Create with Doubao Seedream 4.5 using the modes and parameters currently available in the workspace.
Doubao Seedream 4.5 is suited to product visuals, characters, environments, campaign assets, social content, and early creative exploration.
Generated content will appear here
Facts last reviewed:
Doubao Seedream 4.5 is an image generation model from ByteDance and the 4.5 release in the Seedream image family. It is also written Seedream V4.5, and its task identifiers on this platform are doubao-seedream-4.5, doubao-seedream-4.5-i2i and doubao-seedream-4.5-r2i. On Nano Banana Pro the model runs inside the workspace at the top of this page, so you can write a brief, generate, adjust and generate again without a separate key or a local installation.
The workspace exposes the controls this platform currently supports for the model: eight aspect-ratio presets including auto, an output tier of 1K, 2K or 4K, and up to 14 reference images in one task. Before a request is submitted, the page shows the point cost for that exact combination, so the price of an experiment stays visible while the brief is still moving.
This page documents what is available here rather than what the vendor promises in general. Modes, tiers and reference ceilings follow the live catalog, and the workspace always shows the current state. Nano Banana Pro is an independent multi-model creation platform and is not operated by, affiliated with or endorsed by ByteDance.
Each item below is a control or a behaviour you can verify in the workspace, not a vendor benchmark.
Draft at 1K while the composition is still undecided, keep 2K as the working size, then raise the tier to 4K once a result has to survive a hard crop or go to print. The tier belongs to the request, so changing it is a click rather than a new project.
Long prompts can carry the words you want on the image: a headline, a product name, a small block of caption text. State the characters, the weight and the position, then proofread what comes back, because generated type can still swap, reorder or drop glyphs.
Attach files to steer the subject, the composition, the palette or the material. A ceiling of fourteen references makes it practical to hold a character sheet, a colour board and a product photo inside the same request instead of describing all three in words.
Let auto choose the frame, or pin a fixed ratio from 1:1, 2:3, 3:2, 4:3, 3:4, 16:9 and 9:16. Fixed ratios are what keep a series of images looking like one set when they will sit in the same layout.
Keep the brief, change one variable, run it again. Because the tier, the ratio and the reference set stay visible next to the prompt, you can tell which change actually moved the image.
The estimate follows the mode, the output tier and the number of reference images. Two directions at two different tiers can be compared in points before either one is generated.
The figures below come from the platform configuration for each model, not from vendor benchmarks, and they compare models you can switch between in the same workspace.
| Item | Doubao Seedream 4.5 | Nano Banana Pro | GPT Image 2 |
|---|---|---|---|
| Provider | ByteDance | OpenAI | |
| Aspect ratio presets | 8 presets with auto | 11 presets with auto | 16 presets with auto |
| Default output tier | 2K by default, selectable 1K or 4K | 1K by default, selectable 2K or 4K | 1K by default with low, medium or high render quality |
Each card below puts one job next to a Seedream 4.5 example published by the vendor, so the intended use and a finished result sit side by side.
Vendor-published exampleHold the product shape from a reference photo and change only the scene, the light or the background behind it.
Vendor-published exampleDraft the composition at 2K, then run the same brief again at 4K once the layout stops moving.
Vendor-published exampleKeep one character consistent across several views by attaching the same reference set to every request.
Vendor-published exampleExplore a location or a set before committing to a shoot, using auto framing for the first pass.
Vendor-published exampleCover several placements by re-running one brief at fixed ratios instead of re-composing each one by hand.
Vendor-published exampleHand over the current version plus a mood reference, then describe the single change you want made.
Six prompts, each paired with a Seedream 4.5 example the vendor published on its own model page. The prompts describe what the paired image shows, so read them as copy-ready starting points rather than as the exact inputs behind it.
Replaces only the lettering, leaving composition, framing and lighting untouched.
Rebuild the attached food photograph with every label translated into Chinese and set in a relaxed handwritten face. Keep the plate, the stacked ingredients, the green background and the arrow callouts exactly where they are, and keep each label legible and correctly written. Change nothing else in the frame.
Sets one large headline against a stack of small ingredient claims in the same frame.
Design a skincare key visual on a pink to white vertical gradient. Place one frosted glass jar in the lower half with a soft reflection beneath it. Set the product name on two lines across the top of the frame, then stack four short ingredient claims down the right side as small left-aligned blocks, each carrying a percentage. Keep every line inside the margin and add nothing else.
Holds a narrow palette, a single centred product and one quiet block of small type.
Build a vertical fragrance product page in deep green. Stand one rectangular flacon on a bed of pine needles and moss in the lower third, lit softly from the left. Set the product name, the category line and the three scent notes as separate centred lines in a light serif, and leave a wide empty band above them. Keep the palette to green, off-white and black, and add no other props.
Layers paper, a lace-cut edge and gold lettering for print-led briefs.
Compose a square mockup of a wedding invitation. Lay a lace-edged textured white card over a deep green envelope, then print the names, the date, an address and two short lines of copy in gold foil, centred and evenly spaced. Add an embossed green paper dome in the lower third and two hydrangea blooms at the lower corners. Shoot straight on with soft even light and a faint shadow under the card.
Keeps one layout and one palette intact while the frame runs from 1:1 to 9:16.
Treat the attached banner as the only source of style and elements. Rebuild it as six separate posters with the same collage look and the same title, changing only how the elements are arranged inside each frame. Use the ratios 1:1, 2:3, 4:3, 16:9, 1:2 and 9:16, keep the title legible in every version, and add no new motifs or colours.
Carries a character and an animal through one full scene inside a decorative border.
Illustrate a cover for a bedtime story. Place a small girl in a red cape and a fox at the centre of a dark forest clearing, in front of a cabin with glowing windows. Set a large blue moon in the upper half with drifting stardust, add fireflies above the grass and small white flowers at their feet, then wrap the scene in an ornate bronze storybook border. Keep both figures faithful to the attached references.
Select Doubao Seedream 4.5 in the workspace, then describe the subject, the framing, the light, the material and anything that has to stay out of the image.
Add only files you are allowed to upload, say what each one controls, then set the aspect ratio and the output tier of 1K, 2K or 4K.
Check the mode, the ratio, the tier and the reference count once more, and look at the point cost shown for that exact combination.
Check anatomy, lettering, structure and small details before the image is used, then change one variable and run it again.
This page publishes no fixed price for Doubao Seedream 4.5 and promises no free generations. The workspace shows a live point estimate for the current mode, output tier and reference count before the request is submitted, and that estimate is the number that applies.
Nano Banana Pro does not offer a public API for this model. It provides the online workspace on this page, and the request is sent to the provider through the platform integration. The provider documentation linked below describes the provider side, not an endpoint you can call from here.
The twelve examples on this page, and the six prompt cards built around them, come from the Seedream 4.5 model page published by ByteDance. They are demonstration material released by the vendor rather than results produced on this site, and each prompt describes what its paired example shows instead of reproducing the wording the vendor used. A few examples carry lettering in the language of the original case, and the caption and alt text give the wording in this page language.
Generated images can carry wrong or missing letters, odd anatomy, impossible object relationships and unintended resemblance to existing work. Review every result before it is published, and upload only material you have the right to use.
The provider links document the image generation service and the Seedream working group that publishes the model, and the twelve examples on this page were taken from the vendor model page linked below. Everything on this page about modes, tiers, reference ceilings and point costs describes the Nano Banana Pro workspace as configured today and can change with the live catalog. The facts above were last checked on 2026-09-11.
The platform flagship image model, with eleven aspect presets and a 1K default tier.
Another prompt-led workflow, with sixteen aspect presets and a render-quality control.
A faster iterative option for the stage when the brief is still moving.
Start from a file you are allowed to use and describe the change you want.
Write the brief, attach up to 14 reference images, choose the ratio and the tier, then confirm the live point estimate before you submit.
