Hy Image 3.0 or 3.5 Preview: 3 vs 20 references, size rules, seed

Hy Image 3.0 takes up to 3 reference images and an 8,192-character prompt; Hy Image 3.5 Preview takes up to 20 references and a 100k-token limit through a messages protocol, with a much wider size range. Hy Image 3.5 Preview is not listed in the Sume image catalog (read 2026-10-09). Pick 3.5 for edits and larger output, and 3.0 for simple text-to-image with automatic prompt rewriting.

How do the two models differ?

Everything in this table is from Tencent's page.

What is resize_max_pixels?

It is a 3.5 field for the reference side: an input image whose area is at or under the threshold passes as is; a larger one is scaled to roughly the threshold before the model sees it, so a 6K photo does not blow up the context. The default is 1,048,576 pixels. generate_max_pixels is the output side, with 1K, 1.5K (default) and 2K levels, and only applies when you do not pass size.

What can I call on Sume instead?

Neither Hy model is listed. For edits with many references, GPT Image 2.5 takes up to 16 and Nano Banana 2.1 up to 10. Sume does not serve seed, so a fixed-seed reproduction is not possible there; log your prompt and keep the file instead. Rows and limits are in the image models docs.

• Hy 3.5 for multi-turn edits and 4K via size.
• Hy 3.0 for simple text-to-image with automatic prompt rewriting (about 11 seconds).
• Sume: pick a listed row by reference count and price.

Where is the price?

Tencent's page gives no per-image price, so this post states none. Check your Tencent console before you commit.

The 3.5 size range of 256 to 8192 with an area cap is the biggest practical change from 3.0, which stops at 2048 and the area of a 1024 square. If your use case is a poster or a banner, the size rules alone may decide the model. If it is a square avatar, either model is enough, and price decides.