Hy Image 3.5

Comparison · Tencent Hunyuan image models

Hy Image 3.5 vs Hunyuan Image 3.0

Hy Image 3.5 Preview is Tencent's successor to Hunyuan Image 3.0. The upgrade is less about a single prettier render and more about how many inputs a brief can carry and how revisions work. This comparison uses Tencent's published figures.

Try Hunyuan Image 3.0

In short

  • References: 20 on Hy Image 3.5 Preview versus three on Hunyuan Image 3.0.
  • Editing: 3.5 combines generation and instruction editing in one model with multi-round context; 3.0 treats each request on its own.
  • Output: 3.5 lists direct 4096 × 4096 generation.
  • Tencent states roughly a 30% improvement for 3.5 over 3.0; that figure is Tencent's, not a measurement made here.

Specification comparison

All figures below come from Tencent's release material for each model.

  • Release: Hunyuan Image 3.0 came first; Hy Image 3.5 Preview was announced on September 22, 2026.
  • Reference images per request: 3.0 — three; 3.5 Preview — up to 20.
  • Editing: 3.0 — separate requests, no carried context; 3.5 Preview — multi-round, earlier image and instructions carried forward.
  • Maximum output: 3.5 Preview — 4096 × 4096 when requested.
  • Published API price: 3.5 Preview — $0.024 per 2K output image on Tencent Cloud, references not billed.

When Hunyuan Image 3.0 is still the right choice

If a brief fits in one prompt — one subject, one composition, one light — Hunyuan Image 3.0 already handles it well, and it is the model you can run in this workspace today. Its Instruct variant is tuned for detailed briefs and typography, and the Edit route accepts up to two reference images for a single targeted change.

When Hy Image 3.5 will matter

The 3.5 upgrade pays off on work that needs several inputs or several rounds: a product shot that must match a brand palette and an existing layout, a character that must stay consistent while the scene changes, or an image that a client revises five times. Twenty references and carried-forward context are designed for exactly that kind of iteration.

Getting comparable results from 3.0 now

Until the 3.5 route is connected, you can approximate a multi-round workflow with Hunyuan Image 3.0 by keeping each revision small and explicit.

  1. Generate the base image with a complete paragraph prompt.
  2. Upload that result to the Edit route and ask for one change at a time.
  3. Say what must stay, for example "keep the framing, the light and the bottle shape".

Frequently asked questions

Is Hy Image 3.5 better than Hunyuan Image 3.0?

Tencent reports about a 30% improvement and lists far more references plus multi-round editing. Whether that matters depends on the brief: single-prompt images show less difference than multi-input, multi-revision work.

Can I compare both models side by side here?

Not yet. Hunyuan Image 3.0 runs in this workspace; the Hy Image 3.5 Preview route is listed but disabled until it is connected.

Does Hy Image 3.5 charge for reference images?

According to Tencent Cloud, only output images are billed; references are free.

Updated 2026-10-04. Independent Hy Image 3.5 workspace, not affiliated with Tencent. Model facts are Tencent's published figures. Every generation here uses the model you select in the workspace.