# Free/Open-Source Production Stack

## Preferred benchmark stack

1. **ComfyUI (local engine)** — reproducible node-based workflow.
2. **Wan 2.2 / Wan-S2V** — restrained image-to-video or audio-driven movement.
3. **InfiniteTalk** — dialogue-driven lips, expression, head, and body performance.
4. **Kokoro-class permissive TTS or original human recordings** — temporary voices.
5. **FFmpeg** — final compositing, timing, audio mixing, and export.

## Important licensing rule

Free download does not automatically mean commercially usable. Before production:

- Verify the license for the exact code version and model weights.
- Avoid InsightFace pretrained weights unless commercial rights are confirmed.
- Prefer MediaPipe or another permissively licensed face detector.
- Do not clone a real person's voice without written permission.
- Preserve license and attribution notices required by every component.

## Suggested animation settings

- Use image-to-video with the supplied approved frame as the source.
- Keep motion strength low.
- Lock seed and workflow after approval.
- Animate one character at a time when possible, then composite.
- Use reference conditioning at the highest practical identity strength.
- Disable camera moves generated by the model; the approved composition is locked.
- Generate short sections around each spoken line rather than one long uncontrolled clip.

## Prompt — Daniel

Photorealistic restrained live-action performance. Preserve the exact older Black
bus driver's identity, face, wardrobe, posture, bus interior, terminal background,
camera, and lighting. Daniel notices Kelly lingering, pauses thoughtfully, then
speaks with quiet fatherly reassurance while continuing to handle the final bag.
Minimal natural blinking, breathing, small eye movement, subtle mouth and jaw
motion synchronized to the supplied audio. No theatrical gestures, no smile change,
no camera movement, no identity drift, no added people, no rear exit.

## Prompt — Kelly

Photorealistic restrained live-action performance. Preserve the exact approved tall,
lean blonde Kelly identity, face, wardrobe, seat position, bus interior, terminal
background, camera, and lighting. Kelly is tired, overwhelmed, and reluctant to
leave Daniel's sense of safety. She listens closely; her eyes soften, she releases
one small breath, gives a slight respectful nod, and answers quietly. Minimal natural
blinking and breathing. No crying, exaggerated smile, submissive pose, camera move,
body reshaping, identity drift, added people, or rear exit.

## Assembly

Use the benchmark source clip as the timing base. Replace only the necessary shot
regions. Retain original ambience until the professional sound stage. Mix dialogue
clearly but naturally within the terminal environment; do not make it sound studio-dry.