Call Sheet · Image and pre-production

The best AI models for character design

One character who looks the same in every frame. These are Call Sheet's picks for this job, in order, with the reason for each. Call Sheet is a free tool by Melissa Beery that ranks 101 image, video and sound models for a creative job. Its data was last updated on 20 October 2026.

1

Stable Diffusion / ComfyUI

Stability + open community

Trains on your character, so it stays on model across a whole show.

Maximum control.

Cost
Free / local (your GPU). High setup effort
Where
OpenArt
Rights
You own the pipeline and can train on cleared data, so you control where the data goes.
Watch out
It runs on your own machine, so setup is steep.
How to ask itpositive: [subject, style, detail tags] | negative: [unwanted] | + ControlNet / LoRA / IP-Adapter conditioning

Split it into a positive prompt and a negative prompt.

Keep the positive prompt on subject, style and detail. Use the negative prompt only to remove artifacts.

Open Stable Diffusion / ComfyUI in Call Sheet →
2

Midjourney V8.2

Midjourney

Its character and style references keep one character and one look across the boards.

Unmatched art direction through style references.

Cost
Own platform / subscription
Where
Midjourney's own site · check in app: KREA
Rights
Web‑trained.
Watch out
Won't follow a brief literally.
How to ask it[subject] [doing what], [medium/style], [lighting], [composition] --ar 16:9 --stylize <0-1000>

Keep it under about forty words. This model rewards concision and ignores over‑specification.

Order matters: subject, then medium and style, then lighting, then composition.

Open Midjourney V8.2 in Call Sheet →
3

FLUX.2 (pro / flex / max)

Black Forest Labs

Multi‑reference editing keeps design details locked.

Literal prompt adherence and photoreal control.

Cost
Mid. Local open variant available
Where
Magnific, Higgsfield, Runway · check in app: KREA, OpenArt
Rights
Check terms.
Watch out
Less painterly than Midjourney.
How to ask it[Subject] + [action] + [style] + [context] + [lighting] + [technical]

Describe only what should be present. There is no negative prompt, so "no cars" can put cars in the frame.

Be literal and specific. This model follows exactly what is written, and word order matters.

Open FLUX.2 (pro / flex / max) in Call Sheet →
4

Nano Banana Pro / Imagen (Google)

Google

Up to 14 reference images to composite a character into new shots.

Precision + real‑world knowledge.

Cost
Mid (Pro). A faster/cheaper tier exists
Where
Magnific, Higgsfield, KREA, OpenArt, Runway
Rights
Google terms, plus a SynthID watermark.
Watch out
Less 'art‑directed' personality than Midjourney out of the box.
How to ask it[Subject] + [action] + [location] + [composition/shot] + [style & lighting]. Edits: [what to change] + [what to keep] + [refs]

Write full descriptive sentences, not keyword lists.

Put any on‑image text inside double quotes, spelled exactly as it should appear.

Open Nano Banana Pro / Imagen (Google) in Call Sheet →

Side by side

●●● strong · ●●○ middling · ●○○ weak, as rated in Call Sheet
ModelControlConsistencyEditingCost
Stable Diffusion / ComfyUI●●●●●●●●●Free / local (your GPU). High setup effort
Midjourney V8.2●●○●●●●●○Own platform / subscription
FLUX.2 (pro / flex / max)●●●●●●●●●Mid. Local open variant available
Nano Banana Pro / Imagen (Google)●●●●●●●●●Mid (Pro). A faster/cheaper tier exists

Describe your own character design job in Call Sheet and it ranks every model for it, then writes the prompt.

Open Call Sheet →

Head to head

More jobs