Hunyuan Image 3

HunyuanImage‑3.0 多模态文生图,画质优异、指令遵循性强。支持多种尺寸、种子、JPEG/PNG 输出。

费用:6 积分

输入

试试:

The size of the generated image in pixels (width*height).

Set a specific seed for reproducible results; -1 uses a random seed.

The format of the output image.

输出

生成内容将显示在此处

示例结果

A teenage girl on a rooftop at sunset, ultra realistic details

文生图:

A teenage girl on a rooftop at sunset, ultra realistic details

A professional young woman walks down a street in New York, wearing a well-tailored grey wool blazer over a white silk camisole with matching wide-leg trousers. She is wearing delicate gold earrings and carrying a black leather tote bag. Her expression is confident, with a blurred city scene in the background.

文生图:

A professional young woman walks down a street in New York, wearing a well-tailored grey wool blazer over a white silk camisole with matching wide-leg trousers. She is wearing delicate gold earrings and carrying a black leather tote bag. Her expression is confident, with a blurred city scene in the background.

An old wizard with a long white beard, wearing a classic pointed wizard hat, sits comfortably in a wooden chair on top of his stone wizard tower. He leisurely smokes a pipe, with magical smoke swirling in the air around him, creating faint glowing shapes. The tower's top is high above the clouds, and a vast sky stretches out in the background, giving the scene a tranquil, isolated vibe. At the top of the image, large whimsical text reads: "I'm done procrastinating!" Further down, near the bottom-middle of the image, text in a similar whimsical style reads: "I'm straight up not working anymore!" The text is a key focal point, blending humorously into the laid-back, magical atmosphere. aidmaMJ6.1, hkmagic.

文生图:

An old wizard with a long white beard, wearing a classic pointed wizard hat, sits comfortably in a wooden chair on top of his stone wizard tower. He leisurely smokes a pipe, with magical smoke swirling in the air around him, creating faint glowing shapes. The tower's top is high above the clouds, and a vast sky stretches out in the background, giving the scene a tranquil, isolated vibe. At the top of the image, large whimsical text reads: "I'm done procrastinating!" Further down, near the bottom-middle of the image, text in a similar whimsical style reads: "I'm straight up not working anymore!" The text is a key focal point, blending humorously into the laid-back, magical atmosphere. aidmaMJ6.1, hkmagic.

常见问题

Hunyuan Image 3 是什么?
一款先进的文生图模型,具备强大的多模态理解与高质量成像能力。
支持哪些尺寸?
从 256×256 到 1536×1536(以服务方限制为准)。
如何获得可复现的结果?
设置固定种子。相同提示词 + 种子通常会得到相近结果。
应选择哪种输出格式?
JPEG 体积更小,PNG 适合无损与对透明度/后期处理有要求的工作流。
与 DiT 架构有何不同?
采用统一自回归(autoregressive)多模态架构,联合建模文本与图像,提升语境一致性、推理能力与提示词遵循度。
模型规模有多大?
为 MoE(专家混合)架构,包含 64 个专家,总参数约 800 亿,推理时每 token 激活约 130 亿参数。
提示词遵循性如何?
在多项评测(如 SSAE 与人工 GSB)中表现优异,语义与构图对齐能力强,能稳定地还原关键信息。
是否会自动扩展稀疏提示词?
是。模型具备世界知识推理能力,会对简短提示进行适度细化。建议编写简洁、无冲突的指令以更好地引导扩写。
有何写提示词建议?
推荐:主体 + 属性 + 场景/构图 + 光照/镜头/风格。必要时添加负面描述;避免自相矛盾或同时叠加过多风格。
是否支持图生图或多轮交互?
其开源计划包含图生图与多轮交互。当前应用主要开放文生图常用参数。
许可与商用?
开源许可为 tencent-hunyuan-community。进行商用前请阅读并遵守许可条款。
生成时长?
与服务端负载与生成尺寸有关(256–1536)。通常数秒到数十秒;更大尺寸可能更久。