State what the image must do and what cannot change.
Image generationUpdated 2026.07.29
Image Director
Turn a rough visual idea into an image prompt that can be generated, inspected and corrected without requiring photography terminology first.
For people who know the image they need, but not the camera terms
Turn a rough visual idea into a prompt you can inspect and correct
Describe the purpose, subject and constraints in plain language. The workflow turns them into explicit visual decisions before generation.
- Start in plain languageNo photography vocabulary is required at intake.
- Make decisions explicitSeparate subject, space, camera, light and exclusions.
- Correct one field at a timeRevise the decision responsible for the visible deviation.
Each visual choice becomes independently inspectable.
Generate, compare the result, then correct the responsible field.
Result
Get a prompt you can inspect instead of a prompt you can only retry
The workflow separates requirements that are often mixed together into conditions that can be chosen, checked and revised independently. It replaces vague directions such as “more cinematic” with explicit decisions about subject, shot size, angle, composition, focus, light and constraints.
Start here
Start with the job, subject and non-negotiable constraints
Start with what the image must accomplish, what it must contain and what cannot change. The workflow then selects a task type, structures the visual conditions, chooses the required output depth and checks the prompt before generation.
Input and output
Prompt assembly order
Define what is in the image first, then how it is viewed, and finally the conditions that must not drift.
- 1
Subject and action
- 2
Environment and spatial relationship
- 3
Shot size and camera angle
- 4
Composition and perspective
- 5
Focus, light and colour
- 6
Constraints and exclusions
Use cases
Suitable image tasks
It can begin from a short brief for editorial covers, product concepts, campaign key visuals or reference-image reconstruction, as long as the intended use and non-negotiable constraints are stated.
Method
What was changed
Upstream material provides vocabulary. Image Director adds a decision order, removes conflicting camera language, separates visible facts from inference and makes correction possible without rewriting the entire prompt.
Sources
Where the method comes from
The method combines traceable concepts from three upstream repositories: the HunyuanVideo 1.5 prompt handbook, AI Cinematic Prompt Director and the Seedance 2 prompt library. Video-only and conflicting concepts were removed before the image workflow was assembled.
Limits
What it can do and what it cannot guarantee
The workflow makes image decisions inspectable and correctable. It cannot guarantee identical output across models, recover hidden camera settings from a reference or invent creative intent that the brief does not provide.
Access and version
Access and version
The fixed v1.0.0 package is available under the MIT License. Automated package checks pass on GitHub Actions; independent non-owner usability validation is tracked separately and is not yet claimed.