ByteDance text-to-image

Seedream 3.0 AI Image Generator

Seedream 3.0 is ByteDance’s bilingual text-to-image model for photoreal portraits, product stills, and short Chinese–English typography. On RenderFlow AI it runs at 4 credits per image, with custom sizes from 512 to 2048 pixels and a Guidance Scale control (0–20) that Seedream V4 and Seedream 5.0 do not expose here.

4 credits per image 512–2048 px Guidance Scale 0–20 Seed

Generate with Seedream 3.0 Need a larger canvas? Seedream V4

Seedream 3.0 model input

Prompt

Model

Bytedance

4 credits per image.

ByteDance text-to-image model for photoreal portraits, product stills, and short Chinese–English text. 4 credits per image, with width, height, Guidance Scale, and Seed.

Size

Width

Height

Guidance Scale

Seed

Your generated image will appear here.

Enter a prompt and click 'Generate Image' — or start from an example:

The canvas takes a prompt, width and height from 512 to 2048 pixels (default 1024), Guidance Scale from 0 to 20 (default 3, step 0.1), and a Seed (default -1). Each image costs 4 credits. There is no output-format menu on this model.

Examples

Seedream 3.0 examples

These stills are Seedream 3.0 outputs already published for this model: a crystal portrait, a product splash, a bilingual-style poster, an editorial portrait, graffiti lettering, and a macro. Open any tile to read the prompt.

Portrait

Translucent crystal figure

Prompt: A beautiful transparent woman, porcelain, translucent, crystallized, icy, magical, micro, photography.

Product

Lemon water splash

Prompt: bottle and lemon slices, splashing effect, refreshing summer, water droplet effect.

Poster

Fried-chicken poster

Prompt: A Korean-style fried chicken poster with the title "Crunch, Crunch – Let's Eat Chicken"

Editorial portrait

1990s-style portrait

Prompt: A young beautiful Hong Kong woman. Short - Mid hair, turning her head to the left looking at the camera. 90s Hong Kong style, dark tone

Typography

Graffiti lettering

Prompt: graffiti font, colorful stroke of graffiti on a Brooklyn wall with words "Seedream 3.0 on RenderFlow AI"

Macro

Praying mantis macro

Prompt: macro lens shot of a mantis's upper body, Teal and orange tone

What is Seedream 3.0?

What is ByteDance Seedream 3.0?

ByteDance Seedream 3.0 is a Chinese–English image generation model from ByteDance’s Seed team, announced on April 16, 2025. Official materials highlight native high-resolution output up to about 2K, stronger small-text and poster layout, and improved portrait fidelity versus Seedream 2.0 (ByteDance Seed, technical report).

Seedream 3.0 makes still images. A search for “seedream video” points at a different ByteDance product. Clips with sound on RenderFlow AI are Seedance 2.0, not this model.

Seedream 3.0 specs and controls

Seedream 3.0 AI image generator settings

Seedream 3.0 controls available in the RenderFlow AI canvas
Control	On this page
Credits	4 per image
Width	512–2048 pixels, default 1024
Height	512–2048 pixels, default 1024
Guidance Scale	0–20, step 0.1, default 3
Seed	Integer from -1 to 2,147,483,647, default -1
Output format	No format picker on this model

Credit price is the live catalog price, not a dollar rate. Plan details are on the pricing page.

How to use Seedream 3.0

How to use Seedream 3.0 for digital art

The controls that change the picture are the prompt, the two size fields, Guidance Scale, and Seed. This is the sequence that matches those inputs.

1. Write the subject, light, and frame

Name who or what is in the picture, then the light and the lens. “Editorial portrait, short black bob, looking back over the shoulder, dark 1990s street” is a complete job. Add materials and a background only when they change the result.
2. Quote short text for posters

Put the exact words in quotes and say where they sit: a headline across the top, one short line under it. Seedream 3.0 is a bilingual Chinese–English model, and short strings are the case it handles. Long paragraphs and many separate lines are where letters break.
3. Start Guidance Scale around 2–6

The slider runs from 0 to 20 in steps of 0.1 and defaults to 3. Raise it when every stated detail has to appear. Lower it when you want a looser interpretation. For photoreal portraits and product stills, start between 2 and 6 and move one step at a time.
4. Pin the Seed to iterate

Seed defaults to -1. When a composition is close, replace -1 with a fixed integer and change one thing: the scale, a color, or a line of text. A fixed seed lets you compare prompts instead of rolling a new picture every time.
5. Move up when 2048 pixels is the limit

This model stops at 2048 pixels on each side. Seedream V4 costs 5 credits, documents size support up to 4K, and does not include Guidance Scale. Use Qwen Image when a layout needs more lines of Chinese and English, and SeedEdit V3 or Seedream V4 Edit when you are changing a photo you already have.

Which Seedream 3.0 prompt details to write first

Start a Seedream 3.0 prompt with the subject, the light, and the frame, then add only the materials that change the picture. Put poster copy in quotes, name where each short line sits, and keep the string to a headline plus one supporting line. That is the pattern behind the fried-chicken poster and the graffiti tile above.

Seedream 3.0 or Seedream V4?

Stay here for the 4-credit price and the Guidance Scale dial. Move to Seedream V4 when the frame has to be larger than 2048 pixels or the layout carries more type. V4 is 5 credits per image. Its size field defaults to 2048×2048 and is described as supporting up to 4K. It does not expose Guidance Scale.

Editing a photo you already have is a different model. SeedEdit V3 repairs a picture from one sentence at 4 credits, with a 0–1 guidance control. Seedream V4 Edit is the poster editor: up to four references and output from 1024 to 4096 pixels, at 5 credits. Qwen Image and Qwen Image Edit are the closer fit when the words themselves are the design.

Seedream 3.0 FAQ

What is Seedream 3.0 AI used for here?

Seedream 3.0 AI on this page is for photoreal portraits, product stills, and short Chinese or English text. The examples above were generated from the prompts printed on each card, including a 1990s-style portrait and a translucent crystal figure. It does not make video. Frames larger than 2048 pixels per side belong on Seedream V4.

How should a Seedream AI image generator prompt be written?

Name the subject, the light, and the frame, then put exact words in quotes and keep those strings short. The fried-chicken poster on this page keeps the large English headline readable, while the smaller lines are less reliable. Start Guidance Scale between 2 and 6 (the slider runs from 0 to 20 and defaults to 3). Leave Seed at -1 for a new result, or set a fixed integer to repeat a composition. For dense multi-line Chinese and English, compare Qwen Image or Seedream V4.

Seedream 3 or Seedream V4 on RenderFlow?

People searching Seedream 3 usually mean this 3.0 generator. Stay here for 4 credits per image and a Guidance Scale from 0 to 20. Move to Seedream V4 for a larger canvas or heavier layout work, at 5 credits and without that slider. Photo repair is a different job: SeedEdit V3 for one-sentence fixes, or Seedream V4 Edit when the frame is a poster.

What does the Seedream-3 model id refer to?

Seedream-3 is this page’s URL, /model/seedream-3. The catalog id the canvas sends is bytedance/seedream-v3, which is RenderFlow’s name for the Seedream 3.0 generator at 4 credits per image.

Is there a Seedream 3.0 API on RenderFlow?

No. This page does not offer a public Seedream 3.0 API. You generate in the browser canvas, and the id it sends is bytedance/seedream-v3 at 4 credits per image. BytePlus documents a separate API example, model id seedream-3-0-t2i-250415, and that id is not what this canvas sends (BytePlus).

What sizes can this Seedream image generator output?

Width and height on this Seedream image generator are each set from 512 to 2048 pixels, and both default to 1024. There is no separate file-format control. Seedream V4 is the step up when you need a larger frame: its size field defaults to 2048×2048 and is described as supporting up to 4K, at 5 credits per image.

Ready to generate with Seedream 3.0?

Open the canvas above, set the size, and leave Guidance Scale at 3 until a test tells you to move it.

Back to the Seedream 3.0 canvas