Content operations

Creating Images With AI – FAST – That Don’t Like AI

From social media to website portfolios, the internet is a visual medium, so we need to produce images at speed. AI can do it, but it’s default products are generic and usually scream “AI” which disengages audiences

Domain

Audiences

Workflows

Publication channels

Short version

The Challenge

From social media to website portfolios, the internet is a visual medium, so we need to produce images at speed. AI can do it, but it’s default products are generic and usually scream “AI” which disengages audiences

The Solution

How was the challenge solved? What did I build?

The Outcome

What did I finish and how did it solve the channel? What are the next steps?

The Project story

Conversation Distiller (Claude Skill)

A tool that accepts personal material — pasted text or PDF — and rewrites it in two anonymised forms: a named case study (Florence, a composite character) and an unnamed composite scenario. Both versions preserve the psychological and experiential truth of the material while removing all identifying detail, making the output safe to use in other tools, share with collaborators, or publish directly. Identified during build as a potential standalone offer with a market beyond The Analogue Project.

---
name: conversation-distiller
description: >
  Distils a conversation into four polished documents: a theoretical synthesis (PDF + DOCX)
  and a fictionalised narrative (PDF + DOCX) featuring Florence, a character who carries
  the intellectual and personal material anonymised from the conversation.
  Use whenever the user says "distil this", "run the distiller", "give me the four documents",
  "turn this into a Florence document", or references a conversation and asks for synthesis
  or narrative output. Produces exactly: [title]_synthesis.pdf, [title]_synthesis.docx,
  florence_[title].pdf, florence_[title].docx.
---
 
# Conversation Distiller
 
Produces four documents from a conversation: theoretical synthesis (PDF + DOCX) and
Florence narrative (PDF + DOCX). Read this skill fully before starting.
 
---
 
## Step 1: Extract and Organise the Raw Material
 
Read the conversation carefully. Extract:
 
**For the synthesis:**
- Theoretical positions and arguments (name the thinkers where relevant: Barrett, Jung, Szasz, von Franz)
- Critiques of frameworks or industries (Sedona Method, no-self teaching, self-help industry)
- Distinctions that matter (nourishment vs repair, constructed self vs constructing Self, subject vs object)
- Drafted content pieces (About page copy, LinkedIn posts, YouTube scripts, blog posts)
**For the Florence narrative:**
- Personal experiences that illustrate the theoretical points — anonymise fully
- Moments of rupture or recognition (e.g. an email that revealed a system's true logic)
- Clients, encounters, rooms, practices — all anonymised and fictionalised
- The emotional texture: what it felt like to be inside the methods, to leave them, to see clearly
Do not carry personal names, identifying details, or family specifics into either document.
The Florence narrative should read as fiction, not memoir.
 
---
 
## Step 2: Determine the Title
 
Scan the conversation for its dominant theme or the sharpest single idea.
Use this as the title root — e.g. "The Gap", "Nourishment", "The Constructing Self".
The synthesis becomes `[Title]_synthesis` and the narrative becomes `florence_[title]`.
If no clear dominant theme emerges, use the date: `YYYY-MM-DD_synthesis`.
 
---
 
## Step 3: Generate the Synthesis PDF
 
Use `reportlab` in Python. Install if needed:
```bash
pip install reportlab --break-system-packages -q
```
 
### Design specification
- Page: A4, margins 3cm all sides
- Colours: INK = #1a1a1a, ACCENT = #4a6b5a, RULE = #c8c0b0
- Body font: Times-Roman 10.5pt, leading 16, justified
- Section headings: Times-Bold 13pt, ACCENT colour, spaceBefore 20
- Title: Times-Bold 22pt, ACCENT
- Subtitle: Times-Italic 12pt, INK
- Blockquotes: Times-Italic 10.5pt, leftIndent 20
- Horizontal rules: 0.5pt, RULE colour
### Structure
Number the sections. Typical sections include (adapt to what the conversation contains):
 
1. The Core Framework (Barrett + Jung or whatever the theoretical spine is)
2. The Political Critique (if present)
3. Why [Method/Framework] Fails — Structurally (if present)
4. The Wound the Method Finds (if present)
5. Key Distinctions (nourishment/repair, constructed/constructing, etc.)
6. What the Practice Does
7. Drafted Content (include all copy pieces verbatim, labelled by destination)
Add, remove, or rename sections to fit the conversation. Not every conversation will
have all of these. Some will have different categories entirely.
 
### Output path
`/mnt/user-data/outputs/[title]_synthesis.pdf`
 
---
 
## Step 4: Generate the Florence Narrative PDF
 
Same `reportlab` setup. Novel register — do not summarise or explain, show through scene
and interiority.
 
### Design specification
- Page: A4, margins 3.5cm all sides
- Same colour palette as synthesis
- Title: Times-Bold 26pt, INK
- Subtitle: Times-Italic 13pt, ACCENT
- Chapter headings: Times-Bold 11pt, ACCENT, allCaps effect (use `.upper()` in Python),
  spaceBefore 28, spaceAfter 14
- Body: Times-Roman 11pt, leading 18, justified
- Blockquotes (Florence's internal voice or key realisations): Times-Italic 10.5pt,
  GREY (#666666), leftIndent 24, rightIndent 24
### Florence character notes
- Early forties, never stated explicitly
- Multiple degrees, diverse skills, capable of sustained sophisticated thinking
- Life feels contracted relative to what she senses is possible
- Depression is atmospheric, not dramatic — "a grey Tuesday feeling that lasted for weeks"
- Has tried the methods seriously and for years; not naive, not a victim
- The ADHD-without-naming-it quality: diverse, restless, the gap between capacity and output
- Her relationship with her own body matters (cold water, drawing, making things)
- She is the person who thinks clearly and has been failed by frameworks that couldn't
  match the quality of her thinking
### Narrative chapters
Generate one chapter per major theme or scene extracted from the conversation.
Name chapters with Roman numerals and short evocative titles (e.g. "I. The Inventory",
"IV. The Email", "IX. Nourishment").
 
End on something unresolved. Florence is in process. Do not give her a resolution
she hasn't earned. The last paragraph should feel like a door left open, not closed.
 
Leave structural room for future characters. If a chapter naturally introduces
another person (a founder, a client, a teacher), write them as a minor figure
who could be expanded later.
 
### Output path
`/mnt/user-data/outputs/florence_[title].pdf`
 
---
 
## Step 5: Generate Both DOCX Files
 
Use `docx` via Node.js. Check it is installed:
```bash
npm list -g docx 2>/dev/null | head -2
```
If not installed: `npm install -g docx`
 
### Design specification (both documents)
- Font: Georgia throughout
- Page: A4 (11906 x 16838 DXA), margins 1800 DXA (synthesis) or 2000 DXA (narrative)
- ACCENT colour: "4a6b5a", INK: "1a1a1a", GREY: "555555"
- Body text: size 22 (synthesis) or 23 (narrative), justified
- Never use `\n` — separate Paragraph elements only
- Never use unicode bullets
- Blockquotes: italic, indented left 480 DXA, GREY colour
- Section/chapter headings: bold, ACCENT, Georgia
### Synthesis DOCX
Mirror the PDF structure exactly. Labelled content pieces go in blockquote style.
 
Output: `/mnt/user-data/outputs/[title]_synthesis.docx`
 
### Narrative DOCX
Mirror the PDF structure exactly. Chapter titles in bold ACCENT small caps style
(use `allCaps: true` in TextRun).
 
Output: `/mnt/user-data/outputs/florence_[title].docx`
 
---
 
## Step 6: Present All Four Files
 
```python
# present_files call
filepaths = [
    "/mnt/user-data/outputs/[title]_synthesis.pdf",
    "/mnt/user-data/outputs/[title]_synthesis.docx",
    "/mnt/user-data/outputs/florence_[title].pdf",
    "/mnt/user-data/outputs/florence_[title].docx",
]
```
 
After presenting, note briefly:
- How many synthesis sections were generated
- How many Florence chapters
- Whether any drafted content pieces were included
- One sentence on where the narrative ends and what is left open
---
 
## Quality checks before presenting
 
- [ ] No real names in the Florence narrative
- [ ] No identifying details (specific cities named by the person, family member names, etc.)
- [ ] All four files exist and are non-zero size
- [ ] Florence narrative ends unresolved
- [ ] Synthesis includes all drafted content pieces found in the conversation
- [ ] Title is derived from the conversation's dominant idea, not generic
---
 
## Common failure modes to avoid
 
- **Generic titling**: Do not call it "Conversation Synthesis". Derive the title from the ideas.
- **Over-explaining in the narrative**: Florence shows, she does not explain. If a paragraph
  reads like a summary of the theory, rewrite it as scene or interiority.
- **Premature resolution**: Florence does not get better at the end of a chapter. She sees
  more clearly. That is different.
- **Carrying personal details**: The email anecdote becomes Florence's email. The family
  system becomes Florence's family system, unnamed. The psychiatrist becomes a doctor in
  a city, unnamed.
- **Losing the political critique**: The no-self/neoliberal argument is a spine of this
  project. If the conversation touches it, it belongs in both documents — argued in the
  synthesis, experienced in the narrative.

Structure and analyse a conversation

A visual summary of how this works

I am a slide

Click me

Just another slide

More content to come ..

Learn more

Next steps

Depending on client needs and goals, the next steps would be one or more of the following option:

  • If all that’s required is a structured, organised summary of long unweildy conversations:
    • no further actions are necessary
  • If the information in the conversations is critical to corporate strategy or the knowledge base:
    • The outputs must be tesed to ensure that they accuratly and consistently reflect the raw conversation contents

Pasted in from previous version

[vc_row][vc_column][vc_column_text css=””]

The project

[/vc_column_text][vc_column_text css=””]

The workflow

[/vc_column_text][vc_single_image image=”8215″ img_size=”full” alignment=”center” css=””][vc_column_text css=””]

The tools

Content Analyser (Claude Artifact)

A PDF upload tool that reads a conversation document and returns six structured sections: Observation (detailed account of what the conversation contains), Interpretation (through lenses embedded in the document itself), The Hook (the most original or underexplored idea, with argument for why), Implications and Resolution (where the ideas lead and what they solve), Marketing Applications (targeting angles, taglines, process names), and Critique and Content Strategy (critical audit of confusions, plus sequencing strategy for a non-expert audience). A keyword/topic tag list is generated alongside the six sections. Output is tabbed, copyable per section, and exportable in full. Built in React using the Anthropic API.

[/vc_column_text][vc_column_text css=””]

Link to the artifact:

Content Analyser[/vc_column_text][vc_column_text css=””]

Conversation Distiller (Claude Skill)

A tool that accepts personal material — pasted text or PDF — and rewrites it in two anonymised forms: a named case study (Florence, a composite character) and an unnamed composite scenario. Both versions preserve the psychological and experiential truth of the material while removing all identifying detail, making the output safe to use in other tools, share with collaborators, or publish directly. Identified during build as a potential standalone offer with a market beyond The Analogue Project.

[/vc_column_text][vc_column_text css=””]

Claude Skill
Distils a conversation into four polished documents: a theoretical synthesis (PDF + DOCX) and a fictionalised narrative (PDF + DOCX) featuring Florence, a character who carries the intellectual and personal material anonymised from the conversation. Use whenever the user says "distil this", "run the distiller", "give me the four documents", "turn this into a Florence document", or references a conversation and asks for synthesis or narrative output. Produces exactly: [title]_synthesis.pdf, [title]_synthesis.docx, florence_[title].pdf, florence_[title].docx.

[/vc_column_text][vc_column_text css=””]

Content Strategy

Four cross-cutting editorial principles govern all content: never write from the methodology; treat the reader as experienced; sequence without funnel; use personal voice as argument, not confession.

Foundational blog posts

Three posts establishing the intellectual and positioning ground for all subsequent content:

  • What drawing does — mark-making as interruption of the prediction loop; the I that makes the mark is stronger than the I that thinks about making it
  • What do we mean by coherence — the Jungian Self as the precise term for the organising principle the work is aimed at; convergence with Barrett’s predictive coherence
  • How releasing an emotion tells you something about how it was built — the Sedona-based process as interruption of the construction mechanism; the distinction between deconstructing experience (useful) and deconstructing the self (creates more problems than it solves)

These posts are pasted into a Google stored in the “Claude Blog Posts” folder and a quick skim reveals a LinkedIn style littered with AI cliches, so they will need some brutal editing.

 

[/vc_column_text][/vc_column][/vc_row][vc_row][vc_column][vc_column_text css=””]


Identified but not built

Content Angles Tool

  1. Set up a simple document to capture blog post title ideas on the go.
  2. Create a separate artifact that would suggest perhaps three different angles on the blog post idea
  3. After human review and revision, it would then write a draft post using the documents created by the workflow above and including links to academic sources and references

What the anonymiser suggests as a future offer

The Material Anonymiser has many applications — therapists, coaches, facilitators, researchers — accumulate years of session material they cannot publish as-is. Writers who want to use personal experience as source material often can’t work with it directly. The tool addresses both without requiring the user to think of it as an AI tool — the value proposition is editorial: a process for converting personal material into usable form while preserving psychological truth. This warrants development as a standalone product with its own positioning.

But since I haven’t checked its output yet, it would most likely required some work before this would be viable. And then the question would be why potential clients wouldn’t just do it themselves.

Next steps — for the PM course

  1. Test the Content Analyser with a range of conversation PDFs — different lengths, different levels of theoretical density — and calibrate the prompts against output quality
  2. Test the Material Anonymiser with real personal material and assess how well the psychological truth is preserved across both output versions
  3. Build the Content Angles Tool as a separate artifact, taking two PDF inputs (source conversation + angles document) and generating post outlines
  4. Develop the anonymiser as a standalone offer — write positioning copy, identify the primary market (practitioners vs. writers vs. content creators), and test the framing
  5. Publish the three foundational posts and map response against the content strategy — which stage is the existing audience in, and is the content meeting them there
  6. Build the full content pipeline connecting the three tools: anonymiser → content analyser → angles tool, with a consistent file-naming and storage convention so material doesn’t get lost between stages

What this project demonstrated

That the content production bottleneck is not ideas — for many people there is no shortage of those — but the gap between the quality of thinking in the raw conversations and the usability of that thinking as publishable material. The system built here addresses that gap directly. The tools are functional. The strategy is coherent. The intellectual foundation is in place. What remains is testing, calibration, and the decision about which elements to develop as offers in their own right.

[/vc_column_text][/vc_column][/vc_row][vc_row][vc_column][vc_raw_html]<iframe src=”https://claude.site/public/artifacts/520b0e20-4d87-4d1b-a4ad-89e6ece63df4/embed” title=”PDF Conversation Analyser” width=”100%” height=”600″ frameborder=”0″ allow=”clipboard-write” allowfullscreen></iframe>[/vc_raw_html][vc_raw_html]PHA+SSBhbSByYXcgaHRtbCBibG9jay48YnIvPkNsaWNrIGVkaXQgYnV0dG9uIHRvIGNoYW5nZSB0aGlzIGh0bWw8L3A+[/vc_raw_html][/vc_column][/vc_row]