Workflow: Iterative game-asset concepts with Midjourney

Recommendation (short):
Use Midjourney as your rapid ideation engine but lock down a small set of consistent generation parameters and a naming/folder scheme. Generate a single “base silhouette” using a fixed seed, then produce angle-specific frames (front/3/4/side/back) by reusing that seed + minimal prompt edits. Keep stylize and chaos low to preserve shape consistency, and export upscaled PNGs into a consistent folder structure for handoff.

Why this works
Midjourney is fast and artistically flexible, but it’s nondeterministic by default. Using the same seed + low-stylize/low-chaos + identical camera/lighting language gives repeatable character topology across variants. That yields concept art you can convert into turnarounds or hand off to modelers/texture artists.

Decision criteria (when to use Midjourney for your pipeline)
- Use it if: you need many quick stylized NPC concepts, you have a small art budget, and you can accept 2D-to-3D cleanup by artists.
- Avoid it if: you require final-production meshes/UVs, exact proportions every frame, or strict photogrammetry-style realism.
- Depends on team/budget: subscription cost and time to craft prompts matter. Lone indie can use it to iterate; larger teams may use it for moodboards rather than final assets.

Practical Midjourney settings and seed workflow (step-by-step)
1) Create a clean base silhouette: prompt for a neutral pose (T-pose or A-pose), “neutral gray background”, “orthographic perspective”, “neutral 3-point lighting” and set: --ar 2:3 (tall turnaround), --stylize 25 (or lower), --chaos 5, --quality 2. Generate until you find a silhouette you like.
2) Grab the seed from the job details (shown by Midjourney). Note it.
3) For each view (front, 3/4, side, back): reuse the exact base prompt + change only the view token (“front view”, “left profile”, “rear view”, etc.) and append --seed . Keep stylize/chaos identical. If you want slight variations, use Remix mode or rerun with small prompt changes.
4) Upscale with a non-destructive upscaler (–uplight if available) to retain line/detail.
5) Remove background in an editor if needed (or use transparent options if your plan supports them).

Prompt template (example to copy):
CharacterName, [role], stylized, clean silhouette, T-pose, orthographic perspective, neutral gray background, neutral 3-point lighting, flat camera distance, front view, limited color palette [describe], clean linework --ar 2:3 --stylize 25 --chaos 5 --quality 2 --seed 12345678

Naming conventions & folder structure (practical)
- File name format: ___v__seed_mj.jpg
Example: rogue01_thief_v01_front_seed12345678_mj.jpg
- Folders: concepts/raw/, concepts/approved/, turnarounds/raw/, turnarounds/cleaned/, handoff/ (contains PSDs, reference sheets, and modeler notes).

Checklist for an iteration cycle
1. Write/lock base prompt and parameters. 2. Generate base silhouette; capture seed. 3. Generate each view reusing the seed. 4. Upscale and export PNG. 5. Name files per convention and move to turnarounds/raw. 6. Quick QA: alignment, silhouette match, color palette. 7. Tweak prompt or seed if needed; increment version.

Best-for / Avoid-if
- Best-for: fast stylized NPC exploration, cohesive concept sets, solo or small teams iterating art direction.
- Avoid-if: you need production-ready topology, exact animation-ready turnarounds without manual cleanup.

Optional: use ChatGPT to produce and maintain prompt templates and feedback iterations (it helps automate prompt variants and QA notes). CTA: try Midjourney for the generation steps above.

Quality note: reproducibility depends on Midjourney version and subscription features. Expect a learning curve to tune stylize/seed for your specific style.

Compare Midjourney and Leonardo AI