Seedream 4.5: Guide, Prompts and Examples

Seedream 4.5 is ByteDance’s image model on VdoBloom, and it costs a flat 7 credits per image for both of its modes: text to image for generating from a description, and edit for changing an existing photo. It outputs at 2K on the basic setting or 4K on high, across eight aspect ratios, and the edit mode accepts up to 10 input images in a single instruction.

Updated October 2026.

Disclosure: VdoBloom is our platform.

What Seedream 4.5 is

Seedream 4.5 is ByteDance’s image model, and it sits in the same family tree as the Seedance video models on the platform — useful context, because the two share a habit of following written instructions closely rather than interpreting them loosely. On VdoBloom’s image generator it appears twice, as Seedream 4.5 Text-to-Image and Seedream 4.5 Edit, and both are priced identically.

Its two selling points are resolution and instruction-following. The basic quality setting renders at 2K and the high setting at 4K, so a single generation can already be large enough for print work or for a hero image that will be cropped. The edit mode is the more interesting half: it takes a written instruction plus one or more input images and applies the change, which is the workflow that replaces swatch-by-swatch retouching.

It carries the platform’s HOT content label, the most permissive filtering tier, and VdoBloom’s model picker scores it 4.5.

What it is good at

Clean, finished-looking images. Compositions come back plausibly lit and well framed, which makes it a good default when you want a usable image rather than a sketch of one.
Precise written instructions. Tell it to change a jacket to red leather and keep the face identical, and both halves of that instruction carry weight.
Multiple input images. Edit mode accepts up to 10 images in one request, which covers most composition, product-on-model and merge-photo jobs.
Wide or tall framing. The eight ratios include 2:3, 3:2 and 21:9 on top of the usual square, portrait and landscape options.
Large output without a second tool. Choosing high quality gives 4K in the same run rather than sending the image to an upscaler afterwards.

Credits and specifications

Setting	Seedream 4.5 Text-to-Image	Seedream 4.5 Edit
Cost per image	7 credits	7 credits
Quality tiers	Basic (2K) or High (4K)	Basic (2K) or High (4K)
Input images	None (prompt only)	Up to 10
Aspect ratios	1:1, 4:3, 3:4, 16:9, 9:16, 2:3, 3:2, 21:9	Same eight
Prompt limit	3,000 characters	3,000 characters
File types	—	JPEG, PNG or WebP, up to 10 MB each

The price is flat, which is worth pausing on: high quality at 4K costs the same 7 credits as the 2K basic render. There is no resolution surcharge to weigh up, so the only reason not to reach for 4K is generation time. New accounts start with 10 free credits and no card, and the exact cost is shown on the button before every generation. Pricing covers plans and the one-time credit packs that never expire.

Six example prompts

Product, clean studio. “A frosted glass skincare bottle on a pale sand-coloured plinth, soft key light from the upper left, a single hard shadow falling to the right, muted beige background, plenty of empty space above the bottle for text. Photorealistic, 85mm product photography.”
Portrait with a real face. “Editorial portrait of a woman in a tan trench coat against a plain concrete wall, overcast daylight, natural skin with visible texture, hair slightly windblown, calm expression, 50mm, shallow depth of field.”
Layout with readable text. “A minimal cafe menu board on a dark green painted wall. The heading reads OPEN and below it three lines in a simple sans-serif. Chalk texture, matte finish, straight-on framing, even light.”
Wide banner. “An empty coastal road at dusk shot in 21:9, low sun behind low hills on the right, long empty asphalt leading to the horizon, cool blue shadows, no people, no text.”
Instruction-based edit. “Change the jumper to deep red knitwear and the background wall to warm off-white. Keep the person’s face, hair, pose and lighting direction exactly as in the photo.”
Multi-image composition. “Place the product from the first image on the wooden table in the second image, at the same camera angle and light as the table shot. Match the shadows under the product. Do not change the table or the room.”

Two habits make the difference on edits. Write the instruction as a change, not as a description of the finished picture, and always name what must stay the same — “keep the face, pose and lighting identical” is the clause that stops an edit from quietly reshooting the whole frame. The prompt structure guide covers that ordering in more depth.

Limits worth knowing

Prompt length: 3,000 characters in each mode. Long, but shorter than some models in the roster, so a sprawling brief may need trimming.
Input images: 10 in edit mode, at up to 10 MB each. There is no multi-reference generation mode — text-to-image takes no images at all.
No negative prompt field on this model, so anything you want excluded has to be phrased as a positive instruction that occupies the space.
No seed control exposed in the interface, so an exact rerun is not something you can dial in.
Resolution is a two-step choice, not a free number: basic or high, 2K or 4K.

Seedream 4.5 or another model?

Model	Credits per image	Reach for it when
Seedream 4.5	7	You want 2K or 4K output and up to 10 reference images with strong instruction-following
Seedream 5 Lite	6	You want the newer Seedream generation cheaper, or up to 14 reference images
Nano Banana Pro	18 at 1K and 2K, 24 at 4K	You need the largest output tier and are happy to pay for it
FLUX Kontext Max	10	A precise local change to a single reference image

If your job is text-heavy design work rather than photography, the platform’s comparison of models for text inside images is a faster route to the right one than trying each in turn.

Frequently asked questions

How much does Seedream 4.5 cost on VdoBloom?

7 credits per image, flat, for both text-to-image and edit, at either quality tier. High quality at 4K therefore costs the same as basic at 2K, and the cost is displayed before you generate.

What is the difference between Seedream 4.5 and Seedream 4.5 Edit?

Only the input. Text-to-Image generates from a written description with no images supplied. Edit takes your instruction plus up to 10 input images and changes them, which is the mode for swapping backgrounds, changing clothing, or combining several photos into one composition.

Can Seedream 4.5 really output 4K?

Yes, on the high quality setting. Basic renders at 2K. Both are billed at the same 7 credits, so the choice is about render time rather than price.

Is there a negative prompt?

Not on this model. Only some image models in the roster expose a negative prompt field, and Seedream 4.5 is not one of them. If something keeps appearing in your results, describe the frame without it — naming the empty space you want works better than naming the thing you do not want.

Why did my edit change the whole photo?

Almost always because the instruction described a finished picture rather than a change, or because nothing in the prompt told the model what to preserve. Add an explicit keep clause covering the face, pose, composition and lighting direction, and keep the instruction focused on the one thing you are changing.

Open the image generator, pick Seedream 4.5, and write your prompt with the change in the first sentence and the keep clause in the second. Ten free credits at signup covers a 7-credit 4K render with a little left over, so the first test costs nothing. If you are still choosing between models, Seedream 4.5 Edit is documented in full, and the prompt guide will get you to a usable first attempt faster than trial and error.