Scene Chaining: Seamless Continuity Between Shots

One of the biggest challenges in AI video generation is continuity. When you generate individual clips, each one starts from scratch — characters shift position, lighting changes, and the whole scene feels disjointed. This is especially noticeable when you're trying to tell a story across multiple scenes: every cut feels jarring because nothing visually connects one shot to the next.

Scene chaining solves this by using the last frame of your previous scene as the starting point for the next one. The result: smooth, continuous storytelling that feels like a single take.

How it works

When you add a new scene to your timeline, Paintbrush automatically pulls the final frame from the scene above it. This frame becomes the first frame reference for your new scene's generation.

The AI model sees this frame and generates motion that naturally continues from where it left off — same character positions, same lighting, same camera angle. It's conceptually similar to how traditional animators use key frames, except here the AI fills in everything between your anchor points.

Under the hood, the reference frame is sent alongside your scene description and character references. The model weighs all three inputs — the visual anchor of the previous frame, the text prompt describing what happens next, and the character reference sheets — to produce a clip that feels like a natural continuation.

When to break the chain

Sometimes you want a hard cut. A new location, a time skip, a dramatic reveal. In those cases, simply upload your own first frame or generate a fresh one from scratch. The chain is a default, not a rule.

Hard cuts are also useful when switching between parallel storylines. If your video follows two characters in different locations, you'll want each location change to start from a new first frame rather than chaining from the other storyline's last shot.

Tips for smooth transitions

We've found a few patterns that consistently produce the best results with scene chaining:

• Keep your scene descriptions consistent — if a character is on the left in scene 1, reference that in scene 2
• Use the same setting across chained scenes for maximum continuity
• Shorter clips (3–5 seconds) chain more reliably than longer ones, since there's less drift from the reference frame
• Trim the end of a scene to get the perfect handoff frame — sometimes the last half-second has motion blur or an awkward pose
• Avoid drastic camera angle changes between chained scenes; subtle shifts work better than jumping from a close-up to a wide shot

Standard vs. Pro

Both Standard and Pro generation models support scene chaining, but Pro tends to produce smoother transitions. If continuity is critical — say, a conversation between two characters — Pro is worth the extra credits. For fast-paced montages where small discontinuities are less noticeable, Standard works great.

Common pitfalls

A few things to watch out for when using scene chaining:

Motion blur handoffs. If your scene ends with fast movement, the last frame might be blurry. The next scene inherits that blur as its starting point, which can make the opening look soft or unfocused. Solution: trim the last few frames of the scene before chaining.

Character position drift. Over a long chain of scenes, characters can gradually shift position even with good references. If you're chaining more than 4–5 scenes, consider re-establishing with a fresh first frame every few scenes to reset.

Lighting inconsistency. If your prompt for scene 2 describes different lighting than what's in scene 1's last frame, the model has to reconcile conflicting inputs. Keep lighting descriptions consistent across chained scenes, or use a hard cut when changing lighting dramatically.

What's coming next

Scene chaining is available on all plans today. We're actively working on improvements to the chaining pipeline, including better handling of camera movement, automatic trim suggestions for cleaner handoff frames, and support for audio continuity across scenes. We're also experimenting with "soft chaining" — where the model uses the previous scene as a loose reference rather than a strict starting point, allowing for more natural scene transitions.