Short answer: crowd multiplication turns a few dozen extras into a full venue. Traditionally that meant shooting the same group in many positions and compositing the tiles; AI adds variation, different faces, clothes and movements, so repeated people are not noticeable. The foreground stays real; the depth is filled and varied.
Two approaches
- Tiling real extras: the group is filmed in several sections of the venue with a locked camera, then combined. Very believable, limited by the number of positions shot.
- AI-generated crowd: the far and middle ground is generated to match the venue and light. Fast and flexible, best when figures are small.
Most productions mix them: real people close to camera, tiled extras in the middle, generated crowd at the back.
What breaks crowd shots
- The same person or costume repeating in view.
- Everyone moving in sync, real crowds are never uniform.
- Figures lit differently from the venue.
- Close, lingering shots of generated faces.
Shooting for it
Lock the camera for tiled sections, keep extras’ costumes varied, record the venue empty, and note which areas each tile covers.
Related work
The concert setting of MIDIVOXX shows how a stage world can feel full. Read more on the AI VFX page.
Frequently asked questions
How many extras are needed for a full stadium?
Often a few dozen, filmed in several positions for the foreground and middle ground, with the distance filled by AI or tiles.
Can AI create a whole crowd?
Yes for distant and mid-ground figures. Close faces are usually real people for believability.
How do you avoid a repetitive crowd?
Vary costumes, movement timing and positions, and never show the same figure twice in frame.
Related guides
This work at Leopati
Service: AI Visual Effects Production
