scenario-reve

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Scenario Reve Image

Overview

概述

Reve, Reve AI's image family on Scenario, splits into two members that share a brand and little else: v2.1 is a reasoning-driven generator and instruction editor (a top-2 text-to-image arena entry at authoring time), Remix a fast, cheap compositor that blends several images into one. Discover them with

search

and treat

model_schema_get

as the contract: they disagree on field names, prompt caps, and ratio lists.

Connection and the core loop: see the

scenario

skill in this repo; 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.

Reve是Reve AI在Scenario平台上的图像模型系列，旗下有两款仅共享品牌、其余差异极大的模型：v2.1是推理驱动的生成器与指令编辑器（在本文撰写时位列文生图竞技场前二），Remix则是快速、低成本的合成器，可将多张图像混合为一张。使用

search

查找这些模型，并以

model_schema_get

的返回结果为准：两者在字段名称、提示词长度上限、宽高比列表上均不相同。

连接方式与核心流程：参见本仓库中的

scenario

技能；与模型无关的通用图像处理工作：参见

scenario-image

技能。如果你的可用技能中缺少此处提到的关联技能，请提示用户安装（命令：

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

）；若处于无人值守模式，请基于工具schema继续操作并标记缺失的技能。

Quick reference

快速参考

At authoring time (the live schema wins):

Member	Capabilities	References	Prompt cap	Aspect ratios
Reve v2.1	txt2img, img2img	`references` , optional	4000	18 values, 4:1 through 1:4, default auto
Reve Remix	img2img	`referenceImages` , required, 1 to 6	2560	7 values, 16:9 through 9:16, default 3:2

Three parameters each: the prompt, the reference array,

aspectRatio

. Neither member has a mask, seed, width and height, negative prompt, or batch-count field, so an edit is scoped by wording and size comes only as a ratio.

auto

(v2.1 only) picks a fitting shape; pass an explicit ratio when the deliverable demands one.

Pick by job. v2.1 covers generation from a prompt alone and precise edits: swap the text on a poster, recolor a product, change a material, relight or restyle, merge references while every subject stays consistent. Remix blends photos, illustrations, or design assets into one cohesive composite. Where either would do,

dry_run

both: at authoring time a v2.1 run cost several times a Remix run and took over a minute at p50 against Remix's twenty-odd seconds.

在本文撰写时（以实时schema为准）：

模型成员	支持能力	参考图配置	提示词长度上限	宽高比
Reve v2.1	txt2img, img2img	`references` ，可选	4000字符	18种，范围4:1至1:4，默认auto
Reve Remix	img2img	`referenceImages` ，必填，1-6张	2560字符	7种，范围16:9至9:16，默认3:2

两款模型均包含三个参数：提示词、参考图数组、

aspectRatio

。两款模型都没有mask、seed、宽高设置、负面提示词或批量次数字段，因此编辑范围由提示词的表述决定，尺寸仅能通过宽高比控制。

auto

（仅v2.1支持）会自动选择合适的比例；如果交付物有明确尺寸要求，请传入指定的宽高比。

根据任务需求选型。v2.1支持纯提示词生成和精准编辑：替换海报文字、产品重着色、更换材质、重新打光或调整风格、合并参考图且保持所有主体一致。Remix可将照片、插画或设计素材混合为一张协调的合成图。如果两款模型都能完成任务，可同时运行

dry_run

对比：在本文撰写时，v2.1的单次运行成本是Remix的数倍，p50延迟超过1分钟，而Remix仅需20多秒。

Frame tags wire v2.1 references

帧标签用于关联v2.1的参考图

In a v2.1 prompt, a reference is addressed with a

<frame>N</frame>

tag, numbered from 0 in the order the assets appear in

references

: the first is

<frame>0</frame>

, the second

<frame>1</frame>

. Use the tags whenever more than one reference is in play; without them the model guesses which image you mean. Remix documents no tag syntax: name subjects in plain words ("the bottle from the first image, the lighting from the second").

在v2.1的提示词中，使用

<frame>N</frame>

标签来指代参考图，编号从0开始，对应

references

数组中素材的顺序：第一张参考图是

<frame>0</frame>

，第二张是

<frame>1</frame>

。当使用多张参考图时必须使用这些标签；如果没有标签，模型会自行猜测你指代的是哪张图。Remix没有相关标签语法的文档：请用平实的语言描述主体（"第一张图里的瓶子，第二张图的光线"）。

Edits are instructions, not masks

编辑是指令式而非蒙版式

Neither member takes a mask, so the prompt carries the whole edit: state what changes and what must stay untouched. v2.1's reasoning rewards specific plain-language instructions over keyword lists, and its 4000-character budget leaves room for constraints. For masked inpainting on an exact region, use a model that exposes a

mask

field (see

scenario-image

两款模型都不支持蒙版，因此所有编辑需求都要通过提示词传达：明确说明哪些内容需要修改、哪些必须保持不变。v2.1的推理能力更擅长处理具体的自然语言指令，而非关键词列表，且其4000字符的长度上限足以容纳各种约束条件。如果需要对精确区域进行蒙版式修复，请使用支持

mask

字段的模型（参见

scenario-image

技能）。

Worked example: swapping the text on a poster

实操示例：替换海报上的文字

```
search
```
with
```
target="models"
```
,
```
query="reve"
```
,
```
public=true
```
. At authoring time the hits were
```
model_reve-v2-1
```
(generation and instruction edits) and
```
model_reve-remix
```
(multi-image blends); re-discover each session.
```
model_schema_get
```
on the pick: field names, caps, and the ratio list before anything else.
```
upload_asset
```
the poster (see the
```
scenario
```
skill) to get an asset id.

model_run

with that

model_id

dry_run=true

, and

parameters={"prompt": "Replace the headline on the poster in <frame>0</frame> with 'SUMMER SALE', matching the original font, perspective, and lighting. Change nothing else.", "references": ["asset_x"], "aspectRatio": "auto"}

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
```
to check the swap, then
```
asset_download
```
to save.

调用
```
search
```
，传入
```
target="models"
```
、
```
query="reve"
```
、
```
public=true
```
。在本文撰写时，返回结果为
```
model_reve-v2-1
```
（用于生成和指令式编辑）和
```
model_reve-remix
```
（用于多图混合）；每次会话都需要重新查找模型。
对选中的模型调用
```
model_schema_get
```
：优先确认字段名称、长度上限和宽高比列表。
调用
```
upload_asset
```
上传海报（参见
```
scenario
```
技能），获取素材ID。

调用

model_run

，传入对应的

model_id

、

dry_run=true

以及如下

parameters

以获取成本估算：

{"prompt": "Replace the headline on the poster in <frame>0</frame> with 'SUMMER SALE', matching the original font, perspective, and lighting. Change nothing else.", "references": ["asset_x"], "aspectRatio": "auto"}

再次调用
```
model_run
```
，传入
```
wait=false
```
，然后使用返回的任务ID调用
```
jobs_wait
```
；如果超时，使用
```
pending_job_ids
```
重新调用，绝对不要重复发起
```
model_run
```
。
调用
```
asset_display
```
检查文字替换效果，然后调用
```
asset_download
```
保存结果。

Common mistakes

常见错误

Counting frames from 1:
```
<frame>1</frame>
```
is the second reference; the first is
```
<frame>0</frame>
```
.
Carrying a field name across members: v2.1 takes
```
references
```
, Remix takes
```
referenceImages
```
; each schema lists only its own.
Running Remix without references:
```
referenceImages
```
is required (1 to 6 at authoring time); prompt-only generation belongs to v2.1.
Reusing a v2.1 prompt on Remix without checking length: 4000 characters overflow its 2560 cap.
Sending width, height, a mask, or a seed: none exists on either member;
```
aspectRatio
```
is the only size control.
Treating a v2.1
```
jobs_wait
```
timeout as an error: p50 latency topped a minute at authoring time; re-call with
```
pending_job_ids
```
.

帧编号从1开始计数：
```
<frame>1</frame>
```
对应第二张参考图；第一张是
```
<frame>0</frame>
```
。
跨模型混用字段名：v2.1使用
```
references
```
，Remix使用
```
referenceImages
```
；每个模型的schema仅列出自身支持的字段。
运行Remix时不传参考图：
```
referenceImages
```
是必填项（本文撰写时要求1-6张）；纯提示词生成请使用v2.1。
未检查长度就将v2.1的提示词用于Remix：4000字符会超出Remix的2560字符上限。
传入width、height、mask或seed参数：两款模型都不支持这些字段；
```
aspectRatio
```
是唯一的尺寸控制参数。
将v2.1的
```
jobs_wait
```
超时视为错误：本文撰写时v2.1的p50延迟超过1分钟；请使用
```
pending_job_ids
```
重新调用。