HunyuanImage
TencentHunyuanImage is Tencent's line of image generation models. HunyuanImage 3.0, released in September 2025, is an 80-billion-parameter mixture-of-experts model that generates images inside a single autoregressive model that also reads images, and its Instruct version (January 2026) adds prompt reasoning and image editing.
Scored on HunyuanImage 3.0. Tencent's hosted HunyuanImage 3.5 Preview is not yet established as its successor, so it is left for a later review. The earlier HunyuanImage 2.1 remains on the Hub; HunyuanVideo, Hunyuan3D and HunyuanVideo-Foley are separate entries.
Openness
3 high confidence- weights
- open (HunyuanImage 3.0 and its Instruct and distilled Instruct checkpoints download from Hugging Face without a gate)
- data
- described (the technical report describes the data curation; no training data is released)
- code
- partial (the repository publishes inference code, including vLLM support; no training code)
- license
- Tencent-Hunyuan-Community-License (every checkpoint: licensed for a territory that excludes the European Union, the United Kingdom and South Korea; licensees above 100 million monthly active users must ask Tencent; outputs may not be used to improve other models)
- hosted-candidate
- hosted (HunyuanImage 3.5 Preview (September 2026) runs on Tencent's servers through hosted APIs; no weights or license terms for it were found among Tencent's published releases, and whether it supersedes HunyuanImage 3.0 is not yet established)
HunyuanImage 3.0 and its Instruct versions download without a gate, but Tencent's community license does not apply in the European Union, the United Kingdom or South Korea and needs a separate license for the very largest services. Tencent publishes inference code only, and the training data is described rather than released. The newer 3.5 Preview is hosted, and its relation to 3.0 is not yet settled.
- https://arxiv.org/abs/2509.23951 recorded 2026-11-12
HunyuanImage 3.0 technical report abstract: its key components include "meticulous data curation"; "By releasing the code and weights of HunyuanImage 3.0".
- https://docs.comfy.org/tutorials/partner-nodes/tencent/hy-image-3-5.md recorded 2026-11-12
Comfy's partner-node guide: "Tencent Hunyuan Image 3.5 Preview is Tencent's preview image model, available in ComfyUI through Partner Nodes"; "Both run on Tencent's servers through Comfy API, so no model downloads or local GPU are required."
- https://huggingface.co/api/models/tencent/HunyuanImage-3.0 recorded 2026-11-12
Model API for tencent/HunyuanImage-3.0: "gated": false; cardData license "other", license_name "tencent-hunyuan-community", license_link "LICENSE"; 32 safetensors shards ("model-0001-of-0032.safetensors" onward).
- https://raw.githubusercontent.com/Tencent-Hunyuan/HunyuanImage-3.0/main/LICENSE recorded 2026-11-12
HunyuanImage 3.0 LICENSE body: "TENCENT HUNYUAN COMMUNITY LICENSE AGREEMENT"; "THIS LICENSE AGREEMENT DOES NOT APPLY IN THE EUROPEAN UNION, UNITED KINGDOM AND SOUTH KOREA"; a separate license above "100 million monthly active users"; outputs may not be used "to improve any other AI model".
- https://raw.githubusercontent.com/Tencent-Hunyuan/HunyuanImage-3.0/main/README.md recorded 2026-11-12
HunyuanImage 3.0 README news: "September 28, 2025: HunyuanImage-3.0 Open Source - Inference code and model weights publicly available"; "January 26, 2026": the Instruct and Instruct-Distil checkpoints; the open-source plan checks off inference, the checkpoints and vLLM support, and lists no training code.
Adoption
2 high confidenceAdoption is measured as Hugging Face downloads summed across Tencent's four HunyuanImage checkpoints, most of them HunyuanImage 3.0 Instruct. The current 3.5 Preview has no downloadable weights to count.
- https://huggingface.co/api/models?author=tencent&search=HunyuanImage&sort=downloads&direction=-1&limit=100 recorded 2026-11-12
Listing sorted by downloads: HunyuanImage-3.0-Instruct 26044, HunyuanImage-3.0 1681, HunyuanImage-3.0-Instruct-Distil 532, HunyuanImage-2.1 213; 28,470 in total.
Capability
2 medium confidenceEvery HunyuanImage 3.0 model on the public image arena sits in the lower half of the board, the hosted 3.0 highest. FLUX.2 [dev] places above all of them, in the upper half.
- https://artificialanalysis.ai/image/leaderboard/text-to-image recorded 2026-11-12
Overall board as served (112 models): "HunyuanImage 3.0 (Tencent Cloud)" 58th, Elo 994; "HunyuanImage 3.0 Instruct" 64th, Elo 987; "HunyuanImage 3.0" 78th, Elo 935. The hosted "HunyuanImage 3.5 (Preview)" is 71st, Elo 973.
Verified 2026-11-12