Draft Readable, On-Model Images Fast with Nano Banana 2 Lite
Last verified: November 3, 2026
Nano Banana 2 Lite is the fastest image generation and editing model in Google's Nano Banana family. It is built for speed but keeps reliable prompt adherence, strong character consistency and legible in-image text, so you can take a rough concept to a usable draft with readable lettering and recurring characters. Google released it in June 2026 under the model name Gemini 3.1 Flash-Lite Image. It is based on Gemini 3.1 Flash-Lite, part of Google's natively multimodal Gemini series, and it targets rapid ideation and high-throughput image work.
Google describes it as a significant leap in visual quality and capability over the original Nano Banana (Gemini 2.5 Flash Image) and lists it as the replacement for that legacy model. One model covers both directions natively: text in and image out, or image plus text in and image out. It is tuned for quick local edits such as color swaps, sticker creation and background adjustments, and its character alignment matches the original Nano Banana standard.
Its trade-offs are deliberate. Output stays at the 1K tier, and Search grounding is not supported. The model is also not optimized for multiple reference inputs or multi-turn sequential editing. On this page it runs in Text to image and Image to image modes.
Explore Nano Banana's Models
What the Lite Model Renders and Accepts
Output size, framing, inputs and provenance marks that shape every run.
Output resolution
1K (image size 1024px); 2K and 4K are not supported
Aspect Ratio choices
9 frames: 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9
Upload Images (Image to image)
1–10 JPG, JPEG, WEBP or PNG files, up to 10 MB each, at least 300×300 px
Prompt limit
2,048 characters
Images per run
1
Provenance marks
SynthID watermark, always on plus C2PA Content Credentials
Check These Before You Generate
Six checks that prevent ignored references, cramped layouts and muddy lettering.
Lead with one strong reference
Upload Images takes up to 10 files, but this model is not optimized for multiple reference inputs. Start with the single clearest, best-lit image of your subject. Add more only when each one brings something the first can't.
Match the upload rules
Every file must be JPG, JPEG, WEBP or PNG, no larger than 10 MB and at least 300×300 pixels. Convert HEIC or other formats, and replace small thumbnails with full-size originals, so each upload is accepted and carries real detail.
Name the edit and what stays
In Image to image, state the one change you want, such as 'turn the jacket mustard yellow' or 'replace the backdrop with plain studio gray'. Then say that everything else stays as it is. Spelling out what must not change helps keep the edit to that one area instead of reworking the whole picture.
Set Aspect Ratio before composing
Choose the Aspect Ratio first, then write the layout for that shape. In 4:5 or 9:16, stack the subject and headline vertically. In 5:4 or 16:9, give text its own side of the frame. A layout written for the wrong shape tends to crowd the composition.
Size lettering for the 1K canvas
Output stays at the 1K tier. Put the exact words you want in quotation marks, keep them to a headline or short label, and make the type a dominant element. Long paragraphs and tiny captions get too few pixels to render cleanly.
Spend the 2,048 characters wisely
Start with subject and action, then layout, exact text and style, and cut repeated adjectives. If you expand a rough idea with Enhance Prompt, check that the suggestion you pick keeps your quoted words and key details.
Nano Banana 2 Lite or Flux 3 Klein: Pick the Generator for Your Next Image
The first five rows show what each generator here lets you choose, upload and receive. The last three come from what Google and Black Forest Labs publish about the models themselves.
| Feature/Spec | Nano Banana 2 Lite | Flux 3 Klein |
|---|---|---|
| Modes you can run | Text to image and Image to image | Text to image and Image to image |
| Aspect Ratio options | 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9 | 1:1, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 21:9, 9:21 (default 1:1); Image to image also offers Auto, its default |
| Reference image upload | Upload Images (Image to image): 1–10 files, JPG, JPEG, WEBP or PNG, up to 10 MB each, at least 300×300 pixels | Add Image (Image to image): 1–10 files, JPG, JPEG, PNG or WEBP, up to 10 MB each, at least 300×300 pixels |
| Prompt length | Up to 2,048 characters | Up to 5,000 characters |
| Images per run | 1 | 1 |
| Maximum model output resolution | 1K (1024×1024); 2K and 4K unsupported | Up to 4K |
| Multi-reference generation | Not optimized for multiple reference inputs | Combines up to 10 reference images into one output |
| Developer and release | Google, June 2026 | Black Forest Labs, October 15, 2026 |
Matching Each Generator to the Job
Framing: near-square feeds vs ultrawide panoramas
Seven framings are shared, covering squares, standard portrait and landscape shapes, and 16:9 and 9:16, so everyday layouts work the same on either generator. The difference is at the edges. Nano Banana 2 Lite adds 4:5 and 5:4, which suit portrait feed posts and near-square product cards. Flux 3 Klein adds 21:9 banners and 9:21 tall panels, and its Image to image mode also has an Auto option. Decide where the image will be placed before you write the prompt.
Pixel ceiling and prompt room
Flux 3 Klein's model reaches 4K, and its prompt box takes much longer briefs. That matters for print pieces, large displays and prompts that stack many constraints. With Nano Banana 2 Lite, plan for the 1K tier and a tighter prompt: write concise briefs and aim the output at screen-sized uses such as social posts, thumbnails and slide visuals.
One reference or many
Both generators take the same uploads: one to ten images in the same four formats, with the same size limits. The models differ in how they handle several at once. Black Forest Labs builds Klein to merge up to ten references into one composition, while Google's Lite model is not optimized for multiple reference inputs. For an edit guided by a single photo, either model can do the job. For a composite that pulls products, characters and settings from several uploads, Flux 3 Klein is the better fit.
Framing, Detail Ceiling or Reference Count: Decide What Your Image Needs
Use this quick guidance to pick the best option for your workflow.
How to choose: Choose Nano Banana 2 Lite when the image needs a 4:5 portrait or 5:4 landscape frame, two ratios that Flux 3 Klein's generator doesn't list. Choose Flux 3 Klein when you need any of these: output above 1K (the model reaches 4K), ultrawide 21:9 or tall 9:21 frames, the Auto ratio in Image to image, prompts longer than 2,048 characters (up to 5,000), or composites built from several reference images. Both run Text to image and Image to image, accept the same 1–10 image uploads, and return one image per run.
Make Your First Lite Image in Five Moves
Five steps take you from choosing a mode to a finished image.
Pick a mode in the model list
Select Nano Banana 2 Lite in Text to image to start from words alone, or in Image to image to work from your own pictures.
Add pictures with Upload Images
In Image to image, upload between 1 and 10 images of your subject. Text to image skips this step.
Write in the Prompt box
Describe the image or the edit in up to 2,048 characters. To expand a rough idea, tap Enhance Prompt beneath the box for fuller suggestions.
Choose an Aspect Ratio
Pick the frame that suits where the image will appear: square 1:1, vertical 9:16, widescreen 16:9, or the 4:5 and 5:4 options in between.
Click Generate
Run it and review the image. To explore another direction, adjust the prompt or frame and generate again.