Session 1 · Tool Lab · Lesson 04
Image to Text to Image
Workflow: Visual Translation
A single photograph goes through the machine twice: image becomes description, description becomes image. The drift between them is interpretive reweighting — and it is the material.
Concept
- Every translation loses, invents, and re-emphasizes.
- The machine's description is a reading, not a record.
- Comparing source and output makes the model's priorities visible.
What survives the translation — and what does the system invent to fill the gaps?
Slides
- 14
Transition to Making
- Images become inputs
- Meaning becomes process
- Systems replace compositions
- Photography becomes relational
- 15
Interpretive Workflow Overview
- Archive interrogation
- Visual translation
- Rule-based interpretation
- 17
Workflow 2: Visual Translation
- Input: single photographic image
- Process: image → text → image
- Focus: drift and loss of meaning
- Result: interpretive reweighting
- 19
Instructor Demonstration
- Live walkthrough of one workflow
- Showing failures and misreads
- Adjusting rules rather than prompts
- Connecting output back to concept
Student activity
- 01Ask ChatGPT to describe your source image, then paste that description unedited into Leonardo and generate.
- 02Upload the output, and record what was lost, what was invented, and what was over-emphasized.
Input: single photographic image · Process: image → text → image · Focus: drift and loss of meaning.
01Choose a source image
Pick one photograph you know well — you will compare the round trip against your own memory of it. Keep it open next to this page; nothing is uploaded here.
02Describe it with ChatGPT
Not startedChatGPT
Multimodal chat with a free tier. The workshop's default interpreter — give it an image and it will describe, read, and re-read what it sees.
Upload your source image to ChatGPT, send the prompt below, and copy the reply into Machine description — verbatim, mistakes and all.
Prompt to try“Describe this photograph in two or three plain sentences. What objects, people, and setting do you see, and what is happening? Report only what is there — do not interpret.”
External tool — it has its own privacy policy and may change or require an account.
03Regenerate with Leonardo
Not startedLeonardo
Image generation with a free daily token allowance. The workshop's default generator for the image → text → image translation step.
Paste the machine description into Leonardo as the prompt — unedited — and generate. The drift is the material; do not improve it.
Prompt to tryFill in the machine description above — it becomes your prompt.
External tool — it has its own privacy policy and may change or require an account.
Compare source and output