ToMoviee Image Models - Reference to Image |Skydome Creation Engine

1. Overview

1.1 Service Capabilities

ToMoviee Reference to Image starts with a reference image and enables controlled creative variation while faithfully preserving core visual traits, especially the precise characteristics of subjects or scenes. This feature excels in maintaining subject identity (ensuring consistent facial ID), providing precise body pose control, and finely adjusting sketch details and depth of field (DOF). Unlike basic style transfer, it goes beyond mere imitation to deliver highly realistic image regeneration. By retaining key structural features like character pose and outline, it produces results so vivid that it feels as if the subject is present in front of you. Moreover, this fine control over subject sketch and DOF grants users highly controllable “semi-free creative” flexibility, making it particularly suited for scenarios demanding extreme accuracy, such as product mock-ups.

1.2 Sample Prompts and Outputs

Reference Image	prompt	Output Image
	Subject to Image: A man stands with crossed arms in the Grand Canyon
	Pose to Image: A boy playing basketball on a Japanese anime-style court with sakura petals falling by the sideline
	Sketch to Image: Fantasy fairy-tale style, fruits in fluorescent pink/blue, hazy lighting, a squirrel like a forest spirit
	Depth to Image: Flat illustration of an ancient Chinese pagoda, clean color blocks, bright tones

2. Prompt Engine

Prompt = Reference Description + Preserved Elements + Edit or Add Instructions

The core of Reference to Image lies in modifying or regenerating an image using a reference image. Compared to Text to Image, the prompt framework should also include descriptions of the reference image and clearly specify which elements should be retained and which need modification.
Reference Description: 1–2 sentences summarizing the reference image’s subject, action, and scene.
Preserved Elements: Specify which details should be retained in the output.
Edit or Add Instructions: Clearly describe what should be modified or added.

3. API Requests

3.1 Request URL

https://open-api.wondershare.cc/v1/open/capacity/application/tm_reference_img2img

3.2 Request Parameters

Method： POST

Headers

Parameter Name	Value	Required	Example	Description
Content-Type	application/json	Yes
Authorization		Yes	Basic xxx	Security verification information, in the format of Basic {access_token}, where access_token is a token, generated using the given app_key and app_crit, with the generation method being base64 (app_key: app_crit)
X-App-Key		Yes		Assigned appkey

body:

Parameter Name	Type	Required	Default Value	Description	Other Info
callback	string	No		Callback URL for algorithm results. For example, http://www.wondershare.cn/fx.
params	string	No		透明参数
prompt	string	Yes		Positive prompt.
width	integer	Yes		Width of the output image. Unit: pixels.
height	integer	Yes		Height of the output image. Unit: pixels.
batch_size	integer	Yes		Number of images to generate. Valid values: 1–8.
control_intensity	number	Yes		Control intensity, which is a floating-point number. Valid values: 0–1.
control_type	string	Yes		Control type. 0: outline (sketch) control; 1: pose control; 2: subject control; 3: DOF control.
init_image	string	No		URL of the original image. Image format: JPG or PNG. This field is required when control_type is 2 (subject control).
reference_image	string	Yes		URL of the reference image processed by the model. Image format: JPG or PNG.

3.3 Response

Parameter Name	Type	Required	Default Value	Description	Other Info
code	number	Yes		Error code.
msg	string	Yes		Error message.
data	object	No			Notes: