Seedream 5.0 Lite Edit API

bytedance/seedream/v5/lite/edit
1–10 images · JPEG/PNG · 2K/3K

Seedream 5.0 Lite Edit transforms 1–10 reference images and natural language instructions into high-fidelity revisions and multi-reference composites, supporting mask-free editing, character identity preservation, and cross-scene blending. It accurately executes targeted modifications while preserving source facial features, core compositions, and natural lighting.

Get API Key
Input
Required0/3000
Required0/10
Optional
Optional
Optional
OutputIdle

Generated images will appear here

Estimated cost: —
Continue using

Examples

edit-01-output.png

Combine the three reference images into one natural documentary photograph. Use the elderly woman from image 1 with exactly the same face, wrinkles, eyes, braids tied with red yarn, black montera hat with the red band and magenta shawl with the silver pin. Use the backstrap loom and half-finished textile from image 2, keeping the same red, deep green and golden yellow diamond and zigzag pattern with the small white llama figures. Set the scene in the adobe courtyard from image 3 with its ochre walls, blue door, clay pot of geraniums and the terraces and snowy peaks beyond the low wall. The woman kneels on the straw mat in three-quarter view; the far end bar of the loom is tied to the wooden post, the near bar is attached to a strap around her waist, and her hands pass the weaving sword through the warp threads. Relight her and the loom with the same warm low afternoon sun from the left and matching long shadows, consistent scale and perspective, no text, no watermark.

edit-02-output.png

Extend image 1 into a wide 21:9 panoramic photograph. Keep the kayaker, the red kayak, the yellow paddle, the V-shaped wake and the turquoise-green water exactly as they are, with the same size, angle and relative position. Naturally expand the scene on both sides and upward: continue the steep dark rock cliff on the right to its full height and add a tall thin waterfall pouring down it into the fjord with mist at its base; on the left reveal a far shoreline with green meadows, three small red wooden boathouses at the waterline and layered mountain ridges fading into haze. Show the cliff tops and a soft overcast sky. Seamless continuation of light, reflections and water texture with no visible seams or borders, no text, no watermark.

edit-03-output.jpg

Redraw image 1 in the exact visual style of image 2. Image 1 defines all content: keep the same composition and framing, the two elderly players with their facial features, poses and clothing, the standing onlooker with folded arms, the chess table with the pieces in the same positions, the jacaranda trees and the three pigeons. Image 2 defines only the style: a two-color linocut relief print on off-white textured paper with bold carved black lines, visible gouge marks, parallel hatching for shading, flat rust-red ink for the shirts, the cardigan and the jacaranda blossoms, slightly uneven ink coverage and visible paper grain. Do not copy any boats, harbor, lighthouse or seagulls from image 2. No text, no signature, no watermark.

Seedream 5.0 Lite Edit

Seedream 5.0 Lite Edit is a lightweight, high-performance multimodal image editing and multi-reference blending model developed by the ByteDance Seed team. Sharing a unified reasoning architecture with its text-to-image counterpart, it enables precise natural language edits without manual brush masking (Mask-free), supporting local object modifications, micro-texture replacement, style transfers, and cross-image composition. The model accepts 1 to 10 ordered reference image URLs, allowing users to assign roles such as character identity, wardrobe, props, or background settings using clear "Figure 1" and "Figure 2" references in the prompt. Built on deep feature decoupling and 3D spatial alignment, it follows complex edit briefs faithfully while preserving source facial geometry, core perspective, and ambient lighting reflections across e-commerce, commercial retouching, and creative synthesis.

Why Choose Seedream 5.0 Lite Edit API?

  • 1–10 Reference Blending & Asset AttributionNatively supports 1 to 10 input reference images, allowing prompts to assign character portraits, outfits, items, and backgrounds across images with zero asset upload surcharges.

  • Mask-Free Natural Language Editing & Intent InferenceEliminates manual brush-painting and pixel selection masks, accurately inferring user intent from natural language instructions like removing background clutter or transforming lighting.

  • Subject Identity & Ambient Lighting PreservationDecouples facial likeness from background environments in deep feature space, locking facial geometry and ambient reflections even through dramatic scene or wardrobe changes.

  • 1–15 Batch Variations with Flexible SizingGenerates up to 15 distinct framing and modification variants in a single run across 2K/3K presets and eight aspect ratios, intelligently outpainting backgrounds to fit new formats.

Parameters

ParameterRequirementDescription
promptRequired

Describe the requested changes in 3–3,000 characters, referencing input images as "image 1", "image 2", etc. Leading and trailing whitespace is not counted.

image_urlsRequired

1–10 ordered HTTP(S) reference image URLs. Formats: JPEG, PNG.

sizeOptional

Output size: 2K, 3K, or an aspect ratio. For a custom size, enter WIDTHxHEIGHT (for example 2304x1728) or {"width":2304,"height":1728} with positive integers. Defaults to 1:1.

Default1:12K3K4:33:416:99:163:22:321:9
nOptional

Number of images to generate: integer 1–15. Defaults to 1.

Default123456789101112131415
enable_safety_checkerOptional

Set true to enable safety checking or false to turn it off. Defaults to true.

Defaulttrue

How to Use

  1. Assemble Reference Images and Assign RolesProvide 1 to 10 accessible JPEG or PNG image URLs, deciding which image supplies the main subject, outfit, props, or target background style.

  2. Write Natural Language Edit InstructionsDescribe the desired changes clearly, referencing images as "Figure 1" or "Figure 2" to indicate what to swap, modify, or retain from each asset.

  3. Choose Dimensions and Submit AsynchronouslySet the output size to 2K, 3K, an aspect ratio preset, or custom dimensions, specify n from 1 to 15, and retrieve finished images via task_id polling or callback_url.

Pricing

Each image costs 5 credits ($0.025). The total is 5 × n credits, where n is the number of generated images.

UsageRateDetails
Standard generation5 credits/image · $0.025/image5 × n credits · $0.025 × n
n = 1 / 5 / 10 / 155 / 25 / 50 / 75 credits$0.025 / $0.125 / $0.250 / $0.375

Best Use Cases

  • E-Commerce Model & Product Scene RelocationSeamlessly transfer apparel models or product shots into new studio backdrops or lifestyle settings without re-shooting.

  • Portrait Identity Retention & Stylized RetouchingChange hairstyles, expressions, and wardrobe or migrate subjects into cyberpunk, anime, or cinematic aesthetics while keeping facial likeness intact.

  • Multi-Asset Composition & Commercial Key ArtCombine characters, merchandise, and background elements from separate reference photos into a cohesive scene with unified perspective and lighting.

  • Batch Creative Exploration & Client ApprovalsGenerate 1 to 15 diverse retouching variants concurrently to present multiple visual treatments and expedite client sign-off.

Pro Tips

  • Refer to images clearly in prompts: Use "Figure 1", "Figure 2" (or "image 1", "image 2") corresponding to the array order, such as "Apply the clothing in Figure 2 to the person in Figure 1."
  • State both changes and preserved areas: Explicitly write "keep the facial features and hair color of the person in Figure 1, but replace the background with a sunlit beach."
  • Micro-surface material replacement: Detail the optical qualities of new materials (such as "convert the metallic bottle into frosted translucent sea glass") to trigger accurate light refractions.
  • Outpainting and ratio conversions: Convert horizontal photos into 9:16 vertical or 21:9 ultrawide assets by specifying the target size; the model naturally extends background surroundings.

Notes

  • Reference images required: This endpoint requires image_urls containing 1 to 10 accessible HTTP(S) URLs formatted as JPEG or PNG. For text-only generation, use Seedream 5.0 Lite Text to Image.
  • Zero input asset surcharge: Billing is strictly based on the number of successfully generated output images (5 credits each), regardless of whether you submit 1 or 10 reference images.
  • Automatic refund guarantee: Tasks pre-deduct 5 × n credits upon submission. In the event of validation errors, network issues, or timeouts, pre-deducted credits are automatically refunded in full.

Related Models

Seedream 5.0 Lite Edit API FAQ

What is the Seedream 5.0 Lite Edit API?

Seedream 5.0 Lite Edit is a ByteDance Seed model for high-fidelity image editing and multi-reference blending. It transforms 1–10 reference images and natural language instructions into mask-free revisions, cross-image asset composites, and character-consistent modifications, supporting concurrent batches of 1–15 variations. Built on a unified multimodal reasoning foundation, it executes targeted modifications while preserving source facial features, core compositions, and natural lighting. You can call it programmatically or try it from the playground above.

How many reference images does Seedream 5.0 Lite Edit support?

It supports 1 to 10 ordered HTTP(S) image URLs. All references serve as multimodal context, and you can assign roles to each image in the prompt using labels like "Figure 1" through "Figure 10" with zero extra cost for multiple references.

Do I need to paint a mask when using Seedream 5.0 Lite Edit?

No. The model operates entirely mask-free through natural language instructions. Simply describe the modifications and target aesthetics in text; the model automatically detects focal elements and executes seamless revisions.

How does Seedream 5.0 Lite Edit preserve facial identity across edits?

The model decouples facial likeness from surrounding attire and background in deep feature space. Adding explicit instructions such as "keep the facial features, expression, and identity of the person in Figure 1 unchanged" locks facial structure across diverse scene variations.

Does uploading multiple reference images increase costs in Seedream 5.0 Lite Edit?

No. Pricing is strictly calculated per generated output image at 5 credits ($0.025) each. Uploading between 1 and 10 reference images incurs no additional input asset fees.

Can Seedream 5.0 Lite Edit change dimensions or aspect ratios during an edit?

Yes. You can supply 2K, 3K, eight aspect ratio presets, or custom pixel dimensions in the size parameter; the model preserves core subjects while automatically outpainting and adapting background elements to the new ratio.

Can Seedream 5.0 Lite Edit produce multiple edit variations in one request?

Yes. By setting the integer n parameter between 1 and 15, you can generate up to 15 distinct editing and composition variants in a single call, billed only for successfully produced images.