RESEARCH CHAPTER 12
How images not grounded in a single capture reorganize the sources of time
Generative models rewrite photography’s ‘that-has-been’ as a statistical past of training data, weights, and prompts.
Central question
When an image is not grounded in a single photographic event, how do training materials, model version, prompt, stochastic process, and revision chain constitute its temporal provenance?
Questions still open
- Which actual generative practices can supply a complete evidence chain for model version, training source, prompt, random seed, human revision, and bodily consent?
Argument and analysis
Generative models rewrite photography’s ‘that-has-been’ as a statistical past of training data, weights, and prompts.
The post-photographic question is not whether generated images have time but how the source of time is redistributed: who supplies bodies, places, archives, and labour, and who owns version, consent, and withdrawal? A model’s probability space can be compared with Deleuze’s virtual but cannot be equated with it.
Create provenance cards for model version, training source, prompt, seed, revision chain, and bodily consent.
Scope
Hypothesis: the current film examples are primarily fictional narratives. They can formulate the problem but cannot replace pipeline and provenance evidence from actual generative practice.
- Chinese manuscript locator
- Bearing Time Chinese DOCX, paragraphs 1020–1087.
References
- 01
CHINESE RESEARCH MANUSCRIPT
Bearing time: the expanded thirteen chapters
The complete Chinese chapter and its bibliography.
- Locator
- DOCX paragraphs 1020–1087
- 02
ATLAS ANALYSIS
The next step: real-time production of temporality
Poses the research problem of how generative systems produce tense in real time; it is not a technical audit of existing models.
- Locator
- Generated-image and tense-production sections of The Next Step of the Time-Image manuscript.
- Note
- Operational time-image, temporal apparatus, and platform temporality are research names; model probability space is not Deleuze’s virtual.
- 03
ATLAS ANALYSIS
The Reproduction of Time: expanded revision
Examines training material, weights, versions, and labour as temporal sources of generated images.
- Locator
- Post-photographic and generative-model argument in the revised Reproduction of Time manuscript.
- Note
- P4 materials are primary self-description, not independent scholarly certification; Kakguan lacks a sufficient public archive.
- 04
ATLAS ANALYSIS
Deleuze time-image film catalogue
Compares action, memory, sound, and media material across films.
- Locator
- The corresponding film entries in the Deleuze time-image film catalogue.
Films and concepts
Concepts and terms · 2
- 01
- 02
Films and viewing entries · 2
- 01
- 02
Counterexamples and limits
Artificial intelligence, copied memory, or synthetic characters in fiction do not establish how real generative pipelines operate. Without actual records of provenance, consent, versions, and withdrawal, the chapter remains a question rather than a general politics of generated-image time.
Further reading
Wear, format, resolution, architecture, and the performer’s body let time settle inside the material of the medium.
06Archives preserve materials while later history changes their evidentiary meaningAn archive’s temporality comes from the institutional chain of filming, naming, classifying, storing, accessing, and reusing.
11The future has not happened, yet it already manages the presentPrediction, recommendation, and risk scores press the future into the present, but an algorithmic future is not Deleuze’s open virtual.
13When heterogeneous times rewrite one another rather than merely coexist on screenThe final chapter turns the time-image into a testable relation: at least two heterogeneous times rewrite one another, shot, sound, editing, or apparatus carries the change, and a replayable consequence remains.