Skip to main content

Seedream 5.0 Flash image editing takes one to ten reference images (JPEG, PNG or WebP, up to 30 MB each) and returns one image. Revise a single photo or combine a face, garment and place; choose from fifteen aspect ratios at 1K to 2K.

From $0.029/image

Workbench

Input

Image editing
269/5000

Reordering updates prompt references. Removing media also removes its references; review your prompt.

Up to 5,000 characters. Assign a role to every reference and say what must stay unchanged; unnamed sources get blended. No effect on price. at least 1 characters · at most 5000 characters
3/10

Drop files or media links, paste, or browse · Up to 10 files · Max 10 MB · JPG · PNG · WEBP

  • Image 1
  • Image 2
  • Image 3
3 / 10 files
One to ten reference images (jpeg, png or webp, up to 30 MB each). Order is meaningful: the prompt refers to them as first, second and third. You must hold the rights to every image. The price is per output image and does not grow with the number of references. at least 1 items · at most 10 items
One of fifteen shapes from 1:3 to 3:1, including 4:5, 21:9 and 9:21. Name it when the placement dictates the frame. No effect on price.
1k, 1.5k or 2k; 2k by default. All three tiers cost the same per image, so the default is the sharpest.
Output format
JPEGPNG
Flash exports JPEG for review or PNG for further compositing at the same image price across its three resolution tiers.
Background
OpaqueTransparent
To composite a Flash edit over another background, choose `transparent` with PNG output and exactly one source PNG that already has alpha; other combinations are rejected, while `opaque` is the normal default.
Safety checkerOff

SpicyAPI does not run a safety checker on this model. Your prompt and media go to the model as written, and the model itself decides what comes back.

Checking your account before enabling submission and pricing.

Output

Ready for a new run
three-refs-balcony-dressExample result

Related models and next tasks

Missing the model you need?

Keep the person, change the scene, from up to ten references

Seedream 5.0 Flash's editing endpoint reads one to ten reference images and an instruction, then returns one image. In our tests it kept a face, a pose and the props fixed while changing the outfit completely, moved a rooftop portrait into a rainy neon alley, and built one woman on a Paris balcony out of three separate photos. aspect_ratio defaults to auto, so the result keeps the first reference's shape unless you name one of fifteen ratios.

Open the playground
Keep the person, change the scene, from up to ten references
References
1 to 10 image_urls — jpeg, png or webp, up to 30 MB each
Aspect ratio
auto (first reference's shape) or 1 of 15 fixed ratios
Resolution
1k, 1.5k or 2k (default 2k), same price at every tier
Prompt length
1 to 5,000 characters
Pricing
Flat per output image, whatever the reference count
Content policy
No content filter, no maturity flag
Hold what matters
Name what must not moveThe lace-to-silk test listed face, hairstyle, pose, cup and bed as fixed, and all five survived while the garment and the light changed. Anything you leave unprotected may shift, because every edit renders the whole frame again.
Roles by position
Every reference needs a stated jobThe prompt refers to images as first, second and third in image_urls order. A source with no job still gets blended in, so a ten-image edit without roles averages into a compromise.
Flat per edit
Ten references cost the same as oneThe price is per output image and does not grow with the number of sources, so there is no reason to trim a set below what the edit actually needs.

What the Seedream 5.0 Flash editing API gives you

Change the garment, keep the moment

Change the garment, keep the moment

Send one photo, list what stays and describe the new wardrobe. In our test the outfit changed completely while the face, hairstyle, pose, cup and setting stayed put, and the morning light turned into slanting afternoon sun as asked.

Move a subject into a different world

Move a subject into a different world

The same man, the same open white shirt, the same lean on a rail: the rooftop at dusk became a narrow alley in the rain, the shirt soaked through, pink BAR and teal OPEN LATE signs reflected on the wet ground and on his skin. Describe the new setting and its light, and name the pose as fixed.

  • Face, outfit and place from three photos

    One reference per ingredient, each named by position: the face and hair from the first, a backless red satin gown from the second, a Paris balcony from the third. The result was one woman alone at the wrought-iron rail in that gown, looking back over her shoulder, with warm light from the window behind her.

  • Framing that follows your first image

    aspect_ratio defaults to auto and keeps the shape of the first reference, so a portrait source stays a portrait. Put the image whose framing matters first, or name one of the fifteen ratios when the placement needs a different shape.

  • Wardrobe edits without a switch

    A Mild-level wardrobe swap, relighting a portrait or editing an existing uncensored image runs on this endpoint with no maturity flag and no alternate route. SpicyAPI adds no content filter; the model's own limits still apply, and you need the rights to every image you send.

  • Edit from the original, not the last output

    There is no seed, and each pass re-renders the whole frame, so a chain of edits drifts further from the source every time. Put several changes into one instruction, or go back to the original image for the next variation.

Where Seedream 5.0 Flash editing fits

  • Restyle a shoot without booking another

    Wardrobe and styling

    Restyle a shoot without booking another

    Keep the approved frame and change the clothing, the season or the time of day. Protect face, pose and props by name, and describe the new fabric and light precisely; vague wardrobe words come back as generic outfits.

  • Assemble a scene from separate photos

    Composites

    Assemble a scene from separate photos

    A face from a portrait, a gown from a product shot, a location from a travel photo. Give each image one job in the prompt, in array order, and say whose light the finished frame should take so the pieces read as one exposure.

  • One face across a whole set

    Character series

    One face across a whole set

    Generate a strong portrait once, then reuse it as the first reference in every later edit while the wardrobe and setting change. An identity carries across a set far more reliably this way than by describing the same face again in each prompt.

  • New light on a finished frame

    Relighting

    New light on a finished frame

    Ask for a new source, direction and colour of light and keep the subject fixed. The rooftop-to-alley test shows how far it goes: neon reflected on wet skin and pavement, with the pose and the shirt unchanged.

How to call the Seedream 5.0 Flash editing API

  1. Order the references

    image_urls takes one to ten jpeg, png or webp URLs of up to 30 MB each. Put the image whose face or framing matters first; auto framing follows it.

  2. Give each image one job

    Write something like "Image 1 is the face to keep, image 2 supplies only the gown, image 3 is the location", then describe the result. Unnamed sources get blended.

  3. List what must not change

    Face, hairstyle, pose, props, camera angle: whatever you do not protect may move, because the whole frame is rendered again.

  4. Submit and poll for a terminal state

    Edits in our tests took 35 to 48 seconds. Poll the task until it finishes, then fetch the image from platform storage.

  5. Branch from the source

    Start the next variation from the original references again, not from the previous output, so the drift does not compound.

Seedream 5.0 Flash editing worked examples

  • A wardrobe swap with five things held

    Everything that must survive is listed before the change, and the change of light is stated separately.

    Keep her face, hairstyle, pose, the ceramic cup and the unmade bed exactly as they are. Replace the ivory lace set and linen trousers with a loosely tied black silk robe. Change the light from soft morning light to warm, slanting afternoon sun through the curtains.
  • The same man in a new world

    The pose and the shirt are named as fixed, and the new light is described by its colours and where it reflects.

    Keep the same man, the same open white linen shirt and the same pose leaning on the rail. Move him into a narrow alley at night in heavy rain, the shirt soaked against his chest, a pink neon sign reading "BAR" and a teal sign reading "OPEN LATE" reflected on the wet ground and on his skin.
  • Three references, three jobs

    Roles are assigned by position before the scene is described; this is the brief behind the balcony result.

    Image 1 is the face and hairstyle to keep. Image 2 supplies only the backless red satin gown. Image 3 is the location. Show the same woman from image 1 alone at the wrought-iron railing of image 3, wearing the gown from image 2, looking back over her bare shoulder at the camera. Warm window light behind her.

Seedream 5.0 Flash editing questions

Put the reference whose face or framing matters first, because auto framing follows it. Give every image one job in the prompt and list what must not change; anything unprotected may move. Put several changes into one instruction, and branch new variations from the original, since chained edits drift.