LeemanCheung/dsh-image-gen

Generate images in DeepSeek Harness with OpenAI gpt-image-2, progressive live previews, and a polished Codex-inspired developing animation.

中文说明

The illustration mirrors the shipped card states; live draft frames come from the provider stream.

Highlights

• Registers the Codex-compatible model tool name image_gen.
• Streams up to three real provider partial images instead of showing a fake progress bar.
• Cross-fades each partial over one animated developing plate, then sharpens into the final image.
• Stores the final image in DSH's immutable attachment store and supports replay, lightbox preview, and download.
• Keeps model-facing tool output text-only, so image generation does not make a text-only model route reject conversation history.
• Supports native tool calls and nested Code Mode calls with the same durable card.
• Resolves credentials per operation, rejects redirects, limits response sizes, validates image bytes through DSH, and retries only transient failures.
• Includes Chinese and English UI copy plus prefers-reduced-motion support.

Codex parity and improvements

OpenAI Codex's built-in image_gen tool hardcodes gpt-image-2, records one working state, and saves the completed image under Codex's generated-images directory. Its public source verifies the states in_progress, completed, and failed; it does not expose a distinctive diffusion animation.

dsh-image-gen keeps the compatible image_gen name and model while improving the visible process:

Experience	Codex	dsh-image-gen
GPT Image 2	Yes	Yes
Progressive provider frames	Tool supports them, default UI is generic	Up to 3 live frames in the card
Before first frame	Generic working state	Animated developing plate, scan, and light field
Frame transition	Generic activity	In-place cross-fade and focus development
Final result	Saved image	Durable DSH attachment, replay, lightbox, download
Text-only model compatibility	Not applicable to DSH history	Model receives text; image reference stays in UI metadata
Reduced motion	Platform dependent	Explicitly supported

Primary references:

• Codex image-generation tool
• OpenAI image-generation skill
• GPT Image 2 model
• Image generation guide

Compatibility

Verified against:

• DeepSeek Harness 0.1.0-rc.6
• Node.js 24.15.0
• DSH Web profile on Windows 11

Verification date: 2026-08-15.

Install

Review third-party source before installation and pin a release tag or commit:

dsh plugin --profile web add github:LeemanCheung/dsh-image-gen#v0.1.0

For local development:

git clone https://github.com/LeemanCheung/dsh-image-gen.git
cd dsh-image-gen
npm install
npm run check
dsh plugin --profile web add .

The repository commits lib/index.js and lib/client.js, so a pinned Git install does not need to run a dependency build script. Restart the DSH Host after installation and refresh the Web page.

Credential

The default credential reference is OPENAI_API_KEY. Provide it through DSH's credential store or the Host environment. Do not place a raw secret in cordis.patch.yml, chat messages, Git, or screenshots.

PowerShell example for one Host process:

$env:OPENAI_API_KEY = 'sk-...'
dsh --profile web

The plugin resolves the credential for each generation. Rotating the credential does not require changing plugin configuration.

Use

Ask naturally in a DSH conversation, for example:

Generate a cinematic 16:9 product photograph of a translucent mechanical keyboard on a dark glass desk, violet rim light, no text.

The model calls image_gen. While it runs, the card shows a developing animation and replaces the preview with each real streamed partial. DSH's existing interrupt control cancels the request. The settled card supports preview and download.

Tool options:

• prompt: detailed generation instructions, 1–32,000 characters and at most 64,000 UTF-8 bytes.
• size: auto or arbitrary WIDTHxHEIGHT accepted by GPT Image 2: each edge divisible by 16, no edge above 3840, aspect ratio 1:3–3:1, and 655,360–8,294,400 total pixels.
• quality: auto, low, medium, or high.
• output_format: png, jpeg, or webp.
• output_compression: 0–100 for JPEG/WebP only.
• background: auto or opaque. GPT Image 2 does not support transparent output.

Configure

The bundle inserts the image-gen row with safe defaults. Override it in the selected profile's cordis.patch.yml:

- id: image-gen
  name: dsh-image-gen
  config:
    apiKeyEnv: OPENAI_API_KEY
    baseUrl: https://api.openai.com/v1
    model: gpt-image-2
    defaultSize: auto
    defaultQuality: auto
    defaultOutputFormat: png
    defaultOutputCompression: 90
    defaultBackground: auto
    moderation: auto
    partialImages: 3
    requestTimeoutMs: 120000
    maxRetries: 2
    retryBaseMs: 1000
    maxConcurrent: 2

Configuration fails at load for an invalid provider URL or default image size. Plain HTTP is accepted only for localhost, 127.0.0.1, or [::1] development endpoints. Credential-bearing requests use redirect: "error".

Cost note

OpenAI bills image generation by the selected image quality and size. Each progressive partial image costs an additional 100 image-output tokens according to the provider guide. Set partialImages: 0 to minimize partial-preview cost, or 1–3 to trade cost for a richer live experience.

Data, network, and permissions

• Network: sends the image prompt and selected options to baseUrl; the default is OpenAI's Images API.
• Credentials: reads only the configured DSH credential reference per request. The resolved secret is not stored in plugin state, logs, metadata, or session history.
• Storage: stores only completed images through the DSH attachment service. Partial frames stay in bounded Host memory while the call is active and are then discarded.
• Browser access: uses a loopback-only private RPC. A final image is returned only after the Host finds the exact attachment reference in the requested session and call record.
• Workspace files: does not read or write the session workspace.
• User data: prompts and tool arguments follow DSH's normal session logging. OpenAI receives the prompt under the terms governing the configured API account.

Troubleshooting

No credential is configured for OPENAI_API_KEY

Configure the credential for the DSH Host process, then retry. Never send the key in chat.

Image generation was blocked

Revise the prompt. The plugin does not retry provider moderation or user-input errors.

Preview unavailable after success

Refresh the page. If the card still cannot load, inspect Host logs and verify that the profile still mounts dsh-image-gen and its attachment store is available.

dsh plugin --profile web why dsh-image-gen
dsh --profile web --dump-config

Requests time out

Increase requestTimeoutMs within its 10–300 second range, select a lower quality, or use JPEG. DSH interruption still aborts the upstream request.

Roll back or remove

Before changing a production profile, take a DSH config snapshot when the undo plugin is installed. To uninstall:

dsh plugin --profile web remove dsh-image-gen

Restart the DSH Host and refresh the page. Existing image records remain in session history; their custom card requires the plugin to be installed.

Development

npm install
npm run typecheck
npm test
npm run build
npm pack --dry-run

Tests use deterministic mocked SSE streams and a local redirect server. A real OpenAI smoke test is intentionally not part of the keyless suite. Real provider behavior was not exercised in this checkout because no OPENAI_API_KEY was available.

The build emits:

• lib/index.js: Host Cordis plugin.
• lib/client.js: browser module-loader bundle.
• lib/client.js.map: browser source map.

Known limitations

• Version 0.1 generates new images. Codex-style reference-image editing is not yet exposed because a safe DSH attachment selector and explicit external-upload consent are needed.
• Final previews are intentionally loopback-only. Remote Web clients receive a clear unavailable state rather than image bytes.
• Current DSH credential resolution and attachment saving do not accept cancellation signals. The plugin checks cancellation before and after those stages and waits for them during teardown, but cannot interrupt a provider implementation that stalls inside either service.
• OpenAI may evolve arbitrary-size limits or event fields. The plugin fails closed on incompatible responses instead of guessing.

Security

See SECURITY.md for private reporting. Do not include API keys, private prompts, or generated private images in a public issue.

License

MIT