scenario-grok-imagine-image

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Scenario Grok Imagine Images

Scenario Grok Imagine 图像模型

Overview

概述

Grok Imagine, xAI's image family on Scenario, puts generation and editing in every member: the same model reads an empty

referenceImages

as text-to-image and a filled one as an edit, so the inputs pick the mode and the member picks the tier. Members ship side by side, agreeing on the prompt contract and disagreeing on knobs and price, so discover them with

search

and treat

model_schema_get

as the contract. The Grok Imagine video members belong to the

scenario-grok-imagine-video

skill.

Connection and the core loop: see the

scenario

skill; model-agnostic image work: the

scenario-image

skill. 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.

Grok Imagine是xAI在Scenario平台上的图像模型家族，其所有成员均支持生成和编辑功能：同一个模型会将空的

referenceImages

识别为文生图模式，将有内容的

referenceImages

识别为编辑模式，因此输入决定模式，模型成员决定等级。各模型成员并行提供，遵循统一的prompt规范，但参数和定价不同，因此请使用

search

发现模型，并将

model_schema_get

的返回结果作为约定依据。Grok Imagine的视频模型成员属于

scenario-grok-imagine-video

技能。

连接方式与核心流程：请参考

scenario

技能；与模型无关的图像处理工作：请参考

scenario-image

技能。如果你的可用技能中缺少此处提到的同类技能，请让用户安装（命令：

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

）；若处于无人值守模式，请根据工具schema继续操作并标记缺失项。

Quick reference

快速参考

At authoring time (caps are per member, read the live schema):

Member	Aspect ratios	Notes
2.0	14 values up to 20:9 and 9:20, default auto	adds `quality` (low, medium); strongest text and edits
Quality	same 14, default 1:1	fast at 2k; price near flat across variants
Image	10 values, 2:1 through 1:2	cheapest per image

The rest is shared:

prompt

up to 10000 characters,

referenceImages

capped at 3,

numOutputs

1 to 10,

resolution

at 1k or 2k. No seed, mask, negative prompt, or pixel sizing exists anywhere in the family: size is ratio times resolution tier, and

aspectRatio

decides what is in frame, not just the crop. In editing mode

aspectRatio

is ignored entirely and the output follows the source's proportions. The price spread is why the older members stay live: one image on the unversioned member cost a fraction of a versioned run at authoring time, and 2.0's 2k text-to-image ran well over a minute where the others took seconds. So iterate cheap and fast (the unversioned member, or 2.0 at

quality: "low"

and 1k), spend a medium 2k run on keepers, and

dry_run

two members on the same job before a batch:

numOutputs

quality

resolution

, and references all move price.

在编写本文时（限制因模型成员而异，请以实时schema为准）：

模型成员	宽高比	说明
2.0	支持14种，最高20:9和9:20，默认自动	新增 `quality` 参数（低、中两档）；文字生成和编辑能力最强
Quality	同样支持14种，默认1:1	2k分辨率下速度快；不同变体的价格基本持平
Image	支持10种，范围从2:1到1:2	单张图像成本最低

其余参数为所有成员共享：

prompt

最多10000字符，

referenceImages

最多3张，

numOutputs

范围1-10，

resolution

支持1k或2k。整个模型家族均不支持seed、mask、negative prompt或像素级尺寸设置：图像尺寸由宽高比乘以分辨率等级决定，

aspectRatio

不仅决定裁剪，还决定画面内容。在编辑模式下，

aspectRatio

会被完全忽略，输出将遵循源图像的比例。价格差异是旧模型成员仍在服役的原因：在编写本文时，使用无版本成员生成一张图像的成本仅为版本化模型的一小部分，而2.0版本的2k文生图运行时间远超一分钟，其他模型仅需几秒。因此，请使用低成本高速度的方式迭代（无版本成员，或2.0版本设置

quality: "low"

且分辨率为1k），对满意的结果使用中等质量的2k运行生成，批量处理前先对两个模型成员使用

dry_run

测试同一任务：

numOutputs

、

quality

、

resolution

和参考图都会影响价格。

Write the copy, not a vibe

撰写具体文案，而非模糊感觉

The family's edge is legible in-image text: posters, packaging, infographics, memes. It only pays when the prompt carries the copy verbatim: wrap every literal string in quotes, give each a role ("the title reads ...", "a line beneath reads ..."), keep strings short, and close with "no other text", or the model invents extra labels. Real numbers and statistics go in as quoted copy too, since invented figures look plausible. Render small type at 2k and proofread in

asset_display

before building on the result. For the rest, write a short natural-language brief, subject, setting, composition, light, finish, in 30 to 80 words: quality adjectives ("8K", "masterpiece") are ignored and negatives mostly too, so state positives ("clean empty background", not "no clutter") and keep one aesthetic per prompt.

该模型家族的优势在于可生成清晰可读的图内文字：适用于海报、包装、信息图、表情包。只有当prompt中包含逐字准确的文案时，才能发挥这一优势：将每个字面字符串用引号括起来，为每个字符串指定角色（例如“标题内容为……”、“下方一行小字为……”），保持字符串简短，并以“no other text”结尾，否则模型会生成额外的标签。真实数字和统计数据也需要作为带引号的文案输入，因为模型编造的数字看起来会很像真的。小字号内容请使用2k分辨率渲染，并在

asset_display

中校对后再基于结果进行后续操作。其余内容请撰写简短的自然语言描述，包括主题、场景、构图、光线、质感，长度在30-80词之间：质量形容词（如“8K”、“masterpiece”）会被忽略，否定词也大多会被忽略，因此请陈述正面要求（例如“干净空旷的背景”，而非“没有杂物”），且每个prompt只保持一种美学风格。

Edits are instructions, no mask exists

编辑为指令式，不支持mask

Editing is prompt-driven: pass up to three

referenceImages

and describe the change. Two habits keep it surgical: describe only what changes and name what stays ("keep the pose, background, and lighting unchanged"), and chain one edit per run instead of stacking several. With more than one reference, state each image's role (subject, style, scene) so the blend lands where intended. The output inherits the source's proportions, so reframe before the edit, never with

aspectRatio

编辑功能由prompt驱动：最多可传入3张

referenceImages

并描述修改内容。两个习惯可以让编辑更精准：只描述需要修改的部分，并明确说明保留的内容（例如“保持姿势、背景和灯光不变”），每次运行只做一次编辑，不要堆叠多个编辑请求。如果使用多张参考图，请说明每张图像的作用（主体、风格、场景），以便融合效果符合预期。输出图像会继承源图像的比例，因此请在编辑前调整好源图像的构图，不要使用

aspectRatio

来调整。

Worked example: a branded label onto a product photo

实操示例：为产品照片添加品牌标签

```
search
```
with
```
target="models"
```
,
```
query="grok imagine image"
```
,
```
public=true
```
. Prefer the newest non-deprecated hit, e.g.
```
model_xai-grok-imagine-image-2-0
```
(a live hit at authoring time: re-discover each session).
```
model_schema_get
```
with that id: the reference cap,
```
quality
```
and
```
resolution
```
values, and the ratio list before anything else.
```
upload_asset
```
the product photo (see the
```
scenario
```
skill) to get its asset id.

model_run

with that

model_id

dry_run=true

, and the exact

parameters={"prompt": "Add a minimalist front label to the bottle where the title reads \"LUMEN\" in a clean white serif and a smaller line beneath reads \"Sparkling Citrus Soda\". Keep the bottle shape, glass tint, background, shadows, and lighting unchanged, no other text.", "referenceImages": ["asset_abc"], "quality": "medium", "resolution": "2k"}

; re-estimate after any change to

quality

resolution

numOutputs

, or references.

Repeat
```
model_run
```
with
```
wait=false
```
, then
```
jobs_wait
```
with the returned job id, re-called with
```
pending_job_ids
```
on timeout, never a second
```
model_run
```
.
```
asset_display
```
and proofread the label at full size, then
```
asset_download
```
the keeper.

使用
```
search
```
方法，设置
```
target="models"
```
、
```
query="grok imagine image"
```
、
```
public=true
```
。优先选择最新的未弃用结果，例如
```
model_xai-grok-imagine-image-2-0
```
（编写本文时的有效结果：每次会话都需要重新查询）。
使用该id调用
```
model_schema_get
```
：首先查看参考图数量上限、
```
quality
```
和
```
resolution
```
的可选值，以及宽高比列表。
上传产品照片（参考
```
scenario
```
技能）调用
```
upload_asset
```
，获取其资产id。

使用该

model_id

调用

model_run

，设置

dry_run=true

，并传入准确的

parameters={"prompt": "Add a minimalist front label to the bottle where the title reads \"LUMEN\" in a clean white serif and a smaller line beneath reads \"Sparkling Citrus Soda\". Keep the bottle shape, glass tint, background, shadows, and lighting unchanged, no other text.", "referenceImages": ["asset_abc"], "quality": "medium", "resolution": "2k"}

；如果修改了

quality

、

resolution

、

numOutputs

或参考图，需要重新估算。

再次调用
```
model_run
```
，设置
```
wait=false
```
，然后使用返回的任务id调用
```
jobs_wait
```
，如果超时则使用
```
pending_job_ids
```
重新调用，切勿重复调用
```
model_run
```
。
使用
```
asset_display
```
全尺寸校对标签，然后调用
```
asset_download
```
下载满意的结果。

Common mistakes

常见错误

Setting
```
aspectRatio
```
on an edit: it is ignored and the output keeps the source's proportions; reframe the source first.
Describing copy loosely ("a catchy tagline"): text comes back garbled or invented; quote exact strings and end with "no other text".
Negative prompts: largely ignored; state the positive instead.
A bare string where the schema says array: one reference still goes as
```
["asset_x"]
```
.
Carrying
```
quality
```
beyond 2.0: at authoring time no other member had the field; check the live schema before reusing parameters across members.
Fine print at 1k: small type blurs; render text-heavy finals at 2k.

在编辑时设置
```
aspectRatio
```
：该参数会被忽略，输出将保持源图像的比例；请先调整源图像的构图。
模糊描述文案（例如“一个吸引人的标语”）：返回的文字会混乱或被编造；请引用准确的字符串，并以“no other text”结尾。
Negative prompt：基本会被忽略；请改为陈述正面要求。
schema要求数组时传入裸字符串：即使只有一张参考图，也要以
```
["asset_x"]
```
的形式传入。
在2.0以外的版本中使用
```
quality
```
参数：编写本文时，其他模型成员均不支持该字段；在不同成员间复用参数前，请检查实时schema。
使用1k分辨率渲染小字：小字号会模糊；文字较多的最终成品请使用2k分辨率渲染。