Hy Image3.5 preview API

Technical Specifications of Hy-Image 3.5 Preview

What is Hy-Image 3.5 Preview?

Hy-Image 3.5 is Tencent Hunyuan's professional image generation model designed for text-to-image generation, image-to-image generation, and multi-turn image editing.

Unlike a conventional single-prompt image-generation API, Hy-Image 3.5 Preview uses a Chat/Messages-style protocol, allowing text prompts, reference images, and previous image-generation context to be combined in a conversational workflow. The model can therefore be used not only to create an image from scratch, but also to iteratively modify an existing result across multiple rounds.

Main Features of Hy-Image 3.5 Preview

• Text-to-image generation: Generate images from natural-language prompts and let the model interpret the requested composition, style, subject, and visual details.
• Image-to-image generation: Combine text instructions with reference images to guide the generated result.
• Multi-turn editing: Preserve previous conversation and image context so users can progressively refine an image instead of restarting from scratch.
• Multimodal input: A request can combine textual instructions with reference images, making it suitable for style transfer, product-image editing, character adjustments, and visual iteration.
• 4K output: The official API documentation supports output up to 4096 × 4096 pixels, subject to the documented image-area constraints.
• External search enhancement: The API documentation lists external-search enhancement as a supported capability, allowing image-generation workflows to incorporate additional information when required.
• Flexible image sizing: Developers can specify an exact size, or allow the model to determine an appropriate output size based on the prompt and configured pixel budget.
• Seed control: A seed can be supplied when reproducible generation is required.

Image Generation and Editing Capabilities

Text-to-Image

Hy-Image 3.5 Preview can generate images directly from text instructions. The model accepts a Messages-style request, with the user's text serving as the image-generation instruction.

This makes it suitable for creative tasks such as illustrations, concept art, advertising visuals, presentation graphics, product imagery, and other visual-content workflows.

Image-to-Image

The model can combine text instructions with reference images. Reference images can be supplied through public image URLs or Base64-encoded image data.

The official API documentation allows up to 20 reference images, with a maximum size of 20MB per image.

Multi-Turn Image Editing

One of Hy-Image 3.5 Preview's distinctive capabilities is multi-turn editing. Instead of generating every revision as an independent request, developers can preserve the previous conversation and generated-image context and then submit a new editing instruction.

This enables workflows such as:

1. Generate an initial image.
2. Ask the model to change a specific element.
3. Preserve the rest of the composition.
4. Continue refining the image over multiple rounds.

The API returns an assembled_history structure that can be reused as context for subsequent editing rounds.

Resolution and Image Size

Hy-Image 3.5 Preview supports flexible output dimensions.

When size is explicitly provided, the requested dimensions can be controlled directly, with the official API documenting dimensions from 256 to 8192 pixels per side and a maximum image area of 16,777,216 pixels, corresponding to a 4K-class 4096 × 4096 output.

For 4K generation, the API documentation recommends explicitly specifying the desired size. If no exact size is supplied, the model can determine the output dimensions based on the prompt and configured pixel budget.

Benchmark Performance of Hy-Image 3.5 Preview

Tencent's public launch materials describe Hy-Image 3.5 Preview as a professional-grade image-generation model. Tencent also reported an internal blind evaluation involving hundreds of professional designers, claiming a 30% improvement over Hy-Image 3.0 and comparable results to Seedream 5.0 Pro. These are vendor-reported internal evaluation results, rather than an independently reproduced public benchmark.

Because the official API documentation does not provide a standardized public benchmark table, these reported results should not be presented as directly comparable leaderboard scores.

Hy-Image 3.5 Preview vs. Hy-Image 3.0

Hy-Image 3.5 Preview is therefore positioned as a more conversational and editing-oriented generation system, while Hy-Image 3.0 uses a simpler generation interface centered around a prompt and optional reference images.

Hy-Image 3.5 Preview vs. Other Image Models

Hy-Image 3.5 Preview is primarily differentiated by its combination of image generation, multimodal reference input, conversational editing, and large-context image workflows.

For applications focused on iterative visual creation, the Chat/Messages architecture can be particularly useful because the model can retain the context of previous editing rounds rather than treating each generation as an isolated request.

Compared with models that primarily expose a single image-generation prompt, Hy-Image 3.5 Preview provides a workflow closer to a conversational image editor.

Representative Use Cases

1. Commercial Advertising

Generate campaign visuals, promotional artwork, marketing graphics, and advertising concepts from detailed natural-language descriptions.

2. Product Image Creation

Use reference images and text instructions to create or modify product visuals while maintaining important visual characteristics.

3. UI and Product Design

Generate interface concepts, visual mockups, design explorations, and presentation-ready visual assets.

4. Character and Portrait Editing

Use reference images and iterative instructions to refine characters, portraits, styling, clothing, backgrounds, and other visual elements.

5. Creative Illustration

Create illustrations, concept art, story visuals, and other creative assets from detailed prompts.

6. Multi-Round Visual Production

Use the conversational editing capability to progressively refine an image across multiple instructions, making the model suitable for workflows where the final visual is developed through several iterations.

Limitations and Considerations

Hy-Image 3.5 Preview is currently a Preview model, so API behavior and underlying model versions may change as Tencent continues to update the service.

Generated image URLs returned by the API are temporary. The official documentation states that generated image URLs are valid for 12 hours, so applications that need persistent assets should download and store the generated images rather than relying on the returned URL indefinitely.

Developers using multi-turn editing should also preserve the returned assembled_history when continuing an editing session, because this structure contains the context required for subsequent rounds.

How to Access Hy-Image 3.5 Preview API via CometAPI

CometAPI can provide a unified integration layer for applications that want to access supported AI models through a consistent API workflow.

Step 1: Sign Up or Log In to CometAPI

Create a CometAPI account or sign in to an existing account. After entering the CometAPI console, access the API key management area.

Step 2: Create a CometAPI API Key

Generate an API key from the CometAPI console. Keep the key secure because it is used to authenticate API requests.

Step 3: Find Hy-Image 3.5 Preview

Open the CometAPI model catalog and search for Hy-Image 3.5 Preview. Confirm that the model is currently available through your CometAPI account before integrating it into a production application.

Use the model identifier:

hy-image-v3.5-preview

Step 4: Configure Your Application

Configure your application to use CometAPI's unified API interface and authenticate requests with your CometAPI API key.

For an image-generation workflow, pass the appropriate text prompt, reference-image information, and generation parameters supported by the model.

Step 5: Send Your Image Generation Request

Submit the request through CometAPI using hy-image-v3.5-preview as the selected model.

Depending on your application, you can use the model for text-to-image generation, reference-image generation, or multi-turn editing.

Step 6: Process and Store the Generated Image

Retrieve the generated image from the API response and store it in your own application storage if the image needs to remain available beyond the temporary URL lifetime.