Seedream - ByteDance AI Image Generator and Editor
Supports up to 5,000 characters. If you change the image count, mention the number inside the prompt as well.
Seedream returns all images in a single batch. Keep this slider aligned with the number referenced in your prompt.
Reuse the same seed to reproduce similar outputs.
Each Seedream request consumes 30 credits whether you generate or edit.
About Seedream V4
Seedream V4 is ByteDance's image model, and this page exposes both of its workflows: text to image, which renders a scene from a written prompt, and edit, which refines up to three images you upload. A run can return up to four images from the same prompt, at resolutions from 1K to 4K. The exact credit cost is shown on the generate button.
The two workflows are not two settings of one tool. Text to image builds a picture from nothing but the prompt, with an optional seed when you want a series to share a baseline. Edit takes source images and applies a written instruction to them, which keeps a real product, outfit or location in the result instead of asking the model to imagine it. In edit mode the page accepts up to three images, each under 10 MB and in PNG, JPG or WebP.
Seedream's useful difference is that one prompt can return a batch. Set the images-per-run slider between one and four and the model returns the whole set together, so you can compare variations without rewriting the prompt four times. Nine aspect presets cover square, portrait, landscape and ultrawide 21:9, and each can render at 1K, 2K or 4K. The prompt field holds up to 5,000 characters; if you raise the image count, say how many images you want inside the prompt as well.
Choose this page when you need several variations of one idea, a reproducible seed for a series, or a source image refined at 2K or 4K. Choose Nano Banana Pro when the job is a single keeper frame and you want up to five references guiding one result; choose the standard Nano Banana page while the direction is still undecided. Seedream sits between them: more output per run than either, with a written instruction driving every image in the batch.
The batch control belongs to both workflows, not just text to image. The images-per-run slider and the seed field stay in the shared panel when you switch to edit, so a single instruction against your source images can come back as several variations in one submission. That is the cheap way to compare framings of an edit: keep the same source set and prompt, raise the count, and judge four results side by side instead of paying for four separate runs. Start with two results rather than four: a pair is easier to judge, and the slider can go up once the instruction reads correctly.
Switching back to text to image clears the uploads, so there are no source images left when the mode changes. If you toggle modes to compare an idea midway through an edit, the references are gone and have to be uploaded again; finish or note the edit first. A source set can mix PNG, JPG and WebP in one run, and the same instruction is applied to every file in the batch. The seed field accepts digits and a leading minus sign only; letters and spaces are stripped as you type, so a pasted value needs checking before you submit. Reuse the integer when a series has to share a baseline.
How Seedream V4 works
- 1
Pick text to image or edit
Text to image needs only a prompt. Edit opens the uploader and requires at least one source image; you can add up to three, in PNG, JPG or WebP under 10 MB each.
- 2
Set the batch and the frame
Choose a resolution of 1K, 2K or 4K, one of nine aspect presets, and a batch size between one and four. Type a seed if you want to return to a similar result later, and mention the image count in the prompt.
- 3
Generate once, download the set
One submission returns the whole batch. Review the images in the results panel and download the ones you keep. If the task fails, the credits for that run are returned automatically.
What Seedream V4 is best for
Campaign variations
Describe one concept and receive up to four executions at once, then pick the composition that works before spending time on a final render.
Product photo edits
Upload a real product shot and name the change, such as a new surface or background, while the item itself stays recognisable.
Repeatable series
Keep a seed and a prompt structure fixed so a set of images shares a look across pages, posts or a catalogue.
Wide and large-format placements
Render a 21:9 banner at up to 4K for a site header or a screen, with the composition built for the frame rather than cropped into it.
Seedream V4 at a glance
- Model
- ByteDance Seedream V4, run through KIE
- Workflows
- Text to image and edit, with up to 3 source images in edit mode
- Credit cost
- The exact cost is shown on the generate button
- Resolution
- 1K by default, 2K and 4K available
- Aspect ratios
- Nine presets, from square to ultrawide 21:9
- Batch size
- Up to 4 images per run
Seedream V4 FAQ
- How many credits does a Seedream run cost?
- The exact cost is shown on the generate button, and one charge covers the whole run even when you ask for several images. Free accounts start with 300 bonus credits and can claim 50 more each day they log in; those credits expire after 30 days. Sign in before generating, because the request is tied to your account.
- What is the difference between text to image and edit?
- Text to image creates a new picture from the prompt alone. Edit takes one to three source images and applies your instruction to them, which is the mode for changing a real photo rather than inventing a scene. Edit will not run without at least one uploaded image.
- Can I generate more than one image at a time?
- Yes, up to four in a single run, and they arrive as one batch from the same prompt. Raising the count does not require separate submissions. Mention the number of images in the prompt as well, since the model reads both the slider and the text.
- How do I reproduce a similar image later?
- Enter the same numeric seed and reuse the prompt and settings. The seed makes the output stable enough to continue a series, though it does not promise a pixel-identical copy. Leaving the field blank gives each run a random seed.
- Which resolutions and aspect ratios are available?
- The page offers 1K, 2K and 4K, with 1K as the default. Nine presets cover two square sizes, three portrait ratios, three landscape ratios and ultrawide 21:9. The frame is chosen before generation, so the composition is built for that shape rather than cropped into it afterwards.
- Does the batch size apply to edit mode too?
- Yes. The images-per-run slider sits in the panel shared by both workflows, and the edit request sends it along with the source images, so one instruction can return up to four results from the same set. It is still one charge for the run. Keep in mind that switching back to text to image clears the uploaded images, so finish the edit or upload the set again after a mode switch.