Skip to main content

Guide · facts checked November 2026

How to Write Seedream 5 Pro Prompts (Guide + Examples)

Seedream 5 Pro reads plain, descriptive English. You get the best results by writing the scene the way you would brief a photographer — subject, setting, style, light — then quoting any text and choosing the canvas shape. This guide shows the formula, the special cases (text, references, edits) and examples you can copy.

Seedream 5 Pro at a glance

Made by
ByteDance Seed (ByteDance's AI research team)
Released
July 2026 (public launch); part of the Seedream 5.0 family with Lite and Flash
Resolution
1K, 1.5K and 2K tiers — about 2048×2048, or ~2752×1536 at 16:9. No native 4K.
Aspect ratios
Flexible, from tall portrait to wide landscape
Reference images
Up to 10 per request on major API hosts (fal, Replicate, Runware)
Output
One image per request (Lite can batch)
Strengths
Photoreal images, posters and infographics with in-image text, region-level editing guided by boxes, arrows or sketches
Where to use it
BytePlus ModelArk, Volcano Engine Ark, CapCut and Dola apps, and API hosts such as fal, Replicate, OpenRouter, Runware and ComfyUI
  1. Seedream 3.0 · April 2025
  2. Seedream 4.0 · September 2025
  3. Seedream 4.5 · December 2025
  4. Seedream 5.0 Lite · February 2026
  5. Seedream 5.0 Pro · July 2026

The prompt formula

Write it as connected sentences, most important first. Each layer below adds one kind of control.

Seedream prompt layers with an example
LayerExample
Subject + actionA barista pours latte art into a white cup
Environmentin a small concrete café, morning
Style / mediumeditorial lifestyle photograph
Light + lenssoft window light from the left, 50mm, shallow depth of field
Palette + moodwarm cream and walnut tones, calm
Text (optional)the cup sleeve reads “SLOW MORNING” in small serif capitals

Put together

A barista pours latte art into a white cup in a small concrete café in the morning. Editorial lifestyle photograph, soft window light from the left, 50mm lens with shallow depth of field. Warm cream and walnut tones, calm mood. The cup sleeve reads “SLOW MORNING” in small serif capitals.

Write one in five steps

  1. Start with the subject and action. One sentence that says who or what is in the frame and what is happening: “A ceramicist trims a tall vase on a spinning wheel.”
  2. Place it in an environment. Say where and when: a sunlit studio, a rainy street at night, a seamless grey backdrop.
  3. Add style and medium. Photograph, oil painting, 3D render, flat vector, editorial poster — pick one and say it plainly.
  4. Describe light, lens and palette. Name the light source, the framing (close-up, wide, 85mm) and two or three colours.
  5. Quote any text and set the shape. Put words to render in double quotes, say where they go, and choose the aspect ratio in your app.

Prefer clicking to typing? The Seedream prompt builder walks through the same five steps.

Getting text inside the image right

  • Put the exact words in double quotes: a neon sign that reads “OPEN LATE”.
  • Say where the text sits and how it looks — headline at the top, small caps on a label, hand-painted on a wall.
  • Keep it short. Headlines and labels render far more reliably than paragraphs of body copy.
  • Proofread every result. Small or long text can still contain wrong letters, so regenerate or fix it in an editor.

Examples: poster & typography prompts and infographic prompts.

Using reference images

Upload references (up to 10 on major API hosts) to keep a face, product or style consistent, then refer to each by its number: Put the person from image 1 in the jacket from image 2, standing in the street from image 3. Say what must stay identical and what may change.

Writing editing instructions

  • Lead with the action: replace, remove, add, recolour, extend.
  • Name what stays: Keep the face, pose and outfit unchanged.
  • Be specific about the new part: Replace the background with a foggy pine forest at dawn, matching the light direction.
  • Pro can follow boxes, arrows or sketches drawn on the image for region-level edits, where your app supports it.

More: image editing prompts.

Choosing an aspect ratio

1:1

Logos, icons, album covers, social posts

3:4 / 2:3

Portraits, posters, book covers

4:3

Product shots, general scenes

16:9

Landscapes, thumbnails, slides

9:16

Phone wallpapers, stories, reels

21:9

Cinematic banners and panoramas

Pro outputs up to about 2 megapixels per image (1K, 1.5K and 2K tiers), so the ratio decides the exact pixel size.

Common mistakes

Keyword soup
Strings like “masterpiece, best quality, 8k, trending” add little for natural-language models. Write a sentence instead.
Unquoted text
If the words are not in double quotes, the model may paraphrase or invent letters.
Too many subjects
Four characters each doing something different is hard to get right. Lead with one, add others after.
Vague edits
“Make it better” gives the model nothing to act on. Say what to change and what to keep.
Wrong canvas
A poster prompt on a 16:9 canvas will crop badly. Set the shape before you generate.

Examples to copy

Seedream 5 Pro FAQ

What is Seedream 5 Pro?

Seedream 5.0 Pro is an image generation and editing model from ByteDance Seed. It is built for photoreal images, text-heavy designs such as posters and infographics, and precise region-level editing.

When was Seedream 5.0 Pro released?

ByteDance launched it publicly in July 2026. Earlier versions were Seedream 3.0 (April 2025), 4.0 (September 2025), 4.5 (December 2025) and 5.0 Lite (February 2026).

What is the difference between Seedream 5.0 Pro and Lite?

Pro focuses on precise interactive editing, layer separation and dense layouts, and outputs up to 2K with one image per request. Lite goes up to 4K, can generate batches and offers optional web search.

What is the maximum resolution of Seedream 5.0 Pro?

It offers 1K, 1.5K and 2K tiers — roughly up to 2048×2048 pixels, or the same pixel count at other shapes (about 2752×1536 at 16:9). It has no native 4K output.

How many reference images can I use?

Up to 10 reference images per request on major providers such as fal, Replicate and Runware. Refer to them by number in the prompt, for example “the jacket from image 2”.

Can Seedream 5 Pro render text inside images?

Yes — in-image typography is one of its headline features, in many languages including English and Chinese. Put the exact words in double quotes. Small or long body text can still contain mistakes, so check every word.

Where can I use Seedream 5 Pro?

Through ByteDance's APIs on BytePlus ModelArk and Volcano Engine Ark, in ByteDance apps such as CapCut and Dola, and through third-party platforms like fal, Replicate, OpenRouter, Runware and ComfyUI. Availability varies by region and app version.

How should I write prompts for Seedream 5 Pro?

Describe the subject, action and environment in plain natural language, then add style, lighting and composition. Quote any text to render, number your reference images, set the aspect ratio, and keep the prompt under about 600 English words.

Is AltFTool affiliated with ByteDance?

No. AltFTool is an independent prompt library. Seedream is ByteDance's model; this site is not affiliated with, endorsed by or sponsored by ByteDance.

Sources

AltFTool is independent and not affiliated with ByteDance. Model details change; check the provider before relying on a spec.