YourProject/Plugins/NanoBanana2 - Open UE5, enable the plugin in
Edit -> Plugins - Restart the editor if needed
Open from:
Window -> NanoBanana2
- Enter your Gemini API key and verify it.
- Enter your prompt.
- Optionally add one or more reference images.
BrowseandCapture VPboth append to the existing set. - Choose model, resolution, aspect ratio, and output directory.
- Click generate.
- Result is saved automatically and added to history.
Notes:
- Generation works even without reference images.
- Retry is controlled by
MaxRetries.
- API key visibility toggle and verification status
- Chinese/English UI switching
- Multiple local reference images are supported.
- The old read-only path textbox is removed; references are managed via buttons + tag list.
-
BrowseandCapture VPappend into the current reference set (no implicit clear). - Import from disk or reuse from history.
- Remove specific images from the current reference set.
- Empty reference set is valid (prompt-only generation).
-
Capture VPalso records viewport pose (location/rotation/FOV). The pose can be linked to generated history records.
- Sends one request with current settings.
- Returns status, retry count, token info, estimated cost, response text, and saved image path.
-
AI Rewriteis now placed inside the prompt editor area (top-right). -
Open Latest Imagebutton has been removed from the center workspace.
- Emits text/image chunks progressively.
- Useful for live UI updates while generation is running.
- Build a queue of requests.
- Per-item completion callback and final queue completion callback.
- Queue items can be removed from the UI.
- Rewrite prompt with style mode and thinking settings.
- Supports fallback model and retry on transient failures.
- Auto-saves generated results to history.
- Search, remove record, delete associated file, and double-click to apply.
- Save reusable prompts with thumbnails.
- Search, apply, delete, and assign thumbnail from current preview.
- Prompt thumbnails are now independent from history files.
- Split by rows/columns.
- Supports custom X/Y split ratios.
- Split preview and final split output are both supported.
- When showing viewport overlay, the panel first restores the linked captured pose, then overlays the image.
- If absolute paths differ in format, pose lookup falls back to filename matching for better robustness.
- The workflow no longer depends on spawning/locking camera actors.
Default location:
Project/Saved/NanoBanana2/outputsProject/Saved/NanoBanana2/history.jsonProject/Saved/NanoBanana2/prompt_library.jsonProject/Saved/NanoBanana2/prompt_library_thumbnails
VerifyApiKeyGenerateImageStreamGenerateImageProcessQueueRewritePrompt
- Paths:
GetDefaultOutputDirectory,GetHistoryFilePath,GetPromptLibraryFilePath - History:
LoadHistory,SearchHistory,RemoveHistoryByFilePath,ClearHistory,DeleteGeneratedFile - Prompt library:
LoadPromptLibrary,SavePromptToLibrary,SearchPromptLibrary,RemovePromptFromLibrary - Validation/models:
IsValidAspectRatio,GetSupportedAspectRatios,GetSupportedModels - Cost:
EstimateCostCNY - Image tools:
SplitImage
Current supported models:
gemini-3.1-flash-image-previewgemini-3-pro-image-previewgemini-2.5-flash-image
Supported aspect ratios:
-
Auto,1:1,9:16,16:9,3:4,4:3,3:2,2:3,5:4,4:5,21:9
Yes. Prompt-only generation is fully supported.
Usually a local image decode failure (corrupt file, unsupported payload, or invalid path).
No. Prompt thumbnails are stored independently in prompt_library_thumbnails.
- API key validity
- Network reachability to Gemini endpoints
- Model/parameter correctness
- Output directory write permission
- Keep reusable prompt templates in the prompt library.
- Use queue mode for batch production.
- Validate composition with lower resolution first, then upscale.
- Periodically clean old outputs and history to control
Savedsize.