HunyuanImage 3.0
A solid pick for self-hosting and full data control.
Overview
HunyuanImage-3.0 is a Chinese-built MoE image generator (80B parameters, 13B active per token) relying on a unified autoregressive text-image architecture. It delivers photorealistic, prompt-faithful output, with an Instruct variant adding image-to-image editing. Self-hosted, it keeps all data on your own infrastructure regardless of the vendor's country, but its community licence carries territorial restrictions — some versions cover the EU — that must be checked before deployment, and it requires workstation-class GPU hardware.
Skill profile
Not disclosed
Strengths
- Open-source and self-hostable
Limitations
- API pricing not disclosed
Who is it for
- you want to control cost or self-host
- you have GDPR constraints
Ideal use cases
- Marketing image generation
- Creative visual prototyping
- Image-to-image editing
- Editorial illustration workflows
Access & availability
Key specifications
Privacy
Run it locally
| Quantization | Disk | RAM / VRAM | Typical hardware |
|---|---|---|---|
| Q4 · recommended | 45.6 GB | 52 GB | 64GB Mac / dual 24GB GPUs |
| Q8 · balanced | 85.6 GB | 96 GB | 96-128GB Mac Studio / 4× GPUs |
| FP16 · max quality | 160 GB | 178 GB | Server GPU infra (H100/A100…) |
Estimates for a moderate context. Long contexts need more RAM (KV cache).
Deploy
Copy-ready commands generated from this card. Adjust context length and GPU count to your hardware.
OpenAI-compatible server for production on NVIDIA GPUs.
pip install vllm
vllm serve tencent/HunyuanImage-3.0 \
--max-model-len 32768 \
--tensor-parallel-size 2 \
--dtype autoAdvanced data · for expertsArchitecture, modalities, detailed cost, full benchmarks▾
| Arena Elo | — |
| MMLU | — |
| GPQA | — |
| HumanEval | — |
| SWE-Bench | — |
| MATH | — |