scenario-ideogram

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Scenario Ideogram Image

Scenario Ideogram 图像模型

Overview

概述

Ideogram's image family on Scenario is five single-purpose members, not one model with modes: V4 for generation where in-image text must read, V3 Generate Transparent for native alpha output, V3 Layerize Text for turning a flat graphic into editable text layers, Character for one identity held across scenes, and Remove Background for cutouts. Mechanically they agree on almost nothing: the same concept changes name, casing, and allowed values between members, so discover each with

search

and treat its

model_schema_get

as the contract.

Connection and the core loop: see the

scenario

skill in this repo; model-agnostic image work (sizing families, reference cardinality, masks): 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.

Scenario平台上的Ideogram图像模型家族由5个单一用途的成员组成，而非一个支持多种模式的单一模型：V4用于生成需清晰呈现图像内文本的内容，V3 Generate Transparent用于原生alpha通道输出，V3 Layerize Text用于将平面图形转换为可编辑文本层，Character用于在不同场景中保持同一角色身份，Remove Background用于抠图。从运行机制来看，它们几乎没有共通之处：同一个概念在不同成员中的名称、大小写和允许取值都可能不同，因此请使用

search

查找每个模型，并将其

model_schema_get

的结果作为使用规范。

连接方式与核心流程：参见本仓库中的

scenario

skill；与模型无关的图像处理（尺寸系列、参考数量、蒙版）：参见

scenario-image

skill。如果你的可用skill中缺少此处提到的同类skill，请让用户安装（命令：

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

）；无人值守场景下，请基于工具schema继续操作并标记缺失项。

Quick reference

快速参考

Pick the member by the job (names from the live schemas, caps at authoring time):

Member	Job	Inputs that matter
V4	Text-heavy generation	`prompt` , `imageSize` (six presets), `renderingSpeed` (TURBO, BALANCED, QUALITY), `enablePromptExpansion` , `acceleration` , `numOutputs` (up to 4), `seed`
V3 Generate Transparent	Native alpha output	`prompt` , `negativePrompt` , `aspectRatio` (15 ratios), `renderingSpeed` (adds FLASH), `expandPrompt` , `numOutputs` (up to 8), `seed`
V3 Layerize Text	Flat graphic to layers	`image` , optional `prompt` , `fontName` or `fontFile` per tier (H1, H2, Body, Small), `seed`
Character	Same character, new scenes	`prompt` , `characterReferenceImage` , `styleType` (Auto, Fiction, Realistic), `aspectRatio` or `resolution` , `image` plus `mask` , `renderingSpeed` (Default, Turbo, Quality), `seed`
Remove Background	Cutout to transparent PNG	`image` (10MB cap)

Nothing transfers between members. Sizing is

imageSize

presets on V4 (

square_hd

is the largest),

aspectRatio

ratios on V3 Transparent and Character, and on Character alone an exact

resolution

that overrides the ratio.

renderingSpeed

values are uppercase on the two generators and Title case on Character, whose middle tier is Default, not BALANCED. Only V3 Transparent takes a

negativePrompt

. V4's

acceleration

buys speed on top of the tier at some quality cost; its default

none

is right for finals. Per-asset cost differs several-fold across members and moves with every

cost_impact: true

field, so

dry_run

the same job on the candidates before a batch.

根据任务选择对应成员（名称来自实时schema，编写时的大小写如下）：

成员	用途	关键输入参数
V4	大文本量生成	`prompt` , `imageSize` （6种预设）, `renderingSpeed` （TURBO, BALANCED, QUALITY）, `enablePromptExpansion` , `acceleration` , `numOutputs` （最多4个）, `seed`
V3 Generate Transparent	原生alpha通道输出	`prompt` , `negativePrompt` , `aspectRatio` （15种比例）, `renderingSpeed` （新增FLASH）, `expandPrompt` , `numOutputs` （最多8个）, `seed`
V3 Layerize Text	平面图形转分层	`image` , 可选 `prompt` , 每个层级（H1, H2, Body, Small）对应的 `fontName` 或 `fontFile` , `seed`
Character	同角色，新场景	`prompt` , `characterReferenceImage` , `styleType` （Auto, Fiction, Realistic）, `aspectRatio` 或 `resolution` , `image` 加 `mask` , `renderingSpeed` （Default, Turbo, Quality）, `seed`
Remove Background	抠图生成透明PNG	`image` （最大10MB）

各成员之间没有通用参数。尺寸设置方面，V4使用

imageSize

预设（

square_hd

为最大尺寸），V3 Transparent和Character使用

aspectRatio

比例，而只有Character支持精确的

resolution

参数，该参数会覆盖比例设置。

renderingSpeed

的取值在两个生成器中为全大写，在Character中为标题大小写，其中间档位是Default而非BALANCED。只有V3 Transparent支持

negativePrompt

。V4的

acceleration

参数可在档位基础上进一步提升速度，但会牺牲一定质量；其默认值

none

适合最终输出。不同成员的单素材成本相差数倍，且每个标记为

cost_impact: true

的字段都会影响成本，因此批量处理前请在候选模型上对同一任务执行

dry_run

。

Exact text wants expansion off

需精确呈现文本时请关闭提示扩展

Both generators rewrite the prompt before generating by default, which helps a short exploratory prompt and hurts the family's specialty: the rewrite can paraphrase the exact copy that must render. When the image carries wording, disable expansion (

enablePromptExpansion: false

on V4,

expandPrompt: false

on V3 Transparent), quote each piece of copy, and give it a place and a style ('the headline reads "GRAND OPENING" in bold condensed capitals across the top'). V4 renders multilingual text the same way: quote it.

两个生成器默认都会在生成前重写prompt，这对简短的探索性prompt有帮助，但会损害该模型家族的核心优势：重写可能会改写必须准确呈现的文案。当图像需要包含特定文字时，请关闭扩展功能（V4设置

enablePromptExpansion: false

，V3 Transparent设置

expandPrompt: false

），将每段文案用引号括起来，并说明其位置和样式（例如：'the headline reads "GRAND OPENING" in bold condensed capitals across the top'）。V4生成多语言文本的方式相同：用引号括起来。

Two routes to transparency

两种实现透明背景的方式

Native: V3 Generate Transparent writes the alpha channel directly, so icons, stickers, and UI elements arrive compositing-ready. Cutout: generate on V4, then run Remove Background on the result; it reconstructs edge pixels with partial transparency rather than segmenting, so hair, fur, and glass survive. Go native when the asset is designed as an isolated element; go cutout when typography or overall quality leads, since at authoring time V4 scored well above the V3 line on public text-to-image arena ratings. Prompting "transparent background" on V4 does nothing: its outputs are opaque.

原生方式：V3 Generate Transparent直接写入alpha通道，因此图标、贴纸和UI元素可直接用于合成。抠图方式：先用V4生成，再对结果运行Remove Background；它会重建带有部分透明度的边缘像素而非简单分割，因此发丝、毛发和玻璃边缘都能保留。当素材设计为独立元素时选择原生方式；当排版或整体质量更重要时选择抠图方式，因为在编写本文时，V4在公开文本生成图像评测中的得分远高于V3系列。在V4中提示"transparent background"是无效的：其输出默认是不透明的。

Worked example: a localizable poster

实操示例：可本地化海报

```
search
```
with
```
target="models"
```
,
```
query="ideogram"
```
,
```
public=true
```
. Members return as separate hits; match by name, e.g.
```
model_ideogram-v4
```
for typography (a live hit at authoring time: re-discover each session).
```
model_schema_get
```
with that id: field names, allowed values, and defaults before anything else.

model_run

with that

model_id

dry_run=true

, and

parameters={"prompt": "Retro travel poster, warm dusk palette. The headline reads \"KYOTO IN BLOOM\" in bold serif across the top; caption \"April 2027\" bottom right.", "imageSize": "portrait_16_9", "renderingSpeed": "QUALITY", "enablePromptExpansion": false, "numOutputs": 2}

for the cost estimate.

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
```
the outputs and pick one.
To make the copy editable, discover the Layerize member the same way (
```
model_ideogram-v3-layerize-text
```
at authoring time), read its schema, and
```
model_run
```
with
```
parameters={"image": "<poster asset id>"}
```
: a generated asset's id feeds a file input directly, no re-upload. The output is a text-erased base plus text blocks with role, position, and content.
```
jobs_wait
```
, then
```
asset_display
```
and
```
asset_download
```
.

调用
```
search
```
，参数为
```
target="models"
```
、
```
query="ideogram"
```
、
```
public=true
```
。各成员会作为独立结果返回；按名称匹配，例如用于排版的
```
model_ideogram-v4
```
（编写本文时的有效结果：每次会话都需重新查找）。
使用该id调用
```
model_schema_get
```
：先确认字段名、允许取值和默认值，再进行其他操作。

使用该

model_id

调用

model_run

，设置

dry_run=true

，并传入

parameters={"prompt": "Retro travel poster, warm dusk palette. The headline reads \"KYOTO IN BLOOM\" in bold serif across the top; caption \"April 2027\" bottom right.", "imageSize": "portrait_16_9", "renderingSpeed": "QUALITY", "enablePromptExpansion": false, "numOutputs": 2}

以估算成本。

再次调用
```
model_run
```
，设置
```
wait=false
```
，然后使用返回的job id调用
```
jobs_wait
```
，超时则使用
```
pending_job_ids
```
重新调用，切勿重复调用
```
model_run
```
。
调用
```
asset_display
```
查看输出结果并选择一张。
若要让文案可编辑，用同样的方式查找Layerize成员（编写本文时为
```
model_ideogram-v3-layerize-text
```
），读取其schema，然后调用
```
model_run
```
并传入
```
parameters={"image": "<poster asset id>"}
```
：已生成素材的id可直接作为文件输入，无需重新上传。输出结果为擦除文字的底图加上带有角色、位置和内容的文本块。
调用
```
jobs_wait
```
，然后执行
```
asset_display
```
和
```
asset_download
```
。

Common mistakes

常见错误

Carrying one member's block to the next:
```
imageSize
```
and
```
enablePromptExpansion
```
are V4's; V3 Transparent wants
```
aspectRatio
```
and
```
expandPrompt
```
.
Leaving expansion on with exact copy in the prompt: the rewrite can change the words before the model sees them.
Prompting a transparent background on V4: generate on V3 Transparent or chain Remove Background instead.
Inpainting on Character with
```
image
```
but no
```
mask
```
: they go together (black repaints, white stays, mask resized to the image),
```
characterReferenceImage
```
stays required, and sizing fields are ignored while inpainting.
Setting
```
fontNameH1
```
and
```
fontFileH1
```
together on Layerize: a tier's font comes from a font name or a font file, never both.
```
upload_asset
```
has no font kind, so with only a local font file use the font name route and flag the gap.

混用不同成员的参数：
```
imageSize
```
和
```
enablePromptExpansion
```
是V4的参数；V3 Transparent需要使用
```
aspectRatio
```
和
```
expandPrompt
```
。
prompt中包含精确文案时仍开启扩展功能：重写可能会在模型处理前修改文字内容。
在V4中提示生成透明背景：请改用V3 Transparent生成，或串联使用Remove Background。
在Character上进行inpainting时传入
```
image
```
但未传
```
mask
```
：二者必须配合使用（黑色区域重绘，白色区域保留，mask会调整为与图像匹配），
```
characterReferenceImage
```
仍为必填项，且inpainting期间会忽略尺寸相关字段。
在Layerize中同时设置
```
fontNameH1
```
和
```
fontFileH1
```
：一个层级的字体只能来自字体名称或字体文件，不能同时使用两者。
```
upload_asset
```
没有字体类型，因此如果只有本地字体文件，请使用字体名称路径并标记该缺失项。