Recommendation (short):
If you need repeatable, photoreal actor portraits with tight control over identity across many shots, start with Leonardo for identity-consistency (trained references/persona) and use Midjourney when you need richer artistic photoreal styling. For most casting boards where exact likeness, skin tones and uniform lighting matter more than stylization, Leonardo will usually be the faster path; Midjourney is great when you want a specific photographic aesthetic and are willing to do more prompt/seed management.

Decision criteria (how to pick):
- Consistency of identity: Leonardo (has easier workflows for training/reusing a “model” from many reference photos). Midjourney can match a person with repeated image prompts/seeds but requires more manual tuning.
- Skin tones & lighting accuracy: Both can produce excellent results with correct prompts; Leonardo tends toward cleaner, studio-like renders out of the box; Midjourney can be richer but occasionally hallucinates details without careful prompting.
- Cost & throughput: Leonardo is credit/consumption-based (good for per-image control). Midjourney is subscription-based (good for heavy iterative work). Expect roughly $0.05–$0.50 effective cost per usable portrait depending on subscription/credit plan and how many trials you discard—test both with a small paid run.
- Upscaling & final quality: Both provide built-in upscalers (2x/4x and specialized “upscale” modes). For final casting boards, plan to run the winner through a dedicated upscaler (Topaz/Gigapixel) and minor retouch in Photoshop/Lightroom for perfect skin tone/eyes.
- Workflow & team: Solo artists with prompt skills may prefer Midjourney; teams needing reproducible identity across many images will prefer Leonardo’s training features.

Practical checklist (step-by-step):
1. Collect 8–20 high-quality reference photos (consistent lighting range, clean backgrounds, different angles). Obtain written consent for likeness use.
2. Decide path: quick-reference prompting vs. train a persona/model. Choose Leonardo for persona training; Midjourney if you want rapid stylistic experiments.
3. Build a strict prompt template: camera (50mm/85mm), lighting (softbox / Rembrandt / high-key), skin descriptors (warm beige, natural freckles, specular highlights), expression, wardrobe, background. Keep a fixed core prompt for consistency.
4. Run batches (10–30 generations). Use same seed or saved model where supported. Name files with metadata (model/prompt/seed/date).
5. Pick top candidates, then upscale with the tool’s “max upscale” or export to Topaz Gigapixel for 2–4x. Do manual color-grade and skin-tone match in Lightroom/Photoshop.
6. Validate identities against real refs; adjust prompts or retrain if drift occurs.

Best-for / Avoid-if:
- Best for Leonardo: teams needing consistent identity across many poses, quick persona training, studio-like clean results.
- Avoid Leonardo if you need highly stylized photographic looks or you prefer a flat monthly cost.
- Best for Midjourney: photorealistic, cinematic variations and creative lighting; avoid if you need automated, guaranteed identity fidelity across dozens of shots.

Notes: budget, skill, and stage matter—if you’re at proof-of-concept, test both with a 20-image pilot. If rolling out at scale for casting boards, invest in persona training, a dedicated upscaler, and a small post-production pass.

Compare Midjourney and Leonardo AI