---
title: "Seedream"
description: "Cinematic photos, natural-looking people, and layouts you place by coordinates, from ByteDance's models."
section: "Models"
url: "https://batchbanana.com/docs/models/seedream"
markdown_url: "https://batchbanana.com/docs/models/seedream.md"
last_verified: "2026-10-02"
---
# Seedream
Cinematic photos, natural-looking people, and layouts you place by coordinates, from ByteDance's models.
*Prompt: "Wildlife photograph: a flock of pink galahs bursting off a dead white gum tree at the left edge of the frame and streaming right at dusk, the low sun lighting their pink breasts against a black storm sky. 600 mm lens, wingtips blurred." Model: Seedream 5 Pro.*
## Best for
Seedream is ByteDance's image family. Use Seedream 4.5 for photographs, with realistic people, film looks, light and weather. Use Seedream 5 Pro for layouts, where you can place each thing with coordinates.
**Portrait**
*Prompt: "An environmental portrait of a chocolate maker in her sixties in her small workshop, seated by a copper melanger, cocoa sacks behind her, a tray of cobalt-wrapped bars on her knee. Window light from the left, 85 mm lens, natural skin with lines and freckles." Model: Seedream 4.5.*
**Poster laid out with coordinates**
*Prompt: "A poster for the Lantern Keep game launch. The title "LANTERN KEEP" in chunky stepped pixel letters, lantern gold, in 80 40 920 200. A pixel-art knight holding up a glowing lantern in 250 260 750 760. The line "Out now" in small white letters on a violet ribbon in 300 820 700 900. A ruined castle at dusk behind, pixel art." Model: Seedream 5 Pro.*
**Your product in a new scene**
*Prompt: "Reference target: the Halvard bar in Image 1, its wrapper unchanged. Scene: the bar on the dashboard of an old car parked by a fjord at night, green northern lights in the windscreen." Model: Seedream 5 Pro.*
**Edit by example**
*Prompt: "Image 1 changed to Image 2 by recolouring the wrapper. Apply the same change to the wrapper in Image 3. Keep everything else in Image 3 unchanged." Model: Seedream 5 Lite.*
For posters where every word matters, [Ideogram](https://batchbanana.com/docs/models/ideogram) is the safer pick. For vector files, use [Arrow](https://batchbanana.com/docs/models/arrow). [Which model for what](https://batchbanana.com/docs/models/which-model) shows all five families on the same prompts.
## Which version
| Version | Use it when |
| --------------- | ---------------------------------------------------------------------- |
| Seedream 4.5 | You want a photograph: portraits, film stills, products in real light. |
| Seedream 5 Lite | You're drafting, or copying an edit from one image to another. |
| Seedream 5 Pro | You need things in exact places, or a careful edit of one region. |
Start on 4.5, at 10 credits an image. The same prompt on each version:
*Seedream 4.5*
*Seedream 5 Lite*
*Seedream 5 Pro*
*Prompt for all 3: "An environmental portrait of a chocolate maker in her sixties in her small workshop, seated by a copper melanger, cocoa sacks behind her, a tray of cobalt-wrapped bars on her knee. Window light from the left, 85 mm lens, natural skin with lines and freckles."*
## Write the prompt
**Write sentences, not tags.** ByteDance's guide asks for the subject, what it's doing and where, then style, color, light and composition. Its example: "A girl in a lavish dress walking under a parasol along a tree-lined path, in the style of a Monet oil painting", not "Girl, umbrella, tree-lined street, oil painting texture." Tags still give you a picture, but sentences let you control the parts that tags leave open.
**Be plain, not ornate.** ByteDance says concise, precise prompts beat stacked fancy words. A pile of "breathtaking, ultra-detailed, epic" gave us a storm we never asked for and a motel named "Midnight Dunes". A plain description gave a quiet dusk with the first stars out, though it doubled the word on a second sign.
**Stacked adjectives**
*Prompt: "An absolutely breathtaking, stunningly gorgeous, ultra-detailed, award-winning, hyper-realistic, majestic, ethereal masterpiece photo of a beautiful neon motel sign on a dramatic, moody, epic, atmospheric desert highway at an incredible golden dusk." Model: Seedream 4.5.*
**Plain description**
*Prompt: "A neon motel sign on an empty desert highway at dusk, the first stars out, low warm light from the right, wide shot." Model: Seedream 4.5.*
**Say what the image is for.** "Design a logo for a gaming company" tells Seedream more than "an abstract image of a dog". ByteDance asks for the purpose and type up front.
**Put text in double quotes.** ByteDance's example: 'Generate a poster with the title "Seedream 4.5"', not 'Generate a poster titled Seedream 4.5'.
**Name the camera or the film.** A [published Seedream 5 guide](https://replicate.com/blog/how-to-prompt-seedream-5) recommends naming film stocks, lenses and lighting setups. The galahs at the top of this page came from "600 mm lens, wingtips blurred."
ByteDance recommends no more than 600 words, since a long prompt scatters the details. Most good prompts are far shorter.
## Place things with coordinates
On Seedream 5 Pro you can put an element in an exact box. Write `x1 y1 x2 y2` after it, where the numbers run from 0 at the top left to 999 at the bottom right. `x y` marks a single spot.
> The title "LANTERN KEEP" in chunky stepped pixel letters, lantern gold, in `80 40 920 200`.
ByteDance documents coordinates for edits. In our test they also worked from text alone. Without them, the same poster stacked the title in a corner, moved "Out now" up under it and pushed the knight to the side.
**Without coordinates**
*Prompt: "A poster for the Lantern Keep game launch. The title "LANTERN KEEP" in chunky stepped pixel letters, lantern gold. A pixel-art knight holding up a glowing lantern. The line "Out now" in small white letters on a violet ribbon. A ruined castle at dusk behind, pixel art." Model: Seedream 5 Pro.*
**With coordinates**
*Prompt: "A poster for the Lantern Keep game launch. The title "LANTERN KEEP" in chunky stepped pixel letters, lantern gold, in 80 40 920 200. A pixel-art knight holding up a glowing lantern in 250 260 750 760. The line "Out now" in small white letters on a violet ribbon in 300 820 700 900. A ruined castle at dusk behind, pixel art." Model: Seedream 5 Pro.*
## Edit and combine images
**Name the target and say what stays.** "Dress the tallest panda in pink Peking Opera costume and headgear, keeping its pose unchanged", not "Put that one in pink clothes." ByteDance warns against pronouns like "that one".
**Say what to take from each reference.** Number them: "Dress the character in Image 1 with the outfit from Image 2."
**Copy an edit from one image to another.** Give two images, before and after, and a third to change the same way. This is how the green Halvard wrapper in the row above was made, on Seedream 5 Lite, from a ferry photo before and after its recolour:
> Image 1 changed to Image 2 by recolouring the wrapper. Apply the same change to the wrapper in Image 3. Keep everything else in Image 3 unchanged.
**On 5 Pro, give exact colors as hex codes.** ByteDance's example changes pumpkins to "dark green (#3E4A2E) and turmeric yellow (#DB973E)".
**Mark the spot.** If a region is hard to describe, draw on the image before you attach it, a red box or a scribble, and refer to the mark: "Insert a TV where the red area is marked."
## Settings
**Resolution** is Seedream's only setting besides the aspect ratio. On 4.5, 4K costs the same 10 credits as 2K, and on 5 Lite, 3K costs the same as 2K, so pick the larger one. On 5 Pro, 1K costs 11 credits and 2K costs 22, and an image takes about 90 seconds, so draft at 1K.
| Model | Credits per image | Other settings |
| --- | ---: | --- |
| Seedream 5 Pro | 11 | Resolution: 2K 22 |
| Seedream 5 Lite | 10 | Same price |
| Seedream 4.5 | 10 | Same price |
## Limits
- Seedream 4.5 and 5 Lite take up to 14 reference images, Seedream 5 Pro up to 10.
- No Seedream version makes 5:4 or 4:5 images.
- Seedream 5 Pro tops out at 2K.
- ByteDance calls Seedream 5 Lite a relatively small model with room to improve in structure and realism. Use 4.5 for faces and photographs.
## Sources
ByteDance's own guides: [Seedream 4.0 to 4.5 prompt guide](https://docs.byteplus.com/en/docs/ModelArk/seedream-4-0-5-0-prompt-guide), [Seedream 5 Pro editing with coordinates](https://docs.byteplus.com/en/docs/ModelArk/2582775), and the [Seedream 5 Lite](https://seed.bytedance.com/en/blog/deeper-thinking-more-accurate-generation-introducing-seedream-5-0-lite) and [5 Pro](https://seed.bytedance.com/en/blog/beyond-generation-it-understands-design-introducing-seedream-5-0-pro) launch posts.
Images made in Batch Banana on October 2, 2026. Select Prompt on an image to read the prompt and model that made it. Where a run made more than one image, we show the one we liked best.
---
Last verified October 2, 2026. Web version: https://batchbanana.com/docs/models/seedream