Short videos where your character never changes face
Most AI video tools generate every shot independently, so the character in clip 1 and clip 4 are different people — and next week they are a third. aivid holds characters, products, worlds and motifs as durable entities with locked reference images, and conditions every generation on them.
Free and working now — bring your own Gemini key. It stays in your browser and goes straight to Google.
Four axes, held across runs
Reference conditioning is not one setting. Each axis transfers something different, and each is told what it may and may not contribute.
Character
The same character across clips, runs and months — face, build, palette and signature details held by a locked reference, not re-guessed from a prompt each time.
Product
Your actual product, from your own photographs. Correct geometry, panel layout and markings — not a plausible lookalike. This is the one most tools cannot do.
World
One lighting setup, one lens character, one colour grade. Five different subjects still read as one series.
Motif
Recurring props, logo placement and end-cards that land in the same position every time.
Made by this pipeline
Hindi Shorts produced end to end — subject in, five ten-second clips out, merged and uploaded. Each stars a personified object talking to camera.
Subject in, Short out
Five steps, roughly five to seven minutes per video.
Five distinct real objects from the subject, each with a Hindi script on a five-beat structure.
A 9:16 frame per clip, conditioned on the cast's locked references.
Image-to-video on self-hosted GPU. Your render, your hardware, no third-party quota.
Concatenated to a single 50-second vertical cut.
Uploaded through your own YouTube credentials. You own the channel.
Get told when it opens
The hosted version needs per-tenant credentials and a render queue before it can be safe to open. That is being built now. Leave an address and you will hear once — when it is actually usable.
No newsletter, no sequence. One message when it opens.