Seedream 4 Edit Image to Image: Edit Photos with Text

Last verified: October 19, 2026

Seedream 4 Edit changes the photos you upload based on one plain-language instruction. It can work on a single picture, or take subjects, garments and backgrounds from several uploads and combine them into one finished image. Using Seedream 4 Edit image-to-image is more like briefing a retoucher than masking layers: you write what to add, remove, replace or restyle, and the model makes that change while keeping most of the rest of the frame as it was.

For single-image edits, Seedream 4 Edit is designed to balance four things that tend to pull against each other: following the instruction, staying consistent with the reference, keeping the image's structure intact, and editing text. You need that balance when a change has to work without the face, product or lettering around it shifting. It holds whether you're fixing one detail or moving a whole photo into a new style.

The whole edit goes in the Prompt box: what changes, what stays, and which upload each element comes from. Enhance Prompt turns a rough idea into a fuller instruction. Aspect Ratio and Resolution Quality set the frame and the level of detail of the finished image. Seedream 4 Edit generates high-resolution images natively, so pick one of the higher Resolution Quality settings for anything you plan to print or crop.

Explore more image to image

Image To Image Seededit V3 Nano Banana 2 GPT Image 2 Seedream 4.5 Edit Qwen Image 3.0 Qwen Image 3.0 Pro
Capability Snapshot

Seedream 4 Edit Inputs and Output Options

What the form accepts and what you get back from each run.

Input

Up to 8 images (JPG, JPEG, PNG or WEBP, up to 10 MB each) + a text prompt of up to 2,048 characters

Resolution

1K, 2K, 3K or 4K

Aspect Ratio

1:1, 3:4, 4:3, 9:16, 16:9

Output

1 image per run

Tips for Better Seedream 4 Edit Results

Name the Change, Then Name What Stays

A good edit prompt does two jobs. It states the change as a short, clear instruction, and it lists everything that has to stay as it is. In the Prompt box, write the change first, then add a keep clause that names specifics, for example "replace the overcast sky with a clear sunset; keep the building, the street and the people unchanged." Name every detail you'd notice if it shifted, such as a logo, a hairstyle or the camera angle, rather than hoping the model guesses which parts matter.

Quote the Exact Words for Sign and Poster Edits

When an edit involves lettering, put the new text in double quotation marks. The official prompt guide recommends this for any text that has to appear in the image. Then say what to keep. The creator's own example swaps a festival name and date on a poster while keeping the color scheme, font and alignment the same. Write the old and new wording in full, such as change "SALE" to "NEW SEASON", so it's clear which line you mean.

Relight a Room by Naming the Light Source and the Time of Day

For a lighting change, say what switches on and what stays the same. One of the creator's examples asks for the lights to be turned on in a living room and states that it's still evening outside: one sentence for the new light, one anchor for what the window should keep showing. Use the same pattern in the Prompt box. Name the source (desk lamp, neon sign, candles), say where it sits, and give the time of day that has to remain.

How to Edit a Photo with Seedream 4 Edit

Five steps from upload to download.

1

Add your photos in "Upload Images"

Start with the photo you want to change, then add any references the edit needs: a garment, a backdrop, a product shot or a style sample. Use the sharpest file you have, because everything the edit leaves alone comes straight from it. Keep track of the order you add references in, so your prompt can point to each one clearly.

2

Describe the edit in the "Prompt" box

Describe the change in up to 2,048 characters: what to add, remove, replace or restyle, which upload each new element comes from, and what has to stay as it is. If you're starting from a rough idea, the "Enhance Prompt" button under the box suggests fuller versions. Your text only changes if you pick one, so you can compare each suggestion with your own wording.

3

Set "Aspect Ratio" and "Resolution Quality"

"Aspect Ratio" offers 1:1, 3:4, 4:3, 9:16 and 16:9. Matching your source photo is the safest way to keep its composition; choose a different ratio only when you're reformatting on purpose, such as turning a landscape shot into a vertical story. "Resolution Quality" offers 1K, 2K, 3K and 4K. A lower setting is enough to check whether the edit worked, and 4K is the one to use for print, crops or the final file.

4

Click "Generate"

Before you click "Generate", read the prompt again next to your uploads. Every reference you mention should match a file you added, and every detail you need left alone should be named. Catching a mismatch now saves you a run.

5

Download the result

Each run returns one image. Check it at full size before downloading, especially faces, hands, lettering and the edges where the edited area meets the original, then save the file.

Sketch-Guided Edits: Seedream 4 Edit or a ControlNet Workflow?

How Seedream 4 Edit handles layout guides such as sketches and depth maps, compared with a ControlNet-based setup.

Guiding pose and layout with a sketch, edge map or depth map A sketch, Canny edge map, inpainting mask or depth map goes in as a regular upload, and the same model that reads your prompt uses it as guidance, including simple hand-drawn strokes. A ControlNet-based workflow handles this kind of visual guidance through specialized control models added to the image generator. Choose Seedream 4 Edit when you want to pair a rough drawing with a sentence without setting up extra models; stick with a ControlNet workflow if your existing pipeline is already built around those control models.
For edits guided by a sketch or map, Seedream 4 Edit handles the whole job in one model and one form. A ControlNet setup makes sense for teams that already have that pipeline in place.

Common Seedream 4 Edit Problems and How to Fix Them

What usually goes wrong with an edit, and what to change before you run it again.

My photo won't upload.

Upload Images checks every file against a short list of accepted formats and a minimum and maximum size, all listed in the upload question in the FAQ below. Phone photos saved as HEIC, or files in any other format outside that list, need converting first. Very large exports need resizing or recompressing to get under the file-size limit, and small crops or thumbnails below the minimum dimensions need replacing with a larger version of the same image. If you're adding several references, also make sure you haven't gone over the file limit. Once the files fit, upload them again; your prompt doesn't need to change.

The Prompt box won't take my whole instruction.

The box has a character limit, and long prompts usually hit it because they describe what's already in the photo. The model works from your uploads, so use the space for the change itself and for the details that must stay fixed. Short, clear edit instructions are also what the official prompt guide recommends. Trim the description of the existing scene, keep the change and the keep clause, then generate.

It edited the wrong person in my group photo.

Words like "him", "that one" or "the other guy" leave the model guessing which subject you mean, and the official prompt guide specifically advises against vague pronouns in edit instructions. Identify the target by what sets it apart, such as position, clothing or size: "the man in the grey suit on the far left." Rewrite that phrase in the Prompt box and run the edit again.

I want several variations, not just one image.

The form returns a single image per run, so you get variations by running it again. Keep the uploads and most of the prompt the same, and change one element per run, such as the style word, a color or the background. That keeps the set comparable and shows you what caused each difference. Run as many passes as the project needs.

My restyled portrait doesn't look like the same person anymore.

Restyling from a reference is harder to balance than a simple edit. What should be preserved could be a person's identity, a character design, an art style or a looser concept, and the model has to work out which one you mean. Take out the guesswork: say exactly what to carry over, for example "keep her face and hairstyle recognisable, change only the rendering to watercolor," and say what's free to change. Then generate again.

Frequently Asked Questions

Can Seedream 4 Edit colorize and repair an old photo?

Yes. Restoring an old print is one of the edits documented for the model, and a one-line instruction asking it to colorize the photo and repair the scratches covers both jobs. Upload the best scan you have. If you know the original colors of key details, such as a uniform, a car or a dress, state them so the colorization is based on facts rather than guesses.

Can it change the camera angle or framing of a shot?

Yes. The model handles perspective shifts as well as style changes while keeping the image's structure stable, and the creator's examples include reframing a subject from a close view to a medium view. Describe the new viewpoint in plain terms, such as top-down, eye level or medium shot, and name the subject features that must stay the same through the change.

What files can I upload to Seedream 4 Edit, and how many?

Upload Images accepts JPG, JPEG, PNG and WEBP files, each up to 10 MB and at least 300×300 pixels. You need at least one image and can add up to eight: the photo you want to edit, plus any references such as a garment, a background or a style example.

Do I need to mask or select the area I want to change?

No. Seedream 4 Edit makes high-quality edits from the prompt alone, so you define the area by describing it in the Prompt box: the left-hand window, the sign above the door, the dog on the sofa. The more precisely you locate the target in words, the less the model has to guess.

How is the cost of a run shown, and does Enhance Prompt use credits?

Once you're signed in, the credit cost of the run appears above the Generate button. It's calculated from the options you've chosen, so you see it before you start. Enhance Prompt doesn't use any credits.

References

Sources and citations used to support the content provided above.

Updated: 2026-10-19 23:36:29 6 Sources

arxiv.org

Source Link
https://arxiv.org/html/2509.20427v3

seed.bytedance.com

Source Link
https://seed.bytedance.com/en/blog/seedream-4-0-officially-released-beyond-drawing-into-imagination

docs.byteplus.com

Source Link
https://docs.byteplus.com/en/docs/ModelArk/1824718

arxiv.org

Source Link
https://arxiv.org/pdf/2509.20427

docs.byteplus.com

Source Link
https://docs.byteplus.com/ko/docs/ModelArk/1829186

seed.bytedance.com

Source Link
https://seed.bytedance.com/en/seedream4_0