Troubleshooting
Using Midjourney with Your Agent (Without Getting Stuck)
A practical workflow for when your agent can plan Midjourney prompts but cannot directly run the full Discord image flow.
A common frustration is: "My agent says it can do Midjourney, but then it stalls or asks me to do things manually." In most setups, this is normal. Your agent is great at ideation and prompt iteration, but final image generation usually still happens in your Midjourney Discord flow.
Mental model: use your agent as the creative director + prompt optimizer. Use Midjourney itself for generation and final upscales/variations.
What your agent should do vs what you still do
- Agent does: brand/style discovery, prompt drafting, prompt versioning, negative constraints, iteration planning.
- You do: run
/imagine, choose variations/upscales, and post final assets where needed.
- If it claims full automation: ask for capability proof before you run long prompt loops.
2-minute canary test (before big workflows)
- Ask your agent for 3 Midjourney prompts for the same concept (realistic, cinematic, flat-vector).
- Run one prompt in Midjourney Discord manually.
- Paste result notes (what looked wrong/right) back to the agent.
- Ask for a revised prompt based on that feedback.
Loop-break rule: if your agent keeps repeating "I can publish directly" but gives no proof, stop and switch to manual-generation workflow immediately.
Known-good prompt format
Act as my Midjourney prompt engineer.
Goal:
[what image I need]
Brand/style constraints:
[colors, mood, references]
Output format:
1) Give 4 distinct prompt options
2) Each prompt must include aspect ratio + style controls
3) Add 1 short "when to use this version" note per prompt
4) End with a "next-iteration plan" after I send results
High-friction mistakes to avoid
- Asking for 20+ prompts before testing one concept.
- Skipping style constraints (brand colors, lighting mood, composition).
- Not saving winning prompts with version labels (v1, v2, v3).
- Treating Midjourney generation limits/errors as "agent failure."
Advanced edge cases (when basic flow still fails)
- Wrong Discord account: you generated with one account but tested prompts in a different account/server context.
- Wrong destination context: prompt strategy happens in one chat, but generation/testing happens in another thread and feedback never makes it back.
- Command syntax drift: edited prompt text removes critical Midjourney flags or parameter order.
- Plan/queue confusion: generation throttling/limits can look like agent failure. Confirm queue/plan status first, then continue iteration.
- Retry spiral: re-running broad prompts without feeding concrete image feedback causes low-quality loops and token waste.
Known-good recovery prompt (copy/paste)
You are my Midjourney workflow copilot.
Rules:
- Do NOT claim direct publishing/generation unless you can prove it in this exact context.
- Ask me for exactly one missing input block, then generate prompts.
- Output max 3 prompts, each with: prompt text, parameters, and one-line intent.
- After I share results, give only targeted revisions (not full rewrites).
Before prompts, return this checklist:
1) Automation boundary (what you can/can't execute here)
2) One canary prompt for fast test
3) Success criteria for next revision
Then stop and wait for my test result.
When to escalate to #help
- Your agent used to guide iterations correctly, then stopped producing usable prompt revisions.
- You see persistent platform/tool errors unrelated to prompt quality.
- You need a capability check for a specific connector workflow beyond standard Midjourney-in-Discord usage.
Support-ready evidence packet
Midjourney workflow issue report
Goal:
Prompt used:
Where it failed (agent planning vs generation step):
Expected behavior:
Actual behavior:
One successful/failed sample output:
What changed recently (model/tool/surface):
Best practice: run a quick canary, lock one winning style, then scale output. This keeps token usage down and avoids endless prompt churn.