Notice on the Launch of Hy-Image-3.5-preview

hy-image-v3.5-preview is now available, and the image generation credits from the Mini Program Growth Plan can be deducted on it directly. The model submits image generation requests over the Chat/Messages protocol and returns results synchronously. It supports mixed text and reference images, multi-turn editing, and up to 4096x4096 (4K) output.

Capabilities

Capability	Description
Multimodal input	Mix text and reference images in messages; images can be public URLs or Base64, up to 20 MB each and 20 per request
Multi-turn editing	Pass the previous response's assembled_history into the next round to keep editing the image
Output size	size width and height ∈ [256, 8192], area no larger than 4096x4096; 4K output requires an explicit size
Reproducible results	A fixed seed with the same prompt and size reproduces the same result
Custom watermark	footnote adds a watermark in the bottom-right corner, up to 16 characters

How to Call

import cloudbase from "@cloudbase/js-sdk";

const app = cloudbase.init({ env: "your-env-id" });
const imageModel = app.ai().createImageModel("cloudbase");

const res = await imageModel.generateImage({
  model: "hy-image-v3.5-preview",
  messages: [
    { role: "user", content: [{ type: "text", text: "A golden retriever running on a beach at sunset" }] },
  ],
});

// Image URLs are valid for 12 hours, so transfer them promptly
const { url, width, height } = res.choices[0].delta.image;

For the full parameter list, response structure, and error codes, see Hy-Image-3.5-preview.

Differences from the Legacy Models

Item	HY-Image-3.0-Plus-4090-Tob-v1.0 HY-Image-v3.0-I2I-ToB-v1.0.1	hy-image-v3.5-preview
Model instance	createImageModel('hunyuan-image')	createImageModel('cloudbase')
Request parameters	prompt, size, image_urls	messages (mixed text and reference images)
Response structure	res.data[0].url	res.choices[0].delta.image.url
Multi-turn editing	Not supported	Pass assembled_history back
Maximum output	Determined by the size enum	4096x4096 (4K)

The two calling conventions are incompatible, so they cannot switch automatically and you must adapt your code manually.

The legacy models will be taken offline at 00:00 on December 1, 2026, and calls will fail after that time. See Notice on the Offline of Hunyuan Image Generation 3.0 Models.

Getting Help

If you run into issues during adaptation, visit the CloudBase Community or submit a support ticket for 1-on-1 assistance.