HunyuanImage-3.0-Instruct is a version of Tencent’s AI image model tuned to follow instructions. It was released in January 2026 and it is a both an image generator and editor.
It rewrites prompts on its own and handles detailed instructions better than the regular version. So, it can do more complex stuff like scene changes or object swaps.
It’s built on a setup that mixes text and image understanding. The tool fills in short prompts and figures out what you mean from the full context.
You can use it to make images from scratch or tweak them. That includes changing styles, adding or removing things, blending images together, and improving looks.
It handles long, detailed instructions well. Even when prompts are packed with steps, it follows along and makes visuals that match.
It reasons things out to arrange scenes better. Like placing objects in smart ways, keeping the story clear, or adjusting layout based on what you ask.
Under the hood it runs on a Mixture-of-Experts design with about 80B parts. Only around 13B parts activate each time to keep it fast but powerful.
Compared to the base version, this one’s tuned for instruction-following. It edits creatively, understands layered tasks, and makes images that match complex prompts.
But as usual, Hunyuan Image has a massive size (~169 GB VRAM) making it impractical for most users, which limits it's adoption as a local open-source model. It is available on services like FAL.
If you'd like to access this model, you can explore the following possibilities:
Weights