Unofficial Printing Press CLI for Midjourney's web app API.
This CLI was generated from browser-observed Midjourney traffic and then patched with a small hand-written layer for creation and export workflows that are not represented well by read-only endpoint mirroring.
Important scope note: this is not an official Midjourney API client. It uses the same authenticated browser session surfaces that the web app uses, so endpoint shapes can change without notice.
What Works
Read-only inspection commands:
midjourney-pp-cli explore listmidjourney-pp-cli explore style-likesmidjourney-pp-cli foldersmidjourney-pp-cli generations listmidjourney-pp-cli generations updatesmidjourney-pp-cli moodboardsmidjourney-pp-cli profiles followingmidjourney-pp-cli profiles personalizedmidjourney-pp-cli queuemidjourney-pp-cli rankings contests-countmidjourney-pp-cli rankings model-ratingsmidjourney-pp-cli storage
Creation/export commands added after traffic capture:
midjourney-pp-cli imagine "<prompt>"midjourney-pp-cli rerun <job-id>midjourney-pp-cli download <job-id> --index 0 --out image.png
The mutating commands call Midjourney. Use --dry-run first when testing payloads.
Authentication
Midjourney does not expose a normal public API key for this workflow. The CLI uses an authenticated browser session.
There are two paths:
- Browser-backed commands use Chrome DevTools Protocol against a browser that is already logged into
https://www.midjourney.com.
- Read-only HTTP commands can use
MIDJOURNEY_COOKIE_HEADER when you intentionally provide a cookie header for the logged-in session.
For browser-backed creation/export, keep a Chrome instance with remote debugging enabled and logged into Midjourney, then pass its CDP URL if it is not the default:
midjourney-pp-cli imagine "simple red cube on a white desk" \
--browser \
--browser-cdp http://127.0.0.1:18800 \
--dry-run
For cookie-backed read-only commands:
export MIDJOURNEY_COOKIE_HEADER="<cookie header from your own logged-in browser session>"
midjourney-pp-cli doctor
Do not commit, paste, or print real cookie headers. Treat MIDJOURNEY_COOKIE_HEADER as sensitive credential material.
Quick Start
Verify the CLI:
midjourney-pp-cli doctor
midjourney-pp-cli agent-context --pretty
List recent generations:
midjourney-pp-cli generations list --json
Submit an image job:
midjourney-pp-cli imagine "small red cube on a white studio desk" \
--ar 1:1 \
--version 7 \
--json \
--yes
Preview the same request without sending it:
midjourney-pp-cli imagine "small red cube on a white studio desk" \
--ar 1:1 \
--version 7 \
--dry-run \
--json \
--yes
Export all four images from a completed Midjourney job:
for i in 0 1 2 3; do
midjourney-pp-cli download <job-id> \
--index "$i" \
--out "midjourney-$i.png" \
--json \
--yes
done
Imagine Options
The imagine command builds the same kind of prompt string observed in the Midjourney Create UI, then submits it through POST /api/submit-jobs.
Common flags:
--ar 4:3 appends --ar 4:3--version 7 appends --v 7--niji 6 appends --niji 6 and suppresses --v--style raw appends --style raw--raw is shorthand for --style raw--style cute appends --style cute--quality 2 appends --q 2--stylize 250 appends --stylize 250--chaos 30 appends --chaos 30--weird 25 appends --weird 25--seed 123 appends --seed 123--tile appends --tile--draft appends --draft--sref <url-or-code> appends a style reference--oref <url-or-code> appends an omni reference--image-prompt <url> prepends an image prompt URL--profile-id <profile> appends --profile <profile>--speed fast|relax|turbo sets the request f.mode
Example with style and omni references:
midjourney-pp-cli imagine "pleasant illustrated person organizing notes on a desk" \
--ar 4:3 \
--version 7 \
--oref <omni-reference-url> \
--sref <style-reference-url> \
--profile-id <profile-id> \
--json \
--yes
Example Niji request:
midjourney-pp-cli imagine "tiny friendly robot arranging sticky notes" \
--ar 4:5 \
--style cute \
--niji 6 \
--json \
--yes
Rerun
Rerun submits the observed Midjourney reroll payload:
midjourney-pp-cli rerun <job-id> --json --yes
Dry-run first when integrating:
midjourney-pp-cli rerun <job-id> --dry-run --json --yes
Download
Direct server-side fetches from Midjourney CDN can fail with Cloudflare protection, and browser-side fetch() can be blocked by CORS. The download command avoids that by opening the rendered Midjourney job page through Chrome CDP, finding the rendered image, and saving a cropped PNG.
midjourney-pp-cli download <job-id> --index 0 --out image.png --json --yes
Indexes are 0, 1, 2, and 3 for the standard four-image Midjourney batch.
Agent Usage
This CLI is designed for AI agent consumption:
- Use
--agent for JSON, compact output, no color, no prompts, and confirmation-safe defaults.
- Use
--dry-run before mutating commands.
- Use
which "<capability>" --json to discover commands at runtime.
- Use
agent-context --pretty for the current command surface and profiles.
- Read-only endpoint mirror commands are safe inspection tools.
imagine and rerun mutate remote Midjourney state and should be treated as external actions.download writes a local file and requires an output path.
Exit codes: 0 success, 2 usage error, 3 not found, 4 auth error, 5 API error, 7 rate limited, 10 config error.
Verification
Development checks:
go test ./...
go vet ./...
go build -o ./midjourney-pp-cli ./cmd/midjourney-pp-cli
./midjourney-pp-cli --help
./midjourney-pp-cli doctor --json
For vulnerability scanning when govulncheck is installed:
govulncheck ./...
Printing Press verify mode is respected by mutating commands. When PRINTING_PRESS_VERIFY=1 is set without PRINTING_PRESS_VERIFY_LIVE_HTTP=1, imagine and rerun return the standard synthetic verify noop instead of calling Midjourney.
Discovery Notes
Captured behavior:
- The UI posts
POST /api/prompt-session-log before POST /api/submit-jobs.
POST /api/submit-jobs is the mutating endpoint for tested image-generation variants.- Model and option selection is primarily encoded as prompt suffixes.
--v 7 produced v7_diffusion.--v 7 --style raw produced v7_raw_diffusion.--v 6.1 --style raw produced v6-1_raw_diffusion.--v 6 produced v6_diffusion.--niji 6 produced v6_diffusion_anime.--v 7 --draft produced v7_draft_diffusion.- Draft and tile are incompatible; Midjourney returns
invalid_parameter.
--turbo and --relax prompt suffixes were accepted in UI capture, with response flags resolving to turbo and relaxed.
The capture summaries are intentionally not committed because browser captures can contain account-specific identifiers or session-derived data. Keep raw capture files local and commit only redacted conclusions.
Contributing Back To Printing Press
This is a printed CLI, not a generator change. Per the Printing Press repository guidance:
- Use the supported public-library publishing flow for generated CLIs.
- Run the
/printing-press-publish skill instead of manually copying files into mvanhorn/printing-press-library.
- Do not hand-edit the public-library registry, generated README catalog cells, or mirrored
cli-skills output.
- Include provenance in the PR: source type
sniffed, auth requirement logged-in browser session / cookie header, live smoke evidence, and explicit out-of-scope notes.
- Keep the community PR template intact.
- Use the AI/automation disclosure honestly; for this work,
Human-reviewed is the right category once a human reviews the diff before submission.
- List verification commands actually run.
- For generator-level improvements discovered here, open an issue in
mvanhorn/cli-printing-press instead of hiding them in this printed CLI.
Generated by CLI Printing Press.