Hy-Image-3.5-preview Spicy API

Hy-Image-3.5-preview (Hy Image 3.5 preview) is Tencent Hunyuan's cost-effective, professional-grade image generation and editing model, officially released on September 22, 2026. It continues the Hy Image 3.0 design lineage with a systematic upgrade "from intent understanding to visual expression." It supports text-to-image, image-to-image, and multi-turn conversational editing: a Chat/Messages structure receives text and reference images together, retains context across turns, and lets you keep refining subjects, scenes, and composition via natural language after generating — no restarting from scratch. In Tencent's internal GSB blind test with several hundred professional designers, its overall win rate exceeded Hy Image 3.0 by roughly 30%, level with Seedream 5.0 Pro and slightly ahead of Nano Banana Pro and Qwen-Image-3.0 Pro (vendor's own evaluation).

The model is offered through the iCreat platform as an API using a two-step asynchronous task flow: submit a task to obtain a task_id, then poll the result endpoint until a terminal status. Only output images are billed, with uploaded references free of charge: $0.024 per image at 1K and 2K, and $0.032 at 4K.

Model Positioning

Hy-Image-3.5-preview targets "continuously iterable professional visual production." It is not a one-shot toy but a model built for real workflows that demand revision: commercial posters, episode storyboards, film and TV materials, UI design, creative illustration, brand visuals, and e-commerce assets — plus everyday scenarios like photo retouching, portrait beautification, and knowledge cards. Within Tencent Hunyuan's image family it succeeds Hy Image 3.0 as the new flagship: language and visual understanding improve further, and better generation is delivered at lower inference cost — at $0.024 per 2K image, the pricing is highly competitive among professional-grade peers.

Core Capabilities

Unified Text-to-Image and Image-to-Image

One model covers both directions: a text-only messages structure generates from the prompt; adding image blocks switches to instruction-based editing or composition — the same endpoint, switched as needed.

Multi-Turn Conversational Editing

Context is retained across turns: after generating, keep refining subjects, scenes, composition, and details in natural language without re-describing everything — the working style of chat-based image editors.

Upgraded Text Rendering and Layout

Text rendering and layout improve markedly over the previous generation; multilingual typography stays clean and usable, suited for text-heavy posters, infographics, and UI screens.

Realism and Edit Consistency

Enhanced realism and stylistic expressiveness, with lifelike micro-expressions and material detail; subjects and style stay consistent across editing rounds — no drift while you retouch.

High-Resolution Output

Multiple aspect ratios are supported with output resolutions up to 4K (4096×4096), meeting print-grade and high-density-screen delivery needs.

Pricing

Only output images are billed; uploaded reference images are free of charge. The costUSD field in the response returns the actual cost once the task succeeds.

Application Scenarios

• Commercial posters and brand visuals: professional output refined over multiple rounds
• Episode storyboards and film materials: continuous creation with consistent characters and style across shots
• UI design and interface simulation: interface graphics with clean, usable text and controls
• E-commerce assets: volume production of product and scene imagery plus marketing material
• Portrait beautification and daily retouching: conversational, continuous adjustment for everyday scenarios

Model Comparison

Same-Series Comparison

Cross-Model Comparison

Same-series data comes from Tencent Hunyuan's official releases; cross-model standing reflects Tencent's own GSB blind test, not independent evaluation; this model's pricing and specifications follow the iCreat channel documentation.

Why Choose Hy-Image-3.5-preview?

• Multi-turn conversational editing: keep refining in natural language after generating — never start from scratch
• Text-to-image, image-to-image, and continuous editing unified in one model and endpoint
• GSB blind-test win rate up ~30% over the previous generation, level with Seedream 5.0 Pro (vendor's own evaluation)
• $0.024 per 2K image — a fraction of professional-grade peers, billed on output only
• 1K and 2K at the same price: high-resolution delivery without a premium, and 4K at just $0.032

Specifications

Architecture

Hy-Image-3.5-preview continues the Hy Image 3.0 design lineage with a conversational multimodal architecture: requests are organized as a messages list where text and reference-image blocks pass through one context, and the model generates and edits based on the full conversation history — so text-to-image, image-to-image, and multi-turn editing share a single pipeline. The model is upgraded "from intent understanding to visual expression": language and visual understanding improve over 3.0, with gains across text rendering and layout, realism and stylistic expressiveness, and editing consistency, while optimized inference-resource utilization lowers cost. Before the public announcement it ran in a gradual rollout inside the Yuanbao app for about two months, driving image-generation request volume up by more than 50% during the gray release.

Notes

• Submit and query must be chained with the same task_id
• While processing, result is []; keep polling, with a suggested interval of 2–5 seconds
• FAILED is terminal; check request parameters and reference media before resubmitting
• messages follows a conversation structure; combine text and image blocks as needed
• Reference image URLs must be publicly accessible
• Multi-turn editing retains conversation history; follow-up instructions build on prior results
• size supports common presets and custom aspect ratios; the resolution tier maps to the billing tier
• Only output images are billed; reference uploads are free
• costUSD is returned only when the task succeeds; failed tasks do not produce a cost field

FAQ

How do I get started with the Hy-Image-3.5-preview API?​

Register on the iCreat platform and obtain an API Key from the console, send a generation request to the submit endpoint, then poll the result endpoint with the returned task_id. All requests are authenticated with the Authorization: Bearer header. Keep your API Key safe and never expose it in client code or public repositories.

How is the cost calculated?​

Only output images are billed: $0.024 per image at 1K and 2K, $0.032 at 4K, with reference uploads free of charge. For example, generating 100 2K images costs $2.40. The costUSD field in the response gives the actual cost once the task succeeds.

Does it support generation or editing?​

Both, and continuously. With text-only content blocks it generates from the prompt; adding image blocks enables instruction-based editing or composition of references. Across multiple turns the model retains conversation history — generate first, then keep refining the result without re-describing everything.

What sizes are supported?​

size supports common presets including 1080x1080, 1024x1024, 1280x720, and 2048x2048 plus custom aspect ratios, with resolutions up to 4K; 1K and 2K are the same price, so pick freely per delivery scenario.

How does multi-turn editing work?​

Carry the conversation history and new instructions in messages: the model understands references like "change the background of the last one to a night scene" from the full context, without re-uploading images or restating the whole brief. Each round's result can feed the next round of iteration.

What if the task fails?​

FAILED is a terminal status. Check in order: whether the messages structure is complete (with role and content), whether text blocks are non-empty, whether reference image URLs are publicly accessible, and whether size uses a supported format. Fix any issue and resubmit. The cost field is returned only when the task succeeds.