Tencent's 80 billion parameter mixture-of-experts image model, the largest open-source text-to-image model at release, with readable text baked into the picture. 1 credit per image, five aspect ratios, no watermark, on every Aura AI plan.
Hunyuan Image 3, officially HunyuanImage 3.0, is the text-to-image model that Tencent's Hunyuan team open-sourced on September 28, 2025. It is a mixture-of-experts model with 80 billion total parameters, of which about 13 billion are active for each token, routed across 64 experts. At release it was the largest open-source image generation model available, and within days it climbed to the top of the LMArena text-to-image leaderboard, ahead of several closed models. The weights ship under the Tencent Hunyuan Community License, which permits commercial use with some regional and scale restrictions.
What sets it apart technically is the architecture. Instead of a standard diffusion transformer, Tencent combined a mixture-of-experts language backbone with a "Transfusion" style image head, so text understanding and pixel generation live in one unified autoregressive framework. The practical result is a model that follows long, detailed prompts of a thousand characters or more, reasons about what the scene should contain, and renders Chinese and English text inside the picture with unusual accuracy, from poster headlines to product labels and small annotations.
On Aura AI, Hunyuan Image 3 runs in text-to-image mode for a flat 1 credit per image, in five aspect ratios from 16:9 widescreen to 9:16 vertical, on every plan. It is one of the cheapest models in the studio, which makes it a natural first pass before you spend more credits on Qwen Image 3, FLUX 2 or Nano Banana. All output is delivered without watermark and can be sent straight to the image editor or to image-to-video.
Six things Tencent's largest open image model does well
64 experts and 80 billion total parameters, with about 13 billion active per token. The model has the knowledge of a very large network while spending compute only on the experts a given prompt needs.
Titles, slogans, logos, signage and small labels come out legible in both Chinese and English. This was a headline feature of the release and it holds up on posters, packaging mockups and infographics.
Text understanding and image generation share one autoregressive backbone. The model interprets the prompt the way a language model would, then generates pixels, which improves adherence on complex scenes.
Tencent reports semantic understanding of prompts beyond 1,000 characters. You can describe several subjects, their positions, the lighting and the text you want on screen, and the model keeps track of all of it.
The full checkpoint is public under the Tencent Hunyuan Community License, and the model reached first place on LMArena's text-to-image leaderboard in October 2025. On Aura AI you skip the 80B download and run it hosted.
The lowest price tier in the Aura AI image catalog, shared with FLUX 2 Turbo, Krea 2 Turbo and Gemini 3.1 Flash. Iterate on composition at low cost, then move to a 2 or 3 credit model only if you need an editor.
Four steps from prompt to a downloadable image
A single flat cost, the same in every aspect ratio
| Variant | Mode | Credits per image | Best for |
|---|---|---|---|
| Hunyuan Image V3 | Text to image | 1 credit | Text-heavy posters, dense multi-subject scenes, cheap first drafts |
How Tencent's open model compares to its closest rivals on Aura AI
| Feature | Hunyuan Image 3 | Qwen Image 3 | FLUX 2 | Nano Banana |
|---|---|---|---|---|
| Maker | Tencent | Alibaba | Black Forest Labs | |
| Text-to-image credits | 1 | 2 | 2 (Turbo 1, Max 3) | 3 (Pro or 2) |
| Edit variant on Aura AI | No | ✓ 2 credits | ✓ 2 credits | ✓ 2 or 3 credits |
| Open weights | ✓ Community License | No (3.0 released without weights) | Dev checkpoint only | No |
| Text rendering | Chinese and English, strong | Strong, 12 languages upstream | Good | Good |
| Architecture | 80B MoE, unified autoregressive | Not disclosed for 3.0 | Diffusion transformer | Gemini-based |
| Aspect ratios on Aura AI | 5 | 5 | 5 | 5 |
| Plan | All plans | All plans | All plans | All plans |
Where a 1 credit model with real text rendering pays off
Event posters, sale banners and social announcements where the headline has to be spelled correctly. Put the exact copy in quotes and specify position and typeface style.
Product boxes, bottles and bags with brand names and ingredient text rendered legibly. Useful for pitch decks and early e-commerce concepts before a designer touches them.
Simple explanatory graphics with labelled parts, numbered steps and short captions. The unified architecture keeps the labels attached to the right elements.
At 1 credit per image you can run ten variations of a scene for the price of a few premium generations. Explore compositions here, then refine the winner in a model with an editor.
Common questions about Hunyuan Image 3 on Aura AI
Tencent's 80B open image model, 1 credit per image, five aspect ratios, no watermark. Available on every plan, worldwide.