Skip to main content

Iris

Iris logo
Iris documentation
localized video editing

Prompt-driven editing for reality changes

Iris is built around a simple idea: point at a specific moment, describe what changes, compare multiple interpretations, and keep editorial control over what happens next.

selectpromptacceptpropagate
what Iris is best at
  • Changing one part of one moment without rebuilding the whole shot
  • Comparing variants instead of trusting one blind generation
  • Carrying accepted edits forward through continuity-aware flows
Iris is strongest when you want to change one part of one moment without regenerating an entire scene from scratch.

Localized edits

Change the exact region and moment you care about instead of replacing an entire sequence.

Variant selection

Generate multiple outcomes, compare them quickly, and choose the one that actually works.

Continuity follow-through

Push accepted edits through adjacent frames so the result feels authored, not accidental.

core interaction

The editing loop

Iris is not a one-shot generator. It is an editing system built around timing, selection, comparison, and acceptance.

1

Select the moment

Upload a source clip, scrub to the exact frame or range you care about, and isolate the subject with a box or mask-assisted selection.
2

Prompt the change

Describe the change in plain language. Iris turns that edit intent into candidate video variants instead of asking you to rebuild the entire shot.
3

Choose and continue

Review the interpretations, accept the strongest result, and carry that choice forward through continuity-aware editing.
1
source moment

Every change starts from a precise frame or short range.

N
generated options

The product expects comparison, not blind trust in a single output.

1
accepted result

Acceptance is part of the interaction model, not a fallback.

+
continuity carry

The accepted look can continue into surrounding context.

Iris is for editorial precision. The product shines when a human already knows what should change and wants generation to stay inside that decision.
Iris is not a one-shot text-to-video tool. It is an editing system built around:
  • Localized edits instead of full-scene regeneration
  • Variant selection instead of single-output guessing
  • Continuity propagation instead of isolated frame changes
  • Project memory instead of disposable sessions
what exists today

Current product surfaces

Landing and routing

A landing page, a projects page, and an editor route frame the end-to-end experience.

Editing workflow

Upload, selection, prompt-driven generation, review, acceptance, and continuity tooling.

System support

FastAPI routes, AI observability, object storage support, and an optional GPU worker.

Why the workflow matters

Traditional editors are precise but manual. Generators are expressive but often too global. Iris sits between those two extremes. The goal is not to replace editing. The goal is to make specific editorial changes feel direct.
The defining Iris interaction is: scrub to a moment, select the region, describe the change, review variants, then accept the strongest result.

Start here