Smart Edit
Edit or transform any image with a plain-language instruction, and pick the editing model that fits the job.
Smart Edit changes an image from a text instruction: swap a background, restyle a look, combine several images, or add and remove elements. Upload one or more images, describe the edit, choose an editing model, then generate a few variations to pick from. A photo is optional: with none, Smart Edit makes a new image from your description alone, on any of its models. Different models trade speed, price, and control, so you can iterate cheaply and finish on the highest-fidelity option. The default model is Seedream 5.5.
At a glance
- You provide
- One or more images and a text edit instruction
- You get
- One to four edited image variations, downloadable
- Best for
- Background swaps, restyling, combining images, adding or removing elements
- Image formats
- JPG, PNG, WebP up to 10 MB each
- Input images
- Up to 10 on some models, up to 4 on others
- Outputs per run
- 1 to 4 (2 on some models)
- Default model
- Seedream 5.5 (2K)
Jump to Choosing an editing model for a full comparison with strengths and trade-offs.
How it works
- 1
Upload your image(s)
Drag and drop or Browse Files. Add more images up to the selected model's limit (up to 10 on some models, up to 4 on others). JPG, PNG, or WebP up to 10 MB each. To make a new image from words alone, skip this step: leave the upload area empty and describe the image you want.

- 2
Write your edit instruction
Describe exactly what should change, for example "replace the background with a plain studio grey and keep the product sharp." The prompt is required and has a built-in optimizer to help sharpen it. See the prompting guide for tips.
- 3
Pick an editing model
Open the model selector and choose the model that fits your goal (see the comparison below). The default is Seedream 5.5.

- 4
Set the model's options
Set the controls the chosen model exposes: Aspect Ratio, and where available Output Resolution (1K, 2K, 4K; 1K, 1.5K, 2K on Seedream 5.5), Quality (Low, Medium, High, X-High, Max on GPT Image 2.6 and 2.5; Low, Medium, High on GPT Image 2) and Mode (Quality or Fast on GPT Image 2.6 and 2.5, same price). Only the controls that model supports are shown. Seedream 5.5 has no Aspect Ratio control: it keeps your photo's shape unless your instruction asks for another one.
- 5
Choose how many outputs
Use the slider to set how many variations to generate, from 1 up to the model's maximum (2 or 4). More outputs multiply the credit cost.
- 6
Generate
The Generate button shows the live credit cost first. Click it and you are taken to the results page while your edits render. Finished images are saved to Previous Edits and your Generations history.
Choosing an editing model
These are the actual models in the selector. Each trades price, speed, and control differently. Match the model to what this specific edit needs.
Seedream 5.5
DefaultThe most photographic results (real skin texture, natural colour), exact text, and clean edits that change only what you ask.
- 0.5 credits per output at every size, 1 with 5 or more photos (paid)
- Resolution control: 1K, 1.5K, 2K
- Up to 10 input images and 4 outputs per run
- No ratio to pick: it keeps your photo's shape, or follows a shape named in the instruction. From words alone, name the shape in your description (for example a 16:9 banner)
- Makes a transparent PNG when the instruction asks for a transparent background
- No 4K (2K at most)
- No Quality or Mode controls
- No Aspect Ratio menu, so ask for a new shape in the instruction
GPT Image 2.6
Most controlThe most faithful edits, best prompt-following and the best text on images, with full control over quality and resolution.
- Five quality tiers: Low, Medium, High, X-High, Max
- Quality mode for the most precise edits, Fast mode for quicker results, same price
- Resolution control: 1K, 2K, 4K
- Up to 10 input images and 4 outputs per run
- Renders words on the image reliably (headlines, prices, labels)
- Priced by quality times resolution, so Max or 4K costs the most
- No Match input aspect ratio (needs an explicit ratio)
- Free plans: Low at any resolution, Medium at 1K or 2K
GPT Image 2.5
Previous versionThe same controls and prices as GPT Image 2.6, for anyone who wants to keep the previous version.
- Five quality tiers and Quality or Fast mode
- Resolution control: 1K, 2K, 4K
- Up to 10 input images and 4 outputs per run
- No Match input aspect ratio (needs an explicit ratio)
- Free plans: Low at any resolution, Medium at 1K or 2K
GPT Image 2
Oldest GPT ImageThe original GPT Image model, kept for anyone who prefers it.
- Quality control: Low, Medium, High
- Resolution control: 1K, 2K, 4K
- Up to 10 input images and 4 outputs per run
- Costs more than GPT Image 2.6 for a similar result (High at 2K is 4 credits)
- No Match input aspect ratio
Nano Banana 2.1
Best valueGood-quality edits at a low price, with sharp text and the widest set of aspect ratios.
- 0.5 credits per output at 1K or 2K, 1 at 4K (paid)
- Widest aspect-ratio choice, including 1:4, 4:1, 1:8, and 8:1
- Up to 10 input images
- Up to 2 outputs per run
- No Match input aspect ratio
Nano Banana 2
Widest ratiosFast, high-quality edits with the widest set of aspect ratios and resolution control.
- Widest aspect-ratio choice, including 1:4, 4:1, 1:8, and 8:1
- Resolution control: 1K, 2K, 4K
- Up to 10 input images
- Up to 2 outputs per run
- No Match input aspect ratio
- 4K costs more than 1K or 2K
Nano Banana Pro
Premium edits with resolution control and many reference images.
- Resolution control: 1K, 2K, 4K
- Up to 10 input images to combine or transfer from
- Match input aspect ratio available
- Up to 2 outputs per run
- 4K costs more than 1K or 2K
Nano Banana 2 Lite
Fixed 1024pxQuick, lightweight edits at a fixed 1024px output.
- Flat 0.5 credits per output (paid)
- Up to 4 outputs per run
- Fast, batch-friendly
- Fixed 1024px output, no resolution control
- Up to 4 input images
- No Match input aspect ratio
- A true-to-life photo edit that changes only what you ask? Use Seedream 5.5, the default.
- A transparent PNG? Use Seedream 5.5 and ask for a transparent background in the instruction.
- 4K, or control over quality and speed? Use GPT Image 2.6.
- Words on the image, or the most faithful edit? Use GPT Image 2.6 or Seedream 5.5.
- Good quality for less? Use Seedream 5.5 or Nano Banana 2.1.
- Cheap iteration? Use Seedream 5.5 (0.5 per output with up to 4 photos) or Nano Banana 2 Lite (a flat 0.5 per output).
- Unusual aspect ratios like 8:1 or 1:8? Use Nano Banana 2.1.
- Combining many reference images? Use Seedream 5.5, GPT Image 2.6 or Nano Banana Pro (up to 10 inputs).
- An SVG file (a logo or icon)? Ask the AI Assistant. It draws one with GPT Image 2.6 from your description.
Settings & options
| Setting | Options | What it does |
|---|---|---|
| Edit instruction (prompt) | Free text | Describe the change you want. Required. A built-in optimizer can refine it for you. |
| Model | Seedream 5.5, GPT Image 2.6, GPT Image 2.5, GPT Image 2, Nano Banana 2.1, Nano Banana 2, Nano Banana Pro, Nano Banana 2 LiteDefault: Seedream 5.5 | Drives the price, the available controls, and the input and output limits. |
| Aspect Ratio | Per-model list such as 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9 (Nano Banana 2.1 and Nano Banana 2 also add 1:4, 4:1, 1:8, 8:1) | The available set depends on the model. Nano Banana Pro offers Match input; the GPT Image models, Nano Banana 2.1, Nano Banana 2, and Nano Banana 2 Lite need an explicit ratio. Seedream 5.5 has no Aspect Ratio control: it keeps your photo's shape unless your instruction asks for another one (for example "make it a 16:9 banner"), and with no photo it reads the shape from your description. On Nano Banana Pro, with no photo to match, Default makes a square. Switching models may reset the ratio if the new model does not support it. |
| Output Resolution | 1K, 1.5K, 2K (Seedream 5.5); 1K, 2K, 4K (GPT Image models, Nano Banana 2.1, Nano Banana 2, Nano Banana Pro)Default: 2K | Shown only on models with resolution control. Nano Banana 2 Lite outputs at a fixed 1024px. Seedream 5.5 makes up to 2K, with no 4K. Higher resolution can raise the cost on some models. |
| Quality | Low, Medium, High, X-High, Max (GPT Image 2.6 and 2.5); Low, Medium, High (GPT Image 2)Default: High (Medium on GPT Image 2) | GPT Image models only. Higher quality means sharper detail and better prompt adherence at a higher cost. High is the balanced default. |
| Mode | Quality, FastDefault: Quality | GPT Image 2.6 and 2.5 only. Quality gives the most precise edits; Fast returns everyday edits sooner. Same credit cost either way. |
| Output Images | 1 up to the model's max (2 or 4) | How many variations to generate. Cost scales linearly with the number of outputs. |
Cost is per output image, multiplied by how many outputs you request. The Generate button always shows the live total first. Paid-user prices per output:
- Seedream 5.5: 0.5 at 1K, 1.5K or 2K, or 1 when you add 5 or more photos.
- GPT Image 2.6 and GPT Image 2.5 (quality times resolution, either mode): Low is 0.5 at any resolution; Medium is 0.5 at 1K, 1 at 2K or 4K; High is 1 at 1K, 1.5 at 2K, 2 at 4K; X-High is 1.5 at 1K, 2.5 at 2K, 3.5 at 4K; Max is 2 at 1K, 4 at 2K, 6 at 4K.
- GPT Image 2: Low is 0.5 at any resolution; Medium is 1 at 1K, 1.5 at 2K, 2 at 4K; High is 2 at 1K, 4 at 2K, 6 at 4K.
- Nano Banana 2.1: 0.5 at 1K or 2K, 1 at 4K.
- Nano Banana 2 and Nano Banana Pro: 1 at 1K or 2K, 2 at 4K.
- Nano Banana 2 Lite: a flat 0.5 per output.
Free users pay double these amounts and receive watermarked images.
On a paid plan, your monthly free images pay for Seedream 5.5 outputs first, one per image, so the Generate button shows Free while they last. Other models use credits.
Free vs paid
Free accounts pay roughly 2x the credits per output on every model.
Free outputs carry a watermark. Watermark-free exports are a paid-plan benefit.
On the GPT Image models, free plans can use Low quality at any resolution and Medium at 1K or 2K. High and above are paid-only and show with a lock, and 4K on a free plan needs Low quality.
See Plans & pricing to unlock the paid options.
Best practices
Give a specific, single-goal instruction. One clear change per run reads cleaner than several at once.
Upload multiple reference images on models that support it to combine elements or transfer a style.
Iterate cheaply on Seedream 5.5 (the default, 0.5 per output with up to 4 photos) or Nano Banana 2.1 (0.5 at 1K or 2K), then switch to GPT Image 2.6 when you need 4K or more control over quality.
Only raise Quality or Resolution once the composition is right, since both cost more.
Generate a few outputs and keep the best variation.
Troubleshooting
On a free plan the GPT Image models allow Low quality at any resolution and Medium at 1K or 2K; High and above are paid-only, and 4K needs Low. Upgrade to unlock them, or pick an unlocked combination. See Plans & pricing.
You have hit the selected model's input limit (4 on some models, 10 on others). Remove an image, or switch to a model with a higher input limit.
Each model supports a different ratio set, and some have no Match input option. Reselect a ratio the new model supports.
That is Seedream 5.5: it keeps your photo's shape and makes 1K, 1.5K or 2K. Ask for another shape in the instruction (for example "make it a 16:9 banner"), or switch to a model with an Aspect Ratio control. For 4K, switch to GPT Image 2.6, Nano Banana 2.1, Nano Banana 2 or Nano Banana Pro.
Watermarks apply when only free credits are used. Use a paid plan or purchased credits to remove them.
What to do next
Once your edit looks right: