HunyuanImage 3.0 is Tencent's open image model — an 80-billion-parameter mixture of experts with strong world knowledge. On Kenerate: Hunyuan Image 3 (30 credits), 2.1 (7.5) and 3 Instruct for edits.
Quick answer
Updated
HunyuanImage 3.0 is Tencent's open image model, released on 28 September 2025: an 80-billion-parameter mixture-of-experts model (13B active) that handles text and images together, which Tencent calls the largest open-source image MoE. An Instruct version for editing followed on 26 January 2026. On Kenerate: Hunyuan Image 3 costs 30 credits, Hunyuan Image 2.1 7.5 (native 2K), and Hunyuan Image 3 Instruct 135 with up to 2 reference images.
About the model
Facts from the maker's own announcement and repository, linked below.
| Maker | Tencent Hunyuan |
|---|---|
| HunyuanImage 3.0 | 28 September 2025 — 80B total parameters, 13B active, 64 experts; open weights |
| Instruct | 26 January 2026 — instruction-following edit model (plus an 8-step distilled version) |
| Tencent's highlights | Reasoning and world knowledge, prompt adherence, photorealism |
| On Kenerate | Hunyuan Image 3 (30), Hunyuan Image 2.1 (7.5, native 2K), Hunyuan Image 3 Instruct (135, 2 refs) |
Gallery
Real outputs made on Kenerate. Open any image to copy its prompt and run it yourself.
Models
Every version in the generator, and what each is good at.
From our tests
Hunyuan runs through its provider with its own content rules. For adult art, use open models on the uncensored AI image generator (18+).
How to
Try Hunyuan Image on your own prompt — 25 free credits on sign-up.
In depth
Background, how to prompt it, and when to pick something else.
HunyuanImage 3.0 models text and images together in one autoregressive mixture-of-experts network: 64 experts, 80B parameters in total and about 13B active per token [1]. That size is where its world knowledge comes from.
Pick it for scenes that need knowledge — traditional art forms, historical settings, technical cutaways, maps. For clean text, pick GPT Image 2; for cheap drafts, Flux 2 Dev.
We only publish real, verified reviews — never invented quotes. Send a line about your result (and whether we may quote you), and it will appear here.
FAQ
4 questions
Tencent's open image model released on 28 September 2025 — an 80B-parameter mixture of experts with 13B active parameters.
Hunyuan Image 2.1 is 7.5 credits, Hunyuan Image 3 is 30, and Hunyuan Image 3 Instruct is 135.
Tencent released the weights on GitHub and Hugging Face under its own community licence.
Mixed in our tests: short Chinese calligraphy looked right, but a longer English chalkboard heading came out garbled.
Explore
Generators, design tools, photo edits, models and comparisons — all in one studio, on one credit balance.
Written by the Kenerate AI team
We run these models in the Kenerate generator. Model facts come from the makers' own pages; specs and limits were checked against our code; the examples are our own outputs. Last updated 27 September 2026. Corrections: hello@kenerateai.com.
Ready when you are
25 free credits on sign-up. No watermark.
