Hy Image 3.5 Preview API

Hy Image 3.5 Preview is an image generation and multimodal image editing model developed by Tencent (Hunyuan series), released on the iCreat platform on 2026-11-09. It generates high-quality images from natural language instructions and optional reference images, and accepts a conversational multimodal parameter format built around a messages array.

On the iCreat platform, Hy Image 3.5 Preview is served through the asynchronous endpoint https://api.icreat.ai/v1/task/submit/tencent/hy-image-3.5-preview, with results retrieved from https://api.icreat.ai/v1/task/result. It works in a two-step asynchronous mode (submit a task, then poll for the result using the same task_id). Requests use a flat top-level body where size and messages sit at the same level, with no input or parameters wrapper. Output is billed per image by resolution tier: $0.024 per image at 1K, $0.024 per image at 2K, and $0.032 per image at 4K.

Model Positioning

Hy Image 3.5 Preview targets fast, low-cost, high-quality image generation and reference-guided image editing. It is positioned as a Preview release of the Hunyuan image line, offering 24 preset output sizes spanning 1K, 2K, and 4K tiers across eight aspect ratios, plus support for custom aspect ratios.

Its conversational messages structure lets developers pass a text prompt together with an optional reference image URL in a single request, making text-to-image and image-to-image workflows share one unified call pattern. With 1K and 2K priced identically at $0.024 per image, the model is aimed at high-volume generation and rapid iteration where cost per image matters.

Core Capabilities

Text-to-Image Generation

Hy Image 3.5 Preview generates images directly from natural language prompts supplied as text content blocks inside the messages array. Prompts describe the desired scene, subject, style, and composition, and the model returns a finished image at the requested output size.

Reference-Guided Image Editing

By adding an image_url content block to the messages array, developers can pass a publicly accessible reference image alongside the prompt. This enables image-to-image generation and reference-guided editing, where the reference image steers the output while the text prompt describes the intended change or result.

Conversational Multimodal Parameter Format

The request body uses a structured messages array with role and content fields, mirroring conversational multimodal APIs. Each content block declares a type of either text or image_url, so prompts and reference images can be composed in a single, readable request structure.

Flexible Resolution Control

The size field accepts 24 preset sizes covering 1K, 2K, and 4K tiers across eight aspect ratios, and also accepts custom width-by-height values. This gives fine control over output dimensions without changing any other part of the request.

High Visual Quality with Fast Inference

The model is designed to deliver high visual quality while keeping inference fast, supporting interactive and high-throughput image generation workloads.

Pricing

Total cost = unit price of the selected resolution tier × number of images generated.

Note: 1K and 2K are priced identically at $0.024 per image, while the 4K tier is $0.032 per image. Prices listed here reflect iCreat AI's standard rates. Some models may be eligible for promotional discounts or special offers. Final charges are based on the real-time prices displayed in each model's Playground.

Application Scenarios

• High-volume text-to-image generation where low cost per image is the primary constraint
• Image-to-image generation and reference-guided editing from a supplied reference image
• Marketing and social media visuals produced across multiple aspect ratios from one prompt
• Rapid creative iteration and concept exploration using the 1K and 2K tiers at the same price
• High-resolution asset production at 4K for print, large-format display, and detailed composition work
• Batch generation pipelines that submit tasks asynchronously and poll for results

Model Comparison

Comparison Table 1: Hy Image 3.5 Preview vs Seedream 5.0 Pro

Note: Hy Image 3.5 Preview's 1K and 2K unit price is roughly one quarter of Seedream 5.0 Pro's, and its 4K tier costs about one third as much.

Comparison Table 2: Low-Price Image Generation Model Comparison

Note: Hy Image 3.5 Preview is priced below Nano Banana 2.1 Text-to-Image at every corresponding resolution tier.

Why Choose Hy Image 3.5 Preview?

• Lowest cost per image among comparable image generation models, with 1K and 2K both at $0.024 per image
• 24 preset sizes covering 1K, 2K, and 4K tiers across eight aspect ratios, plus custom aspect ratio support
• Unified conversational messages format that handles both text prompts and reference images in one request
• Reference-guided editing through image_url content blocks for image-to-image workflows
• High visual quality paired with fast inference for interactive and high-throughput use
• Simple flat request body with size and messages at the same level, requiring no nested wrappers
• Asynchronous two-step API pattern that fits naturally into batch and queue-based pipelines

Specifications

Architecture

Hy Image 3.5 Preview belongs to Tencent's Hunyuan image model family and is exposed on the iCreat platform as an asynchronous task-based API. Requests are submitted to https://api.icreat.ai/v1/task/submit/tencent/hy-image-3.5-preview with a flat top-level body containing the required size and messages fields. The messages array follows a conversational multimodal structure: each message carries a role and a content array, and each content block declares a type of either text or image_url.

The size field selects the target output dimensions from 24 preset sizes across 1K, 2K, and 4K tiers and eight aspect ratios, or from a custom width-by-height value. The platform default is 2560x1440 (2K, 16:9). Generation runs asynchronously: the submit call returns a task_id, and the same task_id is used to poll https://api.icreat.ai/v1/task/result. While the task is processing, result is an empty array; on success, result contains Image-type resources with url and download_url fields, and costUSD is returned only on successful completion. A FAILED status is terminal.

Notes

• Both steps of the workflow use the same task_id; while processing, result is an empty array and polling must continue.
• FAILED is a terminal state; check the request parameters or the reference media when it occurs.
• The request body uses flat top-level fields, with size and messages at the same level.
• size accepts 24 platform preset sizes (1K/2K/4K tiers × eight aspect ratios) as well as custom aspect ratios.
• messages is a multimodal conversational structure: text blocks carry the prompt, and image_url blocks carry optional reference images.
• This is a Preview model; capabilities and parameters may change with version updates.
• costUSD is returned only when the task succeeds.

Frequently Asked Questions

How do I call Hy Image 3.5 Preview?

Submit a task to https://api.icreat.ai/v1/task/submit/tencent/hy-image-3.5-preview with Bearer authentication using your iCreat API key, then poll https://api.icreat.ai/v1/task/result with the returned task_id. The request body uses flat top-level fields, with size and messages at the same level.

How is the cost calculated?

Billing is per image by resolution tier: $0.024 per image at 1K, $0.024 per image at 2K, and $0.032 per image at 4K. For example, generating two 2K images costs $0.048. The response returns costUSD only when the task succeeds.

What values does size support?

size accepts 24 platform preset sizes covering 1K, 2K, and 4K tiers across eight aspect ratios (1:1, 2:3, 3:2, 3:4, 4:3, 9:16, 16:9, 21:9). Custom aspect ratios such as 1024x1024, 1280x720, and 2048x2048 are also supported.

How should messages be written?

messages is an array of objects. Each message has role set to user and a content array. The content array contains a text block holding the prompt and, optionally, an image_url block holding a publicly accessible reference image URL.

Does it support reference images?

Yes. Add an image_url block to messages.content to supply a reference image, which enables image-to-image generation and reference-guided editing.

Why is this a Preview version?

Preview indicates a preview release whose capabilities and parameters may be adjusted as the version iterates. Once a stable release is available, the stable version should be treated as the reference.