{"id":1020,"date":"2026-02-24T22:08:34","date_gmt":"2026-02-24T22:08:34","guid":{"rendered":"https:\/\/wiro.ai\/blog\/?p=1020"},"modified":"2026-09-27T19:26:02","modified_gmt":"2026-09-27T19:26:02","slug":"hunyuanworld-text-to-panorama-6-prompt-test","status":"publish","type":"post","link":"https:\/\/wiro.ai\/blog\/hunyuanworld-text-to-panorama-6-prompt-test\/","title":{"rendered":"HunyuanWorld Text-to-Panorama: 6 Prompt Test"},"content":{"rendered":"

HunyuanWorld Text-to-Panorama<\/strong> was tested here for one narrow job: turning a written scene into a 1920×960 panorama that feels continuous from the far left edge to the far right. That matters for 360 viewers, environment concepts, game backdrops, and any image that needs more than a conventional wide crop. The six outputs below are the original model results already hosted on this blog. They are not mockups or retouched replacements.<\/p>\n

The test deliberately mixes enclosed architecture, open weather, stylized sci-fi, dense foliage, extreme contrast, and icy detail. Those are useful failure modes for a panorama model. A regular image model can make an appealing wide frame while still treating both edges as unrelated. A panorama-oriented model has to keep horizon, lighting, scale, and repeated structures believable across the complete width.<\/p>\n

HunyuanWorld Text-to-Panorama test setup<\/h2>\n

Every image used HunyuanWorld Text-to-Panorama on Wiro<\/a>. The renders were set to 1920×960, a 2:1 panorama ratio. The run settings were 35 inference steps, guidance scale 30.0, Blend Extend 6, CFG scale 0.0, and seeds 801 through 806. The current model documentation lists 40 steps as its default, so these tests use five fewer steps than the default rather than presenting a maximum-quality sweep.<\/p>\n

The Wiro documentation does not publish a fixed price for this model, so no cost figure is claimed here. It includes a completed example with a six-second elapsed time, but that is an example task rather than a promise for every queue, prompt, or load condition. Treat it as a useful reference point, not a benchmark. The original test did not record individual task timings, so this article does not invent them.<\/p>\n