DeepSeek Harness 插件

dsh-fal-image-gen

FAL image generation for DeepSeek Harness: a generate_image tool backed by the FAL queue API (default model fal-ai/flux-2/klein/9b). Images render directly in the conversation, are saved to the(英文原文)

跳到安装方式

来源信息

GitHub 仓库
GooDAnDReaDY/dsh-fal-image-gen
最近更新
2026年8月20日
分类
工具与能力
GitHub stars
0
载体类型
plugin
目录证据
上游声明已找到 dsh.bundle
证据路径
package.json#dsh.bundle
核对版本
0.1.0-rc.8
上游核对日期
2026-08-20

该证据由上游目录提供。本站没有安装、运行或安全审核这个插件。

安装

默认先复制一段 Prompt,让 Agent 读 GitHub 仓库和源码;需要自己装时再切到命令。

复制这段 Prompt,发给 DSH、Codex 或其他 Agent,让它先读 GitHub 仓库和源码。

请先不要安装或执行任何命令。阅读这个插件的 GitHub 仓库、README 和关键源码,然后用清楚、直接的方式回答以下问题,帮助我判断它是否适合我的需求:

1. 这个插件是什么,解决什么问题;
2. 适合哪些用户和典型使用场景;
3. 安装后如何使用,并给出一个最小使用示例;
4. 有哪些已知限制,以及隐私、安全、兼容性或维护风险;
5. 给出“推荐 / 有条件推荐 / 不推荐”的明确建议和理由。

请区分仓库明确说明、根据源码推断和未知信息。证据不足时请明确说明,不要猜测或照抄 README。

GitHub:https://github.com/GooDAnDReaDY/dsh-fal-image-gen
插件名:dsh-fal-image-gen
作者:GooDAnDReaDY

检查来源文件

安装前先看这个插件目录里的 README 和其他文件。

文件资源管理器3 个文件
README.md来源说明 · 只读预览

dsh-fal-image-gen

Image generation for DeepSeek Harness (dsh): a generate_image tool with pluggable providers — the FAL REST queue API (default model fal-ai/flux-2/klein/9b) or any OpenAI-compatible images API.

A generated image is:

  • shown inline in the conversation, in the plugin's own tool card;
  • saved into the session working directory (<session cwd>/generated/images/*.png);
  • returned to the model with its path, size and seed.

Providers

providerWhat it isProtocol
fal (default)the FAL queuesubmit, poll, download
customany OpenAI-compatible APIPOST {customBaseURL}/images/generations, one request

The named sizes stay the same whichever provider runs — they are the tool's language. FAL takes them as they are; for an OpenAI-compatible API they are translated (square_hd1024x1024, landscape_4_31024x768, and so on). An API picky about sizes gets customSize, which is sent verbatim instead.

response_format is deliberately not sent: newer OpenAI models reject it, and the answer is accepted either way — base64 inline, or a link that gets downloaded.

Example: OpenAI

- id: dsh-fal-image-gen
  config:
    provider: custom
    customBaseURL: https://api.openai.com/v1
    customModel: gpt-image-1
    customKeyEnv: OPENAI_API_KEY

An empty customKeyEnv means no authorization header at all, for a local gateway that needs none.

Two delivery modes

The picture appears in the conversation either way. What differs is what the chat model receives — and that decides whether the turn survives a text-only model.

link (default)image
The model receivestext and a linkthe image itself
Shown in the chatyes, the card renders it from the linkyes
Works with a text-only chat modelyes, on its ownno — needs dsh-vision-bridge or a vision-capable chat model
The model can reason about the pictureno, only about the prompt and the linkyes
The link points tothis plugin's own route, as durable as the attachmentfal.media, which expires

Pick image when the conversation should be able to discuss what was drawn — "make the cat bluer" needs a model that can actually see it. Without a vision model in the chat that mode fails the turn with does not support image input, which is precisely what dsh-vision-bridge exists to prevent: it swaps the picture for a description from a vision model you choose.

Set it in Settings → Plugins → FAL Image Generation → How the image reaches the chat, or in the profile:

- id: dsh-fal-image-gen
  config:
    deliverAs: image

Install

# From npm after publishing:
dsh plugin --profile web add @goodandready/dsh-fal-image-gen

# From GitHub:
dsh plugin --profile web add github:GooDAnDReaDY/dsh-fal-image-gen

# Locally from a checkout:
dsh plugin --profile web add /path/to/dsh-fal-image-gen

Restart the Web UI afterwards.

Configure (Web GUI)

All settings live in Settings → Plugins → Plugin configuration → FAL Image Generation:

FieldDefaultDescription
providerfalWhich provider generates the image: fal or custom.
modelfal-ai/flux-2/klein/9bFAL model id, called as {baseURL}/{model}. Used when provider is fal.
apiKeyEnvFAL_API_KEYAPI key reference (credentials / env var).
baseURLhttps://queue.fal.runFAL queue base URL.
defaultSizelandscape_4_3Default image size.
defaultFormatpngDefault output format.
pollIntervalMs2000Job status poll interval.
timeoutMs180000Total generation timeout.
deliverAslinklink — the result is text with a link, works with any chat model. image — the result carries the picture, needs dsh-vision-bridge or a vision-capable model.
customBaseURLprovider=custom: API root, e.g. https://api.openai.com/v1.
customModelprovider=custom: model id, e.g. gpt-image-1.
customKeyEnvOPENAI_API_KEYprovider=custom: key reference. Empty means no authorization header.
customSizeprovider=custom: fixed size sent verbatim. Empty means the named size is translated.
outputDirgenerated/imagesOutput folder. A relative path resolves against the session working directory; an absolute path is used as given.

Equivalent values can be set in $DSH_HOME/settings.yaml under dsh-fal-image-gen: — the GUI writes to the same settings document, so both ways are equivalent.

API key

Store your key in Credentials (Web: Settings → Credentials, name FAL_API_KEY) or in $DSH_HOME/.credentials.yaml:

FAL_API_KEY: <your key from https://fal.ai/dashboard/keys>

The plugin prepends the Key auth prefix automatically.

Usage

Just ask the model to draw an image:

> Generate an image: neon cyberpunk city at night in the rain, 16:9

Tool parameters (all except prompt are optional):

ParameterDescription
promptrequired, detailed image description
image_sizesquare_hd / square / portrait_4_3 / portrait_16_9 / landscape_4_3 / landscape_16_9
seedseed for reproducibility
output_formatpng (default) / jpeg / webp
output_namefile name without extension

Structure

dsh-fal-image-gen/
├── package.json            # dsh bundle/plugin metadata + peerDependencies
├── cordis.patch.yml        # bundle layer: inserts the plugin row
├── lib/index.js            # host: generate_image tool, attachment and file handling
├── lib/providers.js        # host: the providers — FAL queue, OpenAI-compatible API
├── lib/client.js           # browser: settings card + the generate_image tool card
├── test/                   # unit tests for the providers, on a fake fetch
├── README.md
└── LICENSE                 # MIT

Why the plugin ships its own tool card

Tool cards in dsh do not render image blocks — only user and assistant messages do — so a picture returned by a tool would otherwise show up as JSON. The plugin registers a keyed tool.call.toolview entry for generate_image and serves the stored bytes from its own route (GET /dsh-fal-image-gen/image), which is what puts the image in the conversation.

No npm runtime dependencies (the @deepseek-ai/* peer deps resolve from the dsh install), no build step — plain ESM.

License

MIT