scenario-skyboxes

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Scenario Skyboxes and 360 Panoramas

Scenario Skybox与360全景图

Overview

概述

Scenario hosts dedicated skybox models that produce seamless equirectangular 360 panoramas, plus a seam-preserving skybox upscaler. Always generate with a skybox-specific model rather than a generic image model: these enforce the seam continuity and pole geometry that ordinary text-to-image output lacks.

Connection and the core generation loop: see the

scenario

skill in this repo. If a sibling skill named here is missing from your available skills, ask the user to install it (

npx skills add scenario-labs/skills --skill <name>

); unattended, proceed from tool schemas and flag the gap.

Scenario提供专用的skybox模型，可生成无缝的equirectangular 360全景图，还有一款保留拼接缝的skybox超分模型。请始终使用skybox专用模型而非通用图像模型进行生成：这些专用模型会保证拼接缝连续性和极点几何一致性，这是普通文本生成图像的输出所不具备的。

连接方式和核心生成流程：参见本仓库中的

scenario

skill。如果你的可用skill中缺少这里提到的同级skill，请让用户安装（命令：

npx skills add scenario-labs/skills --skill <name>

）；在无人值守模式下，请基于工具schema继续操作并标记缺失项。

Quick reference

快速参考


search
model_schema_get
dry_run=true
pending_job_ids
asset_display

Step	Tool	Purpose
1	`search` (target="models", query="skybox", public=true)	Discover current skybox models
2	`model_schema_get`	Exact parameter contract for the chosen model
3	`model_run`	`dry_run=true` to price; `wait=false` to batch
4	`jobs_wait`	On timeout re-call with `pending_job_ids`
5	`asset_display` / `asset_download`	Review inline, then save the file

Model IDs and the parameter facts below were live

search

hits at authoring time. Re-discover them each time: availability differs per team and evolves.

```
model_scenario-skybox-flux
```
: text to 360 panorama, a style preset enum (count drifts; read it from the schema), automatic seam and pole correction, optional reference image with a strength slider.
```
model_scenario-skybox-gpt
```
: text to 360 panorama guided by up to 10 reference images, quality presets, width and height up to 3840 px.
```
model_hunyuan-world-image-to-skybox
```
: one photo of a place to a seamless 360 skybox.
```
model_sc-upscale-flux-skybox
```
: 2x to 8x skybox upscale that preserves the seamless wrap.


search
model_schema_get
dry_run=true
pending_job_ids
asset_display

步骤	工具	用途
1	`search` (target="models", query="skybox", public=true)	查找当前可用的skybox模型
2	`model_schema_get`	获取所选模型的精确参数约定
3	`model_run`	`dry_run=true` 用于估算费用； `wait=false` 用于批量处理
4	`jobs_wait`	超时时使用 `pending_job_ids` 重新调用
5	`asset_display` / `asset_download`	在线预览，然后保存文件

以下模型ID和参数信息是撰写本文时

search

的实时结果。每次使用时请重新查找：模型可用性因团队而异，且会不断更新。

```
model_scenario-skybox-flux
```
：文本生成360全景图，包含风格预设枚举（数量会变动，请从schema中读取），自动校正拼接缝和极点，支持可选参考图及强度调节滑块。
```
model_scenario-skybox-gpt
```
：支持最多10张参考图引导的文本生成360全景图，提供质量预设，宽高最高可达3840像素。
```
model_hunyuan-world-image-to-skybox
```
：将一处场景的单张照片转换为无缝360 skybox。
```
model_sc-upscale-flux-skybox
```
：支持2倍至8倍的skybox放大，可保留无缝拼接特性。

Worked example: generate, iterate, export

实操示例：生成、迭代、导出

Example: a stylized forest skybox for a game scene.

```
search
```
with target="models", query="skybox", public=true. Pick a text-to-skybox model, for example
```
model_scenario-skybox-flux
```
.
```
model_schema_get
```
for that model. Expect fields like prompt, style, negativePrompt, image, strength, numOutputs, geometryEnforcement, seed. negativePrompt is inert unless negativePromptStrength is above 0.
```
model_run
```
with parameters={"prompt": "ancient pine forest at dawn, mist between trunks, god rays", "style": "cinematic", "numOutputs": 2}. Take style preset names from the schema response.
```
dry_run=true
```
first prices the run; launch with
```
wait=false
```
.
```
jobs_wait
```
with job_ids=[the returned job_id]. Its ~180s timeout is not an error: re-call with the returned
```
pending_job_ids
```
as
```
job_ids
```
, never a second
```
model_run
```
. Then
```
asset_display
```
each output.
Iterate on mood: copy the seed from the best result and change only style (cinematic, oil-painting, cyberpunk, and more). To keep composition while shifting look, pass the favorite as image with low strength (0.2 to 0.4). To steer mood from concept art instead, switch to
```
model_scenario-skybox-gpt
```
and pass referenceImages.
Export: run
```
model_sc-upscale-flux-skybox
```
with image=asset_id and the smallest upscaleFactor that reaches target (the upscale can cost several times the generation, and its
```
dry_run
```
can only run once the input asset exists, so the chain cannot be priced up front). baseModel defaults to FLUX.1-dev (stylized); a Krea-based realism option exists, and strength defaults to 0.6, which invents detail: lower it when the goal is the same panorama at higher resolution. Read the exact allowed values from
```
model_schema_get
```
before switching. Then
```
asset_download
```
the final asset.
Verify the seam at no cost: compare the saved PNG's leftmost and rightmost pixel columns; on a seamless panorama they differ by near zero while columns a quarter-turn apart differ by an order of magnitude more.

Engine format notes: Skybox Flux outputs equirectangular panoramas; keep the default sizing (1536x768 at authoring time) and read the real dimensions off the returned asset rather than assuming them. Skybox GPT's catalog lists equirectangular 2:1 plus cubemap strip 6:1 and cubemap cross 4:3 layouts, but its schema exposes only width and height, so confirm the layout contract with

model_schema_get

before relying on a cubemap layout. Beyond flat backdrops, the same search surfaces

model_hunyuan-world-skybox-to-splat

, and a separate search (query="world") finds the Marble world models; both turn a finished panorama into a navigable 3D Gaussian splat scene, the pipeline the

scenario-3d-worlds

skill teaches end to end.

示例：为游戏场景生成风格化森林skybox。

调用
```
search
```
，参数为target="models"、query="skybox"、public=true。选择一个文本生成skybox的模型，例如
```
model_scenario-skybox-flux
```
。
调用
```
model_schema_get
```
获取该模型的信息。预期包含prompt、style、negativePrompt、image、strength、numOutputs、geometryEnforcement、seed等字段。negativePrompt默认无效，除非negativePromptStrength大于0。
调用
```
model_run
```
，参数为{"prompt": "黎明时分的古松树林，树干间雾气弥漫，阳光穿透", "style": "cinematic", "numOutputs": 2}。请从schema响应中获取风格预设名称。先使用
```
dry_run=true
```
估算运行费用；然后使用
```
wait=false
```
启动任务。
调用
```
jobs_wait
```
，参数为job_ids=[返回的job_id]。其约180秒的超时不是错误：请使用返回的
```
pending_job_ids
```
作为
```
job_ids
```
重新调用，切勿再次调用
```
model_run
```
。然后调用
```
asset_display
```
查看每个输出。
氛围迭代：从最佳结果中复制seed，仅修改style（cinematic、oil-painting、cyberpunk等）。如果要在改变风格的同时保留构图，请将最喜欢的结果作为image传入，并设置较低的strength（0.2到0.4）。如果想要基于概念图调整氛围，请切换到
```
model_scenario-skybox-gpt
```
并传入referenceImages。
导出：运行
```
model_sc-upscale-flux-skybox
```
，参数为image=asset_id，并使用能达到目标的最小upscaleFactor（放大的费用可能是生成费用的数倍，且其
```
dry_run
```
仅在输入资产存在后才能运行，因此无法预先估算整个流程的费用）。baseModel默认为FLUX.1-dev（风格化）；还有一个基于Krea的写实风格选项，strength默认为0.6，该值会生成新细节：如果目标是获得更高分辨率的同一张全景图，请降低该值。切换前请通过
```
model_schema_get
```
读取确切的允许值。然后调用
```
asset_download
```
下载最终资产。
免费验证拼接缝：对比保存的PNG文件最左侧和最右侧的像素列；无缝全景图的这两列差异几乎为零，而相隔四分之一圈的像素列差异会大一个数量级。

引擎格式注意事项：Skybox Flux输出equirectangular全景图；请保持默认尺寸（撰写本文时为1536x768），并从返回的资产中读取实际尺寸，而非假设尺寸。Skybox GPT的目录列出了equirectangular 2:1、cubemap strip 6:1和cubemap cross 4:3布局，但其schema仅暴露宽和高，因此在使用cubemap布局前，请通过

model_schema_get

确认布局约定。除了平面背景外，同样的搜索还会找到

model_hunyuan-world-skybox-to-splat

，另外搜索（query="world"）可以找到Marble世界模型；两者都可以将成品全景图转换为可导航的3D Gaussian splat场景，该完整流程在

scenario-3d-worlds

skill中有端到端的讲解。

Common mistakes

常见错误

Prompting a generic image model for a "360 panorama": edges will not wrap and poles smear. Use a dedicated skybox model.
Upscaling with a generic upscaler: it breaks continuity at the wrap seam. Use the skybox upscaler.
Hardcoding model IDs in scripts or docs: re-discover with
```
search
```
; the catalog changes.
Fighting seam or pole distortion through prompt wording on Skybox Flux: raise geometryEnforcement above 0 instead, and only when distortion is actually visible.
Requesting a non 2:1 width to height ratio on Skybox GPT while expecting equirectangular output: keep 2:1 (for example 2048x1024) for correct 360 viewing.
Skipping
```
model_schema_get
```
: skybox models carry model-specific fields (style, geometryEnforcement, quality) that generic assumptions miss.

向通用图像模型提示生成「360全景图」：边缘无法拼接，极点会模糊。请使用专用的skybox模型。
使用通用超分工具放大：会破坏拼接缝处的连续性。请使用skybox专用超分工具。
在脚本或文档中硬编码模型ID：请使用
```
search
```
重新查找；模型目录会变动。
在Skybox Flux中通过调整prompt措辞来解决拼接缝或极点变形问题：请改为将geometryEnforcement提高到0以上，且仅在变形确实可见时才这么做。
在Skybox GPT中请求非2:1的宽高比却期望得到equirectangular输出：请保持2:1的比例（例如2048x1024）以获得正确的360观看效果。
跳过
```
model_schema_get
```
步骤：skybox模型包含特定于模型的字段（style、geometryEnforcement、quality），通用假设会遗漏这些字段。