Dreamina Design
Plugin marketplaces
This plugin belongs to AIGC content creation.
| Category | Marketplace | Purpose |
|---|
| Full-stack development | Full Stack Plugins | Architecture and UI design, code understanding, quality checks, code review, workflow governance, and server operations |
| AIGC content creation | Full AIGC Plugins | Image, video, audio, music, 3D, and multimodal content creation |
Dreamina Design — create images and videos with control
Create and edit Dreamina images and videos from your coding agent, with a capability snapshot from the live CLI, one explicit approval per paid call, and submissions you can resume by identifier.
English | 简体中文 · Install · Quick start · Skills · MCP tools · Troubleshooting
Positioning
dreamina-design is a strict wrapper around the official Dreamina CLI. It discovers what the installed CLI can actually do, validates each generation request against that snapshot, asks for a native confirmation before any paid call, and queries an uncertain task by its submit_id instead of submitting again. No tool accepts free-form shell input.
Since 0.7.0 the plugin runs a dual rail: paid submit tools default to the Canvas runtime (dreamina-canvas CLI with quote-bound approval), while runtime: "legacy" keeps the frozen dreamina path. The legacy binary stops being maintained after 2026-11, so new work should use the Canvas rail.
Installation
Prerequisites
- Python 3.13 for the MCP server.
- The
dreamina-canvas CLI, enrolled and trusted with an absolute path and SHA-256 digest. The legacy dreamina CLI is optional and only for the frozen rail.
- Optional, reference-video project tools only:
ffmpeg, ffprobe, a Whisper-compatible ASR runtime, and macOS say.
- A Dreamina account with credits; generation spend is never included in the plugin.
From the plugin marketplace
codex plugin marketplace add partme-ai/full-aigc-plugins
codex plugin add dreamina-design@full-aigc-plugins
Restart the host and open a new task. Then confirm the server and the CLI trust record:
codex mcp list
Read-only tools run without a prompt; paid and mutating tools raise a native confirmation whose default action is Cancel.
Quick start
flowchart LR
A[Intent] --> B[Capability snapshot]
B --> C[Validated request]
C --> D{Paid?}
D -->|no| E[Read-only tool]
D -->|yes| F[Native confirmation]
F -->|approved| G[Submit once with submit_id]
G --> H[Query by submit_id]
H --> I[Verified download plus receipt]
F -->|cancelled| J[Stop, nothing spent]
Create a Dreamina image from this brief: a matte ceramic teapot on a linen table,
soft morning light. Show the exact parameters before you spend anything.
Expected observation: a capability snapshot, a validated request, a confirmation dialog listing the exact parameters, then a single submission with a submit_id. Asking to query it downloads the artifact and verifies it on arrival.
What you can build
| Workflow | Built-in path | Verifiable output |
|---|
| Image generation and edits | dreamina-prompt-text2image, dreamina-prompt-image2image, dreamina-submit_image | Submitted task plus verified image |
| Video generation | dreamina-prompt-text2video, dreamina-prompt-image2video, dreamina-submit_video | Submitted task plus verified video |
| Iterate toward a target | dreamina-visual-target, dreamina-vision-judge, dreamina-visual-loop | Target receipt, round receipt, judge verdict |
| Full video project | dreamina-video-production, dreamina-video-shots, dreamina-video-evaluator, dreamina-video-sync | Shot analysis, evaluated batches, composed export |
| Browser-only members | dreamina-opencli-text2image, dreamina-opencli-image2image, dreamina-opencli-text2video, dreamina-opencli-image2video | Generated artifacts through the Jimeng web session |
Skills
| Skill | What it does |
|---|
dreamina-design-use | Routes an unspecified image or video request to the right entry point. |
dreamina-design-harness | Invocation spec: generation families, shot annotation, evaluation and paid submit-once gates. |
dreamina-auto-seedance | Turns a validated DCC preview into an approved Seedance video. |
dreamina-seedance-resume | Resumes a paused 3D job from its ledger state without re-paying. |
dreamina-prompt-text2image | Writes and refines image-generation prompts. |
dreamina-prompt-image2image | Writes edit prompts for style, background, detail, colour and compositing. |
dreamina-prompt-text2video | Writes video prompts with motion and camera direction. |
dreamina-prompt-image2video | Covers single-image, first-last frame, storyboard and multimodal reference modes. |
dreamina-vision-judge | Scores a candidate against a target on composition, lighting, materials and detail, out of ten. |
dreamina-visual-target | Obtains or synthesizes the target image a visual loop needs. |
dreamina-video-production | Runs a complete video project from a local reference video. |
dreamina-video-shots | Breaks a finished clip into a measured shot table with deterministic gates. |
dreamina-video-sync | Composites shot metadata onto the source video so it cuts with the picture. |
dreamina-shot-annotator | Produces shot-annotation schema content only. |
dreamina-video-evaluator | Produces shot-evaluation schema content only. |
dreamina-opencli-text2image | Automates Jimeng text-to-image for standard members through the browser bridge. |
dreamina-opencli-image2image | Browser-bridge image editing when no CLI is available. |
dreamina-opencli-text2video | Browser-bridge text-to-video for standard members. |
|
Slash commands
| Command | What it does |
|---|
/dreamina | Entry point; routes text/image and video families. |
/dreamina-setup | First-run setup: CLI install, authentication and preflight. |
/dreamina-text2image | Submit, poll and troubleshoot image generation. |
/dreamina-image2image | Local edits, stylization and upscale. |
/dreamina-text2video | Text-to-video including Seedance. |
/dreamina-image2video | Reference image, video or audio driven generation. |
/dreamina-prompt | Prompt refinement across the four modes. |
/dreamina-visual-loop | Target lock → paid generation → independent review → bounded retry. |
/dreamina-auto-seedance | Validated preview → one submit → verified download. |
/dreamina-seedance-resume | Resume a paused Seedance job from the ledger. |
/dreamina-video-production | Reference video → analysis → prompts → submit → evaluation. |
MCP tools
.mcp.json declares one stdio server, dreamina_design, started with python3 -m scripts.dreamina_mcp_server. It exposes 22 typed tools; read-only ones are pre-approved and paid ones always prompt.
| Group | Tools |
|---|
| Read-only | dreamina_capability_snapshot, dreamina_cli_status, dreamina_account, dreamina_query_task, dreamina_list_tasks, dreamina_diagnose, dreamina_quote_video_batch |
| Paid or mutating | dreamina_submit_image, dreamina_submit_video, dreamina_visual_loop, dreamina_cli_install_or_upgrade, dreamina_auth, dreamina_session |
| Video project | dreamina_video_project, dreamina_analyze_reference_video, dreamina_validate_shot_analysis, dreamina_create_redesign, dreamina_evaluate_video_batch |
| Batch and export | dreamina_approve_video_batch, dreamina_execute_video_batch, dreamina_compose_video, dreamina_export_video_project |
Failures return a structured envelope with error_type, message, retryable, requires_user_action and a next_action of request_user_action, query_same_submit_id or correct_request.
Configuration
| Setting | Location | Notes |
|---|
| MCP server | .mcp.json | stdio; 10-second startup timeout, 3600-second tool timeout |
| Trusted CLI record | ~/.config/dreamina-design/trusted-cli.json | Mode 0600; stores path and digest, never a credential |
| State root | ~/.local/share/dreamina-design/ | Holds operations/ and approvals/ |
| CLI logs | ~/.dreamina_cli/logs | Read only through the bounded, redacted diagnostic tool |
Architecture and security
Codex owns orchestration and approvals; the Dreamina CLI owns authentication and the remote API; this plugin owns validation, approval receipts and the operation ledger. The authors receive no artifact.
- A paid call requires a native confirmation bound to the session and request; receipts are single-use and expire after five minutes.
- CLI trust requires an absolute non-symlink path, a trusted owner and mode, and a SHA-256 digest.
- Approval persistence strips credential-like keys before writing; diagnostics are bounded and redacted.
- Reference-video uploads are contained: at most 50 MiB per image and 512 MiB per media file.
| Evidence | Status |
|---|
| CLI runtime contract, existing-task recovery | Recorded in docs/verification/dreamina-cli-runtime.md |
| Paid canary | Separately approved and recorded in docs/verification/paid-canary-2026-09-13.md |
| Reference-video project chain | Registered, but the ten runtime gate lines are NOT_RUN |
Troubleshooting
| Symptom | Check first | Resolution |
|---|
| Tools are missing | codex mcp list | Enable the plugin and open a new task |
| A paid call is blocked | The confirmation dialog | Confirm explicitly; the default is Cancel |
| The CLI is not trusted | The trust record | Re-enrol the CLI and confirm the digest |
| A task looks stuck | The task state | Query by submit_id; never submit again |
| A video mode is refused | Web-side prerequisites | Complete the prerequisite on the web; the plugin will not bypass it |
| A reference-video tool does nothing | The runtime gate record | Those gate lines are NOT_RUN until they pass |
Upgrade
codex plugin marketplace upgrade full-aigc-plugins
codex plugin add dreamina-design@full-aigc-plugins
Re-run the capability snapshot after upgrading: model parameters and command help come from the installed CLI, not from this repository. If you still call runtime: "legacy", plan the move to the Canvas rail before the 2026-11 sunset.
Current packaged version: 0.8.1+codex.20261018 (development preview).
Contributing and support
Open functional issues at https://github.com/full-aigc-plugins/dreamina-design-plugin/issues. State the CLI version you verified against, whether the change alters the approval envelope or the receipt format, and include the affected gate output.
License
Apache-2.0 — see LICENSE, NOTICE and THIRD_PARTY_NOTICES.md.
