Commission a Machine
Name a subject and the studio drafts it as Leonardo would: a titled mechanism, a drawn blueprint, and the exact prompt used to render it.
Script Written Backwards
Leonardo wrote right to left, readable only in a mirror. Write plainly here; the studio returns the script as it would sit in the codex, and can read mirrored script back again.
The Mirrored Page
Reversed character by character, as the hand wrote it. The reflection view shows the same page held to a looking glass.
Measured Against Phi
Every subject is measured against the Divine Proportion, 1.618, and the canon of the body: four fingers a palm, four palms a foot, four cubits the height of a man.
The Bound Notebook
Everything the studio produces can be bound here. The Codex is held for this sitting; download it as a manuscript file to keep it beyond the session.
DaVinci Protocol: Text-Stability Pipeline
Video models smear intent over time. This pipeline locks glyph geometry before motion is applied: typography is rendered once, perfectly, as an anchor frame, and the video model is asked only to read it, never to write it.
The Typesetter
Render perfect typography into pixels with a text-accurate image model. Verify spelling, kerning, and edge clarity in a single still frame. That frame is the immutable truth for everything downstream.
Text-to-Image, 4K / 8K still
The Animator
Feed the anchor as first-frame conditioning so the model computes only deltas. Constrain motion to low-magnitude camera and light shifts. No 3D rotations: rotation is where glyphs warp and morph.
Image-to-Video, first frame locked
The Polisher
Recover the edge fidelity that low-resolution diffusion and compression spent. Upscale with a refiner tuned for text edges, denoise the high-contrast regions, and interpolate to 60fps for readability.
Upscale to 8K · Interpolate 60fps
Why It Works
Standard text-to-video hallucinates glyphs fresh in every frame, so letters melt. Lock the pixels in space first, then move them in time.
Failure State · Standard T2V
Root cause: stochastic hallucination of glyphs, re-decided per frame. The sign never had a fixed shape to hold.
The DaVinci Bridge: the anchor frame crosses from image to video timeline with its geometry locked.
Solved State · DaVinci Pipeline
One rendering of the word, carried through motion and polish untouched. Output spec: 8K / 60fps.
Phase 1 · Build the Anchor Prompt
The exact words for the sign, the scene around them, and the look. This prompt goes to Ideogram or Flux, never to the video model.
Phase 2 · Build the Motion Prompt
Upload the anchor still as the first frame. The motion prompt asks for deltas only, and names the text as untouchable.
Phase 3 · Polish Settings
The reconstruction pass, in order. Run all three; each recovers a different kind of loss.
- Upscale to 4K / 8K in Topaz Video AI with the Proteus or Iris model, tuned for text edges rather than faces or foliage.
- Frame interpolation to 60fps so letterforms hold steady through the motion instead of strobing at 24.
- Edge-focused denoise on high-contrast regions only; global denoise softens the very edges the pipeline exists to protect.
Operational Guardrails
Production rules from the framework. They are not suggestions; each one maps to a failure already seen in the wild.
- The anchor typography is the only source of truth. Text is never re-rendered in the video stage.
- If text degrades: reduce motion magnitude, shorten the duration, or re-run with a stricter “keep text static” instruction. In that order.
- Prefer rigid-body text motion (translate, scale) over warp or rotate. Perspective rotation is the primary melt trigger.
- Verify the anchor still before any motion is generated. A misspelled anchor produces a perfectly stable misspelling.