Seedream 4.5

bytedance/seedream-4-5 — a image model by ByteDance. Call it through the gateway’s standard OpenAI-compatible door using the catalog slug; the full parameter schema is reproduced in full below.


Catalog slug	`bytedance/seedream-4-5`
Kind	`image`
Vendor	ByteDance

How to call

Through the Sociaro SDK / OpenAI-compatible door

from sociaro_ai import Sociaro

client = Sociaro(api_key="gw_live_...")

img = client.images.generate(
    model="bytedance/seedream-4-5",
    prompt="a red fox in the snow",
    # size / n / negative_prompt / seed … — see the schema below
)

Raw door: POST https://api.sociaro.com/v1/images/generations with {"model": "bytedance/seedream-4-5", "prompt": "…"}. See media doors.

Parameters — full schema

The complete, one-to-one API schema for this model, transcribed from the official source. Nothing omitted. Examples in the schema may use a representative model of the family — for this model use the slug bytedance/seedream-4-5 as shown in How to call above.

Coverage map (per-model capability surface)

Model (catalog slug)	In request/response reference	In per-model examples	Supports streaming
`seedream-5-0` (non-lite)	NOT named in any per-parameter “supported on” clause	YES — every Seedream-5.0 example uses the 5.0 slug	YES
`seedream-5-0-lite`	YES — primary “new” model in the “seedream-5-0-lite, 4-5 and 4-0” clauses	YES — lite is the headline 5.0 model	YES
`seedream-4-5`	YES	referenced in image-to-image examples	YES
`seedream-4-0`	YES	YES — used in a 4-0 example	YES
`seedream-3-0-t2i`	YES (size/seed/guidance_scale clauses)	no example	not supported (no streaming)
`seededit-3-0-i2i`	YES (size `adaptive`/seed/guidance_scale clauses)	no example	not supported (no streaming)
`seededit-3-0-t2i`	YES (image/output_format clauses)	no example	not supported (no streaming)

The four primary models for this family — seedream-5-0, seedream-5-0-lite, seedream-4-5, seedream-4-0 — share ONE request/response schema on the standard image-generations door. The 3-0/seededit models use the same schema with a reduced capability set; they are included here in full for completeness because their per-parameter behavior is what defines the “supported on” exclusions for the 4.x/5.x models.

Critical per-model differences (one-to-one with the model API)

Aspect	seedream-5-0 (non-lite)	seedream-5-0-lite	seedream-4-5	seedream-4-0	seedream-3-0-t2i	seededit-3-0-i2i	seededit-3-0-t2i
`image` (reference/edit input)	yes (single + multi 2–14)	yes (single + multi 2–14)	yes (single + multi 2–14)	yes (single + multi 2–14)	n/a (t2i)	yes, single only	NOT supported
`size` presets	2K/3K/4K (same shape as lite)	2K / 3K / 4K	2K / 4K	1K / 2K / 4K	pixels only, default 1024x1024	only `adaptive`	(size column shown for seededit-3-0-t2i but pixel/preset menu not separately enumerated — treat same pixel rules as the t2i group)
`size` pixel-method range (Method 2)	same as lite/4-5	[2560x1440 = 3,686,400 , 4096x4096 = 16,777,216], ratio [1/16,16]	[3,686,400 , 16,777,216], ratio [1/16,16]	[1280x720 = 921,600 , 16,777,216], ratio [1/16,16]	total px [512x512 , 2048x2048], default 1024x1024	`adaptive` (53-row lookup)	—
`seed`	NOT supported	NOT supported	NOT supported	NOT supported	supported (default -1)	supported (default -1)	not in clause
`guidance_scale`	NOT supported	NOT supported	NOT supported	NOT supported	supported, default 2.5	supported, default 5.5	not in clause
`sequential_image_generation` (+ `_options.max_images`)	yes	yes	yes	yes	no	no	no
`stream`	yes	yes	yes	yes	no	no	no
`output_format` (png/jpeg)	examples pass `png`; reference restricts custom format to lite only — AMBIGUOUS, flag for live verification	yes (png/jpeg, default jpeg)	no (jpeg fixed)	no (jpeg fixed)	n/a	no (jpeg fixed)	no (jpeg fixed)
`optimize_prompt_options.mode`	standard shape	standard only	standard only	standard + fast	no	no	no
`data[].size` in response	yes	yes	yes	yes	no	no	no
Streaming (SSE)	yes	yes	yes	yes	no	no	no

output_format on non-lite seedream-5-0 — documented contradiction. The request/response reference states “Only seedream-5-0-lite supports this parameter” and a tip that for seedream-4-5/4-0, seededit-3-0-i2i and seededit-3-0-t2i the format is jpeg-fixed (it does NOT mention non-lite 5-0). Yet every per-model example for seedream-5-0 passes "output_format":"png". The per-parameter “supported on” clauses never enumerate a bare non-lite seedream-5-0, so attributing the lite support matrix to non-lite 5-0 is an inference, not a documented fact. Treat non-lite-5-0 output_format support as ambiguous; flag for live verification.

optimize_prompt_options.mode — exact wording. Reference: “Only seedream-5-0-lite/4-5 (only supports standard mode) and seedream-4-0 support this parameter.” Examples: “seedream-4-0 allows you to set optimize_prompt_options.mode to fast … seedream-5-0-lite and seedream-4-5 focus on high-quality image generation and only [support standard mode].” So fast is valid on seedream-4-0 only; standard is the only mode on 5-0-lite and 4-5.

seed / guidance_scale are legacy. The reference explicitly limits both to seedream-3-0-t2i and seededit-3-0-i2i. They are NOT supported on any 4.x/5.x model. A transparent-passthrough gateway must not silently inject them for Seedream 4.x/5.x.

Endpoint, auth

Endpoint (standard image-generations door): POST https://api.sociaro.com/v1/images/generations Auth: header Authorization: Bearer $SOCIARO_API_KEY. Content-Type: application/json

Call a Seedream model by passing its catalog slug in the model field (e.g. bytedance/seedream-5-0). The gateway routes the request to ByteDance’s image API on your behalf.

Image generation capabilities by model (verbatim from reference)

seedream-5-0-lite (new), seedream-4-5, seedream-4-0:

Generate multiple images in sequence (a batch of related images) — set sequential_image_generation to auto.
- Batch from multiple reference images (2–14) + text prompt (input + output images ≤ 15).
- Batch (up to 14) from a single reference image + text prompt.
- Batch (up to 15) from text prompt.
Generate a single image — set sequential_image_generation to disabled.
- Single from multiple reference images (2–14) + text prompt.
- Single from a single reference image + text prompt.
- Single from text prompt.

seedream-3-0-t2i: Generate a single image from a text prompt.

seededit-3-0-i2i: Generate a single image from a single reference image + text prompt.

Request body shape

{
  "model": "bytedance/seedream-5-0",      // required — catalog slug
  "prompt": "string",                     // required — recommend < 600 English words
  "image": "url-or-base64" | ["url1","url2", ...],  // optional (i2i / multi-ref 2–14); seededit-3-0-i2i = single only; seededit-3-0-t2i = NOT supported
  "size": "2K" | "WIDTHxHEIGHT" | "adaptive",       // optional, preset OR explicit pixels OR (seededit-3-0-i2i) adaptive
  "seed": 12345,                          // optional — ONLY seedream-3-0-t2i / seededit-3-0-i2i; default -1
  "sequential_image_generation": "disabled" | "auto",   // optional, default "disabled"; ONLY 5-0-lite/4-5/4-0 (+5-0 per tutorial)
  "sequential_image_generation_options": {              // optional — only when sequential_image_generation = "auto"
    "max_images": 4                                      // int [1,15], default 15
  },
  "stream": false,                        // optional, default false; ONLY 5-0-lite/4-5/4-0 (+5-0 per tutorial)
  "guidance_scale": 2.5,                  // optional — ONLY seedream-3-0-t2i (def 2.5) / seededit-3-0-i2i (def 5.5); range [1,10]
  "output_format": "png" | "jpeg",        // optional, default "jpeg"; reference: ONLY seedream-5-0-lite
  "response_format": "url" | "b64_json",  // optional, default "url"
  "watermark": true,                      // optional, default true
  "optimize_prompt_options": {            // optional — 5-0-lite (standard only), 4-5 (standard only), 4-0 (standard+fast)
    "mode": "standard" | "fast"           // default "standard"
  }
}

Request parameters — full table

Name	Type	Required	Default	Allowed values / range	Supported on	Description
`model`	string	required	—	A valid catalog slug	all	The catalog slug used for image generation. e.g. `bytedance/seedream-5-0`, `bytedance/seedream-5-0-lite`, `bytedance/seedream-4-5`, `bytedance/seedream-4-0`.
`prompt`	string	required	—	free text; recommend < 600 English words	all	Text prompt for generation. Overly long prompts may scatter info and cause the model to overlook details / miss elements.
`image`	string OR array of strings	optional	—	URL or Base64 data URI	5-0/5-0-lite/4-5/4-0 = single OR multiple (2–14); seededit-3-0-i2i = single only; seededit-3-0-t2i = NOT supported	Reference/edit image(s). URL: must be publicly accessible. Base64: `data:image/<format>;base64,<encoding>`, `<format>` lowercase, e.g. `data:image/png;base64,...`. See input-image constraints below.
`size`	string	optional	pixel method default `2048x2048` (3-0-t2i default `1024x1024`); preset method has no single default	per-model preset/pixel rules (see Size section)	all (rules differ per model)	Output image dimensions. Two mutually-exclusive methods: Method 1 = resolution preset (model picks W/H from prompt-described aspect/shape); Method 2 = explicit `WIDTHxHEIGHT` px. For seededit-3-0-i2i only `adaptive`.
`seed`	integer	optional	`-1`	[-1, 2147483647]	ONLY seedream-3-0-t2i and seededit-3-0-i2i	Random seed controlling randomness. `-1`/unspecified = random → different outputs each call. Same seed + same request → similar (not guaranteed identical) results.
`sequential_image_generation`	string	optional	`disabled`	`auto`, `disabled`	ONLY seedream-5-0-lite, 4-5, 4-0 (and non-lite 5-0 per tutorial)	Whether to disable batch generation. `auto`: model decides whether/how many images to return based on the prompt. `disabled`: only one image generated.
`sequential_image_generation_options`	object	optional	—	object with `max_images`	ONLY seedream-5-0-lite, 4-5, 4-0 (and 5-0)	Config for batch generation; effective only when `sequential_image_generation = auto`.
`sequential_image_generation_options.max_images`	integer	optional	`15`	[1, 15]	as parent	Max images to generate this request. Actual count = jointly by `max_images` and number of input reference images: (input reference images + generated images) ≤ 15.
`stream`	boolean	optional	`false`	`true`, `false`	ONLY seedream-5-0-lite, 4-5, 4-0 (and 5-0)	Streaming output mode. `false`: all images returned at once. `true`: each image returned immediately after it is generated (works for both single & batch). See Streaming Response section.
`guidance_scale`	float	optional	seedream-3-0-t2i: `2.5`; seededit-3-0-i2i: `5.5`	[1, 10]	ONLY seedream-3-0-t2i and seededit-3-0-i2i; seedream-5-0-lite/4-5/4-0 NOT supported	How closely the image follows the prompt (controls creative freedom). Higher = less freedom, more adherence.
`output_format`	string (new)	optional	`jpeg`	`png`, `jpeg`	reference: ONLY seedream-5-0-lite	Output image file format. Tip: for seedream-4-5/4-0, seededit-3-0-i2i and seededit-3-0-t2i the default is jpeg and custom settings are not supported. (Non-lite seedream-5-0: tutorial curls pass `png` — ambiguous, flag for live verification.)
`response_format`	string	optional	`url`	`url`, `b64_json`	all	How images are returned. Generated image is JPEG. `url`: download link, valid 24 hours after generation. `b64_json`: Base64-encoded image string inside JSON.
`watermark`	boolean	optional	`true`	`true`, `false`	all	`true`: adds an “AI generated” watermark in the bottom-right corner. `false`: no watermark.
`optimize_prompt_options`	object	optional	—	object with `mode`	ONLY seedream-5-0-lite/4-5 (standard only) and seedream-4-0	Prompt optimization config.
`optimize_prompt_options.mode`	string	optional	`standard`	`standard`, `fast`	`fast` valid only on seedream-4-0; 5-0-lite & 4-5 = standard only	`standard`: higher quality, longer generation time. `fast`: faster, more average quality.

Input image constraints (`image` parameter, verbatim from reference)

Format: jpeg, png. seedream-5-0-lite, 4-5 and 4-0 (new) also support webp, bmp, tiff, gif, heic, heif.
Aspect ratio (width/height):
- within [1/16, 16] for seedream-5-0-lite, 4-5 and 4-0
- within [1/3, 3] for seededit-3-0-i2i and seededit-3-0-t2i
Width & height (px): each > 14.
Size: up to 30 MB.
Total pixels: no more than 6000×6000 = 36,000,000 (limit is the product W×H, not each dimension).
Max reference images: up to 14.

Batch failure-handling contract (part of the documented API surface)

When generating a batch with seedream-5-0-lite / 4-5 / 4-0, if an image fails:

Content-filter rejection: the next generation task is still requested; other tasks in the same request are unaffected.
Internal service error (500): the next image generation task is not requested. (This same contract is restated in the streaming image_generation.partial_failed event description.)

Size parameter — per-model detail (Method 1 presets + Method 2 pixel rules)

The reference renders size as per-model tabs. Each is reproduced exactly below.

seedream-5-0-lite (and, per tutorial, non-lite seedream-5-0)

Method 1 presets: 2K, 3K, 4K (describe aspect/shape/purpose in the prompt; model picks W/H).
Method 2 pixels: default 2048x2048; total-pixel range [2560x1440 = 3,686,400, 4096x4096 = 16,777,216]; aspect ratio range [1/16, 16]. Both constraints must hold simultaneously.
- Valid example: 3750x1250 (4,687,500 px ✓, ratio 3 ✓).
- Invalid example: 1500x1500 (2,250,000 px < 3,686,400 ✗) — fails pixel min even though ratio is fine.

Recommended W×H presets:

Resolution	Aspect ratio	W×H px
2K	1:1	2048x2048
2K	4:3	2304x1728
2K	3:4	1728x2304
2K	16:9	2848x1600
2K	9:16	1600x2848
2K	3:2	2496x1664
2K	2:3	1664x2496
2K	21:9	3136x1344
3K	1:1	3072x3072
3K	4:3	3456x2592
3K	3:4	2592x3456
3K	16:9	4096x2304
3K	9:16	2304x4096
3K	2:3	2496x3744
3K	3:2	3744x2496
3K	21:9	4704x2016
4K	1:1	4096x4096
4K	3:4	3520x4704
4K	4:3	4704x3520
4K	16:9	5504x3040
4K	9:16	3040x5504
4K	2:3	3328x4992
4K	3:2	4992x3328
4K	21:9	6240x2656

seedream-4-5

Method 1 presets: 2K, 4K (no 3K).
Method 2 pixels: default 2048x2048; total-pixel range [3,686,400, 16,777,216]; aspect ratio [1/16, 16]. Both must hold (same valid/invalid examples as 5-0-lite).

Recommended W×H presets (2K and 4K rows identical to the 5-0-lite table; there is no 3K tier):

Resolution	Aspect ratio	W×H px
2K	1:1	2048x2048
2K	4:3	2304x1728
2K	3:4	1728x2304
2K	16:9	2848x1600
2K	9:16	1600x2848
2K	3:2	2496x1664
2K	2:3	1664x2496
2K	21:9	3136x1344
4K	1:1	4096x4096
4K	3:4	3520x4704
4K	4:3	4704x3520
4K	16:9	5504x3040
4K	9:16	3040x5504
4K	2:3	3328x4992
4K	3:2	4992x3328
4K	21:9	6240x2656

seedream-4-0

Method 1 presets: 1K, 2K, 4K.
Method 2 pixels: default 2048x2048; total-pixel range [1280x720 = 921,600, 4096x4096 = 16,777,216] (lower minimum than 5-0-lite/4-5); aspect ratio [1/16, 16].
- Valid example: 1600x600 (960,000 px ✓, ratio 8/3 ✓).
- Invalid example: 800x800 (640,000 px < 921,600 ✗).

Recommended W×H presets:

Resolution	Aspect ratio	W×H px
1K	1:1	1024x1024
1K	4:3	864x1152
1K	3:4	1152x864
1K	16:9	1280x720
1K	9:16	720x1280
1K	3:2	832x1248
1K	2:3	1248x832
1K	21:9	1512x648
2K	1:1	2048x2048
2K	4:3	2304x1728
2K	3:4	1728x2304
2K	16:9	2848x1600
2K	9:16	1600x2848
2K	3:2	2496x1664
2K	2:3	1664x2496
2K	21:9	3136x1344
4K	1:1	4096x4096
4K	3:4	3520x4704
4K	4:3	4704x3520
4K	16:9	5504x3040
4K	9:16	3040x5504
4K	2:3	3328x4992
4K	3:2	4992x3328
4K	21:9	6240x2656

seedream-3-0-t2i (pixels only; legacy)

No presets. Set W×H in pixels. Default 1024x1024. Total-pixel value range [512x512, 2048x2048].

Aspect ratio	W×H px
1:1	1024x1024
4:3	1152x864
3:4	864x1152
16:9	1280x720
9:16	720x1280
3:2	1248x832
2:3	832x1248
21:9	1512x648

seededit-3-0-i2i (`adaptive` only; legacy)

Only adaptive is supported. The system compares the input image’s dimensions to the preset table below and selects the first available aspect ratio with the smallest difference from the original image’s ratio.

Full preset lookup (Width/Height ratio → Width × Height px), 53 rows verbatim:

Width/Height	Width	Height
0.33	512	1536
0.35	544	1536
0.38	576	1536
0.4	608	1536
0.42	640	1536
0.47	640	1376
0.51	672	1312
0.55	704	1280
0.56	736	1312
0.6	768	1280
0.63	768	1216
0.66	800	1216
0.67	832	1248
0.7	832	1184
0.72	832	1152
0.75	864	1152
0.78	896	1152
0.82	896	1088
0.85	928	1088
0.88	960	1088
0.91	992	1088
0.94	1024	1088
0.97	1024	1056
1	1024	1024
1.06	1056	992
1.1	1088	992
1.17	1120	960
1.24	1152	928
1.29	1152	896
1.33	1152	864
1.42	1184	832
1.46	1216	832
1.5	1248	832
1.56	1248	800
1.62	1248	768
1.67	1280	768
1.74	1280	736
1.82	1280	704
1.78	1312	736
1.86	1312	704
1.95	1312	672
2	1344	672
2.05	1376	672
2.1	1408	672
2.2	1408	640
2.25	1440	640
2.3	1472	640
2.35	1504	640
2.4	1536	640
2.53	1536	608
2.67	1536	576
2.82	1536	544
3	1536	512

Response shape (non-streaming)

{
  "model": "bytedance/seedream-5-0",   // catalog slug used
  "created": 1757323224,            // unix seconds, request creation time
  "data": [                         // array of output image objects (or error objects)
    {
      "url": "https://...",         // present when response_format=url (24h expiry)
      "b64_json": "...",            // present when response_format=b64_json
      "size": "1760x2368"           // <width>x<height>; only 5-0-lite/4-5/4-0 (+5-0)
    }
    // OR, for a failed image in a batch:
    // { "error": { "code": "...", "message": "..." } }
  ],
  "usage": {
    "generated_images": 1,          // # successfully generated (billing basis)
    "output_tokens": 16280,         // round( sum(W*H) / 256 )
    "total_tokens": 16280           // == output_tokens (input tokens not counted)
  },
  "error": {                        // top-level request error, if any
    "code": "...",
    "message": "..."
  }
}

Response parameters — table

Field	Type	Description
`model`	string	Model ID used (model name-version).
`created`	integer	Unix timestamp (seconds) of the request creation time.
`data`	array	Output images. Each element is either an Image-info object or an Error-message object. Batch failure rule: content-filter rejection → next task still requested; internal 500 → next task NOT requested.
`data[].url`	string	Image download URL; present when `response_format=url`. Expires within 24h of generation — save before expiry.
`data[].b64_json`	string	Base64 image data; present when `response_format=b64_json`.
`data[].size`	string	`<width>x<height>` px, e.g. `2048x2048`. Only seedream-5-0-lite/4-5/4-0 (and 5-0) return this.
`data[].error`	object	Error object for a failed image (within `data`).
`data[].error.code`	string	Error code for the failed image. See Error Codes.
`data[].error.message`	string	Error message for the failed image.
`usage`	object	Usage info for the request.
`usage.generated_images`	integer	# images successfully generated (excludes failures). Billing is per successfully generated image.
`usage.output_tokens`	integer	Tokens consumed by generated images = round( sum(image_width × image_height) / 256 ). These are IMAGE tokens, not text tokens.
`usage.total_tokens`	integer	Total tokens for the request; equals `output_tokens` (input tokens not currently counted).
`error`	object	Top-level request error (if the whole request failed).
`error.code`	string	Error code. See Error Codes.
`error.message`	string	Error message.

Streaming response (`stream=true`) — SSE event schema

Supported only on seedream-5-0-lite, seedream-4-5, seedream-4-0 (and non-lite 5-0 per tutorial). When stream=true, the server pushes events via Server-Sent Events (SSE) as the response is generated. There are three event types plus a top-level error.

Event `image_generation.partial_succeeded`

Returned when any single image is successfully generated.

Field	Type	Description
`type`	string	Always `image_generation.partial_succeeded`.
`model`	string	Model ID used (`<model_name>-<version>`).
`created`	integer	Unix timestamp (seconds) of the request creation time.
`image_index`	integer	Index of the image for this event. Starts at 0 and auto-increments by 1 for BOTH `partial_succeeded` and `partial_failed` events, regardless of success.
`url`	string	Image download URL; returned when `response_format=url`.
`b64_json`	string	Base64 image; returned when `response_format=b64_json`.
`size`	string	`<width>×<height>` px, e.g. `2048×2048`.

Example:

{
  "type": "image_generation.partial_succeeded",
  "model": "bytedance/seedream-5-0",
  "created": 1589478378,
  "image_index": 0,
  "url": "https://...",
  "size": "2048×2048"
}

Event `image_generation.partial_failed`

Returned when any image fails to generate. Same failure contract: content-filter rejection → next task still requested; internal 500 → next task NOT requested.

Field	Type	Description
`type`	string	Always `image_generation.partial_failed`.
`model`	string	Model ID used (`<model_name>-<version>`).
`created`	integer	Unix timestamp (seconds) of the request creation time.
`image_index`	integer	Index of the image for this event. Starts at 0; auto-increments for both partial events.
`error`	object	Error reason for this event.
`error.code`	string	Error code. See Error Codes.
`error.message`	string	Error message.

Example:

{
  "type": "image_generation.partial_failed",
  "model": "bytedance/seedream-5-0",
  "created": 1589478378,
  "image_index": 2,
  "error": {
    "code": "OutputImageSensitiveContentDetected",
    "message": "The request failed because the output image may contain sensitive information."
  }
}

Event `image_generation.completed`

Final event in the stream; returned after all requested images (successful or failed) have been processed.

Field	Type	Description
`type`	string	Always `image_generation.completed`.
`model`	string	Model ID used (`<model_name>-<version>`).
`created`	integer	Unix timestamp (seconds) of the request creation time.
`usage`	object	Usage info for the request.
`usage.generated_images`	integer	# successfully generated images (excludes failures). Billed only for successes.
`usage.output_tokens`	integer	round( sum(picture_height × picture_width) / 256 ).
`usage.total_tokens`	integer	Total tokens; equals `output_tokens` (input tokens not counted).

Example:

{
  "type": "image_generation.completed",
  "model": "bytedance/seedream-5-0",
  "created": 1589478378,
  "usage": {
    "generated_images": 2,
    "output_tokens": 16280,
    "total_tokens": 16280
  }
}

Top-level streaming `error`

Returned for the current request if a request-level error occurs.

Field	Type	Description
`error`	object	Error message for this request.
`error.code`	string	Error code. See Error Codes.
`error.message`	string	Error message.

Example:

"error": {
  "code": "BadRequest",
  "message": "The request failed because it is missing one or multiple required parameters. Request ID: {id}"
}

Examples

Minimal — text-to-image

Request:

curl https://api.sociaro.com/v1/images/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SOCIARO_API_KEY" \
  -d '{
    "model": "bytedance/seedream-5-0",
    "prompt": "Vibrant close-up editorial portrait, model with piercing gaze, wearing a sculptural hat, Vogue cover aesthetic, dramatic studio lighting."
  }'

Response:

{
  "model": "bytedance/seedream-5-0",
  "created": 1757323224,
  "data": [ { "url": "https://...", "size": "2048x2048" } ],
  "usage": { "generated_images": 1, "output_tokens": 16384, "total_tokens": 16384 }
}

Full A — single image, common knobs (verbatim tutorial example, seedream-5.0)

curl https://api.sociaro.com/v1/images/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SOCIARO_API_KEY" \
  -d '{
    "model": "bytedance/seedream-5-0",
    "prompt": "Vibrant close-up editorial portrait, model with piercing gaze, wearing a sculptural hat, rich color blocking, sharp focus on eyes, shallow depth of field, Vogue magazine cover aesthetic, shot on medium format, dramatic studio lighting.",
    "size": "2K",
    "output_format": "png",
    "watermark": false
  }'

Response:

{
  "model": "bytedance/seedream-5-0",
  "created": 1757323224,
  "data": [ { "url": "https://...", "size": "1760x2368" } ],
  "usage": { "generated_images": 1, "output_tokens": 16280, "total_tokens": 16280 }
}

Full B — image-to-image (single reference, verbatim tutorial)

curl https://api.sociaro.com/v1/images/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SOCIARO_API_KEY" \
  -d '{
    "model": "bytedance/seedream-5-0",
    "prompt": "Keep the model'\''s pose and the flowing shape of the liquid dress unchanged. Change the clothing material from silver metal to completely transparent clear water (or glass). Through the liquid water, the model'\''s skin details are visible. Lighting changes from reflection to refraction.",
    "image": "https://example.com/assets/seedream4_5_imageToimage.png",
    "size": "2K",
    "output_format": "png",
    "watermark": false
  }'

Full C — multi-reference image-to-image, single output (verbatim tutorial)

curl https://api.sociaro.com/v1/images/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SOCIARO_API_KEY" \
  -d '{
    "model": "bytedance/seedream-5-0",
    "prompt": "Replace the clothing in image 1 with the outfit from image 2.",
    "image": ["https://example.com/assets/seedream4_imagesToimage_1.png", "https://example.com/assets/seedream4_5_imagesToimage_2.png"],
    "sequential_image_generation": "disabled",
    "size": "2K",
    "output_format": "png",
    "watermark": false
  }'

Full D — batch / sequential output from a single reference image (verbatim tutorial)

curl https://api.sociaro.com/v1/images/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SOCIARO_API_KEY" \
  -d '{
    "model": "bytedance/seedream-5-0",
    "prompt": "Using this LOGO as a reference, create a visual design system for an outdoor sports brand named GREEN, including packaging bags, hats, cards, lanyards, etc. Main visual tone is green, with a fun, simple, and modern style.",
    "image": "https://example.com/assets/seedream4_imageToimages.png",
    "size": "2K",
    "sequential_image_generation": "auto",
    "sequential_image_generation_options": { "max_images": 4 },
    "stream": false,
    "output_format": "png",
    "response_format": "url",
    "watermark": false
  }'

Full E — batch from multiple references (verbatim tutorial)

curl https://api.sociaro.com/v1/images/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SOCIARO_API_KEY" \
  -d '{
    "model": "bytedance/seedream-5-0",
    "prompt": "Generate 3 images of a girl and a cow plushie happily riding a roller coaster in an amusement park, depicting morning, noon, and night.",
    "image": ["https://example.com/assets/seedream4_imagesToimages_1.png", "https://example.com/assets/seedream4_imagesToimages_2.png"],
    "sequential_image_generation": "auto",
    "sequential_image_generation_options": { "max_images": 3 },
    "size": "2K",
    "output_format": "png",
    "watermark": false
  }'

Full F — seedream-4-0 with optimize_prompt fast mode (note `fast` mode, which is 4-0 only)

curl https://api.sociaro.com/v1/images/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SOCIARO_API_KEY" \
  -d '{
    "model": "bytedance/seedream-4-0",
    "prompt": "Generate a series of 4 coherent illustrations focusing on the same corner of a courtyard across the four seasons, presented in a unified style that captures the unique colors, elements, and atmosphere of each season.",
    "size": "2K",
    "sequential_image_generation": "auto",
    "sequential_image_generation_options": { "max_images": 4 },
    "optimize_prompt_options": { "mode": "fast" },
    "stream": false,
    "output_format": "png",
    "response_format": "url",
    "watermark": false
  }'

Gateway-relevant notes

Billing: per usage.generated_images (successful only); output_tokens = round(sum(W*H)/256), total_tokens == output_tokens. These are image tokens, not text tokens.
Output URL expiry: 24h. Persist before expiry.
Do not inject seed/guidance_scale for Seedream 4.x/5.x — they are legacy params for seedream-3-0-t2i / seededit-3-0-i2i only and are unsupported on the 4.x/5.x models.
output_format on non-lite seedream-5-0: reference vs. example contradiction (jpeg-fixed per reference clause vs. png in every example). Flag for live verification before relying on it.
optimize_prompt_options.mode=fast: valid on seedream-4-0 only; 5-0-lite and 4-5 are standard-only.

Seedream 4.5

How to call

Through the Sociaro SDK / OpenAI-compatible door

Parameters — full schema

Coverage map (per-model capability surface)

Critical per-model differences (one-to-one with the model API)

Endpoint, auth

Image generation capabilities by model (verbatim from reference)

Request body shape

Request parameters — full table

Input image constraints (image parameter, verbatim from reference)

Batch failure-handling contract (part of the documented API surface)

Size parameter — per-model detail (Method 1 presets + Method 2 pixel rules)

seedream-5-0-lite (and, per tutorial, non-lite seedream-5-0)

seedream-4-5

seedream-4-0

seedream-3-0-t2i (pixels only; legacy)

seededit-3-0-i2i (adaptive only; legacy)

Response shape (non-streaming)

Response parameters — table

Streaming response (stream=true) — SSE event schema

Event image_generation.partial_succeeded

Event image_generation.partial_failed

Event image_generation.completed

Top-level streaming error

Examples

Minimal — text-to-image

Full A — single image, common knobs (verbatim tutorial example, seedream-5.0)

Full B — image-to-image (single reference, verbatim tutorial)

Full C — multi-reference image-to-image, single output (verbatim tutorial)

Full D — batch / sequential output from a single reference image (verbatim tutorial)

Full E — batch from multiple references (verbatim tutorial)

Full F — seedream-4-0 with optimize_prompt fast mode (note fast mode, which is 4-0 only)

Gateway-relevant notes

Input image constraints (`image` parameter, verbatim from reference)

seededit-3-0-i2i (`adaptive` only; legacy)

Streaming response (`stream=true`) — SSE event schema

Event `image_generation.partial_succeeded`

Event `image_generation.partial_failed`

Event `image_generation.completed`

Top-level streaming `error`

Full F — seedream-4-0 with optimize_prompt fast mode (note `fast` mode, which is 4-0 only)