HunyuanWorld-1.0 Image to Panorama
Builds a full 360-degree equirectangular panorama around a single photo, generating the parts of the scene the camera never captured. You set the field of view the photo covers on the sphere.
Overview
HunyuanWorld 1.0 Image to Panorama takes one image, projects it into an empty equirectangular canvas at the field of view you choose, and paints the rest of the sphere around it. The weights are Tencent's HunyuanWorld-PanoDiT-Image LoRA on FLUX.1 Fill [dev], Black Forest Labs' inpainting and outpainting model.
Tencent scores faithfulness to the input with CLIP-I, an image similarity measure, and reports 85.1 for HunyuanWorld 1.0 on image-to-panorama, against 80.8 for MVDiffusion and 73.9 for Diffusion360.
Choosing a variant: pick this page when the panorama has to start from a picture you already have. With only a description, or to steer the scene with words, go to HunyuanWorld 1.0 Text to Panorama instead.
What you can build
- A 360-degree VR or game background grown out of one interior or landscape shot.
- An environment that has to match concept art your team already approved.
- The first step of Tencent's image-to-world workflow, which starts from this panorama.
Inputs
Upload one image in JPG, JPEG, PNG, GIF, WEBP or HEIC. There is no prompt field, so the new surroundings come from the image alone.
Three numbers set the placement:
- FOV (0 to 180, default 80) is the angle your image spans: horizontal for a landscape image, vertical for a portrait or square one.
- Theta (0 to 10, default 0) moves the image sideways by that many degrees.
- Phi (0 to 10, default 0) moves it vertically by that many degrees.
Blend Extend (default 6) sets the overlap that joins the left and right edges, and Shifting Extend (default 0) rolls the canvas sideways during the first denoising steps. Steps default to 40, guidance scale to 30 and Cfg Scale to 0. Width and height go up to 2048 each (default 1920 by 960), and a seed of 0 means random.
Outputs
An equirectangular panorama, 1920 by 960 at the default size. With Theta and Phi at 0 your image sits in the center of the frame, and the generated scene continues from its edges all the way around, including behind the camera.
Recommended settings
Set FOV to the real field of view of the camera that took the photo. Tencent's paper estimates it with a 3D reconstruction model; here you enter it yourself. Too low a value shrinks the photo into a small patch; too high a value stretches it. For image input, Tencent's reference script uses FOV 80, Theta and Phi at 0, 50 steps, guidance 30 and Cfg Scale 2.0. It also appends quality keywords to the prompt and sets a negative prompt that excludes people and blur. This page leaves both the prompt and the negative prompt empty, so Cfg Scale has nothing to guide against here and the same numbers will not match Tencent's results. Keep the output at 2:1.
Limitations
- Everything outside the original frame is generated, so the result is not a record of the real place.
- You cannot describe the new surroundings in words. For a different fill, change the seed.
- With Theta and Phi capped at 10 degrees, the photo stays near the center of the panorama.
Safety & compliance
If your photo shows a real person, get that person's consent before uploading it. Under the Tencent HunyuanWorld-1.0 Community License, the model and its outputs may be used only outside the European Union, the United Kingdom and South Korea. Organisations above 1 million monthly active users at the release date need their own license from Tencent. The license also forbids using outputs to improve any other AI model and attaches Tencent's Acceptable Use Policy. The FLUX.1 Fill [dev] base weights come under Black Forest Labs' FLUX.1 [dev] Non-Commercial License, which lets you use outputs commercially but not to train, fine-tune or distill a model that competes with FLUX.1 [dev].
API quick start
Run HunyuanWorld-image-to-panorama with a single API call.
{
"inputImage": "https://your-cdn.com/input.png",
"steps": 40,
"scale": 30.0,
"theta": 0
}curl -X POST "https://api.wiro.ai/v1/Run/tencent/hunyuanworld-image-to-panorama" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"inputImage": "https://your-cdn.com/input.png",
"steps": 40,
"scale": 30.0,
"theta": 0
}
JSONOther models you might like
Explore similar image generation models.



