Turn 1: the basic skeleton
The first prompt should ask for just the structure. Elements in the right place, correct text, without worrying about timing, easing, or aesthetics. The goal of Turn 1 is to validate the composition — whether the elements are in place and the sequence makes sense.
💡 Why start simple?
If you ask for timing and aesthetics in Turn 1 and the composition is wrong, you waste time refining something that will be redone. Approve the skeleton first, then add layers.
Turn 2: add timing
With the composition approved, Turn 2 is about timing and sequence. How long each element stays on screen, when it enters, when it exits, and the delay between elements. Claude remembers the entire structure from Turn 1 and applies the timing to it—it doesn’t start from scratch.
✓ What to specify in Turn 2
- Total video duration
- Entry/exit timestamps for each element
- General easing (ease-in, ease-out, linear)
- Delays between elements
✗ What to leave for later
- Overshoot and spring (Turn 3)
- Advanced stagger (Turn 3)
- Colors and gradients (Turn 4)
- Shadows and glow (Turn 4)
Turn 3: refine the motion
With the timing approved, Turn 3 adds expressiveness in motion. This is where the animation gains personality—overshoot, spring, asymmetric stagger between elements. Subtleties that make the difference between "functional" and "fluid".
The same animation evolves in a loop—from the static structure to the final polish.
💡 Spring vs. ease-out
Spring adds natural overshoot — the element “passes” its destination and comes back. Ease-out simply slows down. For product animations, spring tends to feel more alive and less mechanical.
Turn 4: final polish
Turn 4 is the finishing touch. Final colors, typography, gradients, shadows, glow. What turns "works" into "impressive." It only makes sense once the motion has been approved in the previous turns.
Each turn builds on the previous one — quality isn’t replaced; it’s built.
The “no one-shot” rule
A single long, detailed prompt rarely beats 4 incremental prompts in the same session. Why? Because it's impossible to perfectly describe in words something you haven't seen yet. You need to see the skeleton to know what to change.
✗ One-shot
- Long prompt full of requirements
- Claude tries to do everything at once
- Mediocre result that doesn’t impress
- Difficult to fix without redoing everything
✓ 4 Turns
- Each turn focuses on one dimension
- Approves each layer before advancing
- Great result that impresses
- Easy to adjust: only the relevant turn changes
Context that shapes quality
Claude remembers all the code generated in the session. Each turn builds on the approved state—it doesn’t start from scratch. This means the quality is cumulative, not replaced. That’s why the same session produces a better result than separate sessions.
Basic skeleton
Structure created. Claude keeps all the code in session memory. You approve the composition.
+ Applied timing
Claude reads the T1 context and adds the timing about the structure. The skeleton stays intact; only the timing is new.
+ Expressive motion
Spring, overshoot, and stagger added to the T1+T2 state. None of the previous layers are lost.
= Polished result
Colors, typography, and glow applied to T1+T2+T3. The final result includes all the layers accumulated during the session.
💡 Separate sessions break the flow
If you start a new session in Turn 2, Claude has no context from T1. You lose the accumulated memory. Always keep all 4 turns in the same conversation to make use of the cumulative context.
Module summary
Next module:
4.3 — Recap & final map: the complete library matrix, the 4 universal rules, and why this is production now.