Method and limitations
A styling concept is useful only when its reasoning stays inspectable
Image quality alone is not our evaluation target. A polished tree that no longer relates to the source is a poor source-specific preview. Our method therefore keeps provenance, visible structure, controlled variation and limitation disclosure together.
Read the source. Compare the direction. Keep the limits visible.
Every guide is designed to return the decision to the real tree rather than a generic finished silhouette.
What matters most
- The source image defines the visible evidence available to the system and reviewer.
- Four outputs explore variation inside one selected style rather than pretending to deliver one correct answer.
- Evaluation rewards continuity and useful alternatives, not biological prediction.
1. Source contract
The user supplies one front-facing photo with the base, full trunk and primary branches visible. That photograph becomes the reference for identity, dominant movement and branch evidence.
We recommend a quiet background and level camera because overlap and distortion increase uncertainty. The source cannot prove hidden roots, health, live veins or branch flexibility, so those remain outside the concept contract.
2. Controlled design inputs
The user selects one of six supported styles, a 1, 3 or 5-year visual horizon and one presentation. These inputs separate direction, perceived maturity and display context so comparisons can remain interpretable.
The horizon controls visual maturity, not promised biological time. Presentation can include a pot or show setting, but it does not function as a pot recommendation service.
3. Four materially different concepts
A generation produces four alternatives that share the source and selected direction but vary branch emphasis, negative space, canopy distribution and refinement. Variation is useful only when each option still reads as the same source tree.
The set is not ranked as four increasingly correct answers. It is a small decision space intended to reveal trade-offs and create more precise questions for critique.
4. Human evaluation and disclosure
Reviewers compare trunk continuity, taper, branch origins, depth, negative space and unsupported invention. A result can be visually attractive and still fail the source-specific test.
We disclose that concepts are visual inspiration and recommend direct professional review before irreversible work. Indexed examples require provenance and cannot be presented as guaranteed transformations.
Questions people ask
What does structure-aware mean?
It means the visible source trunk and major branch relationships are treated as anchors for evaluation, even though an image model can still invent or alter details.
Why produce four concepts?
Multiple concepts expose trade-offs and reduce the false impression that one generated image is the uniquely correct future for a living tree.
Continue the decision
About Bonsivo
Make the styling conversation visible before the tree changes
Bonsivo helps people explore structure-aware styling directions from one real tree photo while keeping the source visible, the product boundary explicit and important horticultural decisions in human hands.
Read this guideDesign principles
Bonsai design principles are relationships, not decoration rules
Understand practical bonsai design principles—trunk identity, taper, branch hierarchy, asymmetrical balance, rhythm, depth and negative space—and use them to evaluate visual concepts against a real tree.
Read this guideSource photo guide
A useful bonsai photo records evidence, not atmosphere
Prepare a clear bonsai source photo for styling visualization: choose a front, set camera height, reveal the base and primary branches, control background and light, and avoid angles that distort structure.
Read this guideSource-to-concept evidence
Compare every future concept with its source structure
Explore six transparently labelled AI demonstration pairs, inspect which trunk and branch relationships remain visible, and learn how to judge a Bonsivo styling concept without mistaking it for a growth forecast.
Read this guideUse the real source
Evaluate the method with your own source
Upload one clear tree photo, keep it beside all four results and inspect what was preserved or invented.