AI models
Wan 2.2 Image Trainer + CutScene AI
trains dual Wan LoRA weights with automatic face alignment so creative teams can personalize Wan 2.2 for recurring characters and signature styles.
Overview
Why creators use Wan 2.2 Image Trainer
Wan 2.2 Image Trainer trains dual Wan LoRA weights with automatic face alignment so creative teams can personalize Wan 2.2 for recurring characters and signature styles.
CutScene compresses your curated stills, triggers Wan’s trainer, and returns low- and high-noise weights plus the config file straight into the character profile in under ten minutes.
Highlights
- Built-in face detection, cropping, and masking deliver clean subject framing even from candid references.
- Outputs both diffusers (low-noise) and high-noise LoRAs plus the config JSON so inference endpoints stay in sync.
- Integrated credit tracking and webhook callbacks let teams monitor progress without leaving CutScene.
Workflow
Wan 2.2 Image Trainer inside CutScene's free editor
Wan 2.2 Image Trainer integrates with CutScene's professional video editor—which is 100% free—so you can generate AI content and edit it seamlessly. Only pay for AI generation; editing, timeline tools, and exports cost nothing.
Producers, illustrators, and editors share the same training sets and trigger phrases, ensuring each department stays on-model without swapping ZIP files outside the platform.
Recommended steps
- Collect polished portraits or keyframes inside CutScene and review them with stakeholders.
- Start Wan 2.2 Image Trainer from the character drawer; CutScene packages the dataset and launches the job with your trigger phrase.
- Use the returned LoRA pair inside Wan’s text-to-image workflow for perfectly aligned storyboards and marketing art.
Optimization
Prompting tips for Wan 2.2 Image Trainer
Get better results from Wan 2.2 Image Trainer with clear direction and reference shots to match your creative vision.
Upload a diverse set of 15-25 shots covering lighting, angles, and expressions—keep `use_masks` enabled unless you are training style LoRAs so Wan can lock onto the subject cleanly.
Tips to try
- Blend tight headshots with waist-up compositions to help Wan generalize wardrobe and posture.
- Craft a distinctive trigger phrase (e.g., `acme_hero_alpha`) to avoid collisions with other LoRAs in your library.
- Leave `include_synthetic_captions` off when your dataset already includes balanced descriptions.
Keep exploring
Discover more ways CutScene helps your team build cinematic stories with AI-powered workflows.
Frequently asked questions
Answers to the most common questions about this workflow and how to maximize CutScene for your team.