BABUNI.DEV

Anskar Grotzschnapf / Overview

Anskar Grotzschnapf: Overview

Anskar Grotzschnapf

Local poetry archive and daily artist. Once a day this machine can:

  1. Choose one cleared poem from the archive
  2. Generate a 1600×1600 artwork locally (sd-cli)
  3. Prepare a print pack
  4. Publish to Instagram (@anskargrotzschnapf)
  5. List a quantity-1 SEK digital download on Etsy when the shop is ready

Version 1.0 is that automated loop with the noon wake timer and Etsy maintenance armed. Manual Instagram and Etsy publish commands still require --confirm.

Permanent agent instructions: AGENTS.md. Short history: CHANGELOG.md. Design/task records: docs/tasks/.

Setup

Python 3.14 provides SQLite in the standard library. setuptools is not installed, so the project is not installed as a package and nothing is fetched from the network.

python3 -m venv --without-pip .venv

The virtual environment at .venv was created with that command.

Daily automation (hands-off)

Prefer one noon publisher:

Also enable the Etsy maintain user timer (sales, day-30 discount, day-42 removal).

PYTHONPATH=src .venv/bin/python -m anskar schedule status
PYTHONPATH=src .venv/bin/python -m anskar schedule install --enable
PYTHONPATH=src .venv/bin/python -m anskar schedule prepare-system

After prepare-system, activate the system wake timer (and disable the user noon timer) with the commands that command prints. Do not leave both noon timers enabled.

PYTHONPATH=src .venv/bin/python -m anskar daily-status
PYTHONPATH=src .venv/bin/python -m anskar daily-run          # one-off live run
PYTHONPATH=src .venv/bin/python -m anskar daily-resume <id>  # finish an incomplete run

daily-run auto-resumes an incomplete live run. At most one live Instagram artwork per Swedish calendar day (Europe/Stockholm). Desktop notifications fire on major stages; set ANSKAR_NOTIFY=0 to silence them.

Commands

Import the Project Gutenberg collection:

PYTHONPATH=src .venv/bin/python -m anskar import data/collections/gutenberg-public-domain.json

Select a poem. The same pending poem is shown on every later run until it is confirmed or skipped.

PYTHONPATH=src .venv/bin/python -m anskar select
PYTHONPATH=src .venv/bin/python -m anskar select --confirm
PYTHONPATH=src .venv/bin/python -m anskar select --skip

--confirm marks the pending poem used. --skip dismisses it and leaves it unused. The database is data/archive.sqlite. Set ANSKAR_DB to use a different file.

Generate one PNG from the pending poem. The stored poem is not rewritten. By default Ollama writes one Surreal Abstract visual prompt, Painterly shapes its surface, and Z-Image Turbo makes the image. The poem stays pending. Ollama is unloaded before image generation.

PYTHONPATH=src .venv/bin/python -m anskar generate --pending --dry-run
PYTHONPATH=src .venv/bin/python -m anskar generate --pending
PYTHONPATH=src .venv/bin/python -m anskar generate --pending --print-inches 10
PYTHONPATH=src .venv/bin/python -m anskar generate --pending --style none
PYTHONPATH=src .venv/bin/python -m anskar generate --pending --style dreamlike
PYTHONPATH=src .venv/bin/python -m anskar generate --pending --model z-image

--style accepts painterly (the default), dreamlike, symbolic, or none. --composition accepts wrapped (the default), style_text, or text. none and text send the stored poem by itself. The other compositions ask Ollama for the image prompt. --abstraction accepts gentle, surreal (the default), or experimental. --model accepts an installed model. The default is Z-Image Turbo. The default image is 1600×1600, an 8×8 inch print at 200 PPI. --print-inches sets a square side at 200 PPI and rounds up to a multiple of 8. --width and --height are exact pixels, multiples of 8 from 64 through 2400. A request is not silently reduced. A poem that already has a successful artwork is not generated again.

The generator is the installed sd-cli. Paths and sampler settings can be overridden with ANSKAR_SD_CLI, ANSKAR_SD_MODEL_DIR, ANSKAR_SD_DIFFUSION_MODEL, ANSKAR_SD_TEXT_ENCODER, ANSKAR_SD_VAE, ANSKAR_SD_BACKEND, ANSKAR_SD_STEPS, ANSKAR_SD_SAMPLER, ANSKAR_SD_CFG, ANSKAR_SD_WIDTH, ANSKAR_SD_HEIGHT, ANSKAR_SD_OFFLOAD_TO_CPU, ANSKAR_SD_FLASH_ATTN, ANSKAR_SD_TIMEOUT, and ANSKAR_ARTWORKS_DIR. Images are written to data/artworks/. A JSON file beside each new image keeps the poem and the visual prompt. Approving that file for print is a separate command.

Open the local prompt lab in a browser. It generates from text you type and saves each result under data/experiments/. Direct generation sends that text through the house style. Interpret with Ollama asks the local qwen3:8b model for a visual prompt, which stays editable, and then sends that prompt by itself. Abstraction can be Gentle, Surreal Abstract, or Experimental. The model list includes Z-Image Turbo, Z-Image, RealVisXL V5, and DreamShaper XL when those files are installed. DreamShaper XL is for experiments and local previews until its commercial license is documented. House style defaults to Dreamlike; Painterly and Symbolic stay available under Advanced. It does not change the poetry archive. Ollama is expected at 127.0.0.1:11434.

Prepare a local print pack from one saved image. The pack copies that file as the print master. It does not upscale or generate a second image. It classifies 4, 6, 8, 10, and 12 inch sizes from the real pixel count, writes a digital-download listing, and makes a smaller Instagram preview when the long edge is above 1080. Preparing a pack does not publish. A size below 150 PPI is marked unsuitable.

PYTHONPATH=src .venv/bin/python -m anskar print data/experiments/<experiment>/<file>.json

PYTHONPATH=src .venv/bin/python -m anskar lab

The lab page is http://127.0.0.1:8765/. Use another port with --port.

Instagram

Instagram uses the official Instagram API with Instagram Login. The token stays in ~/.config/anskar/instagram.json, outside this repository. Put the token in a file by itself and install it from that file. The token is not printed.

PYTHONPATH=src .venv/bin/python -m anskar instagram install-token --from /path/to/token-file
PYTHONPATH=src .venv/bin/python -m anskar instagram status
PYTHONPATH=src .venv/bin/python -m anskar instagram prepare <pack-id>
PYTHONPATH=src .venv/bin/python -m anskar instagram publish <pack-id> --dry-run
PYTHONPATH=src .venv/bin/python -m anskar instagram publish <pack-id> --confirm

status only reads the account. A public https image URL (often via R2) must be configured before a live post. When fewer than seven days remain on the token, instagram renew refreshes it through Meta's ig_refresh_token grant. Meta “API access blocked” is a developer-dashboard restriction, not a missing local file.

Etsy

Etsy Open API credentials stay in ~/.config/anskar/etsy.json. Register the exact callback https://localhost:8743/etsy/callback on the Etsy app, store the keystring and shared secret, then authorize once:

PYTHONPATH=src .venv/bin/python -m anskar etsy setup
PYTHONPATH=src .venv/bin/python -m anskar etsy connect
PYTHONPATH=src .venv/bin/python -m anskar etsy status
PYTHONPATH=src .venv/bin/python -m anskar etsy test
PYTHONPATH=src .venv/bin/python -m anskar etsy set-state ready
PYTHONPATH=src .venv/bin/python -m anskar etsy maintain
PYTHONPATH=src .venv/bin/python -m anskar etsy refresh-descriptions

OAuth alone does not enable publishing — seller verification is separate (etsy set-state ready). Buyer listing copy is short: digital download note, intended print size, the source poem, and one local-AI disclosure. Gallery photos are preview-based; the paid file is the print PNG. Tokens refresh automatically. Secrets are never printed.

One-time wake rehearsal

This suspends the computer. It does not change the noon timer. --publish creates one real Instagram post; Etsy is queued only.

PYTHONPATH=src .venv/bin/python -m anskar schedule test-cycle
PYTHONPATH=src .venv/bin/python -m anskar schedule test-cycle --publish

The result is kept in data/test-cycle/latest.json. Cancel before the wake with anskar schedule test-cycle --cancel.

After pulling timer unit changes, refresh installed units (schedule install / prepare-system + sudo copy for the system timer).

Tests

PYTHONPATH=src .venv/bin/python -m unittest discover -s tests -v

Development documentation

The journal distinguishes source-based reconstruction, author direction, AI assistance, and independently verified behavior.

Documentation audit #36 — 28 October 2026

The project is an ongoing local-first fictional artist created by Babuni.dev, not a claim that its software wrote the source poems. Source poets/translators retain attribution. This audit indexes the artistic presentation, poem/pack provenance, scheduling rules and recorded failure-recovery history without modifying artwork or publishing behavior.

See design, content and provenance, architecture, decisions, development journal and AI collaboration. The README/changelog describe an operational 1.0.0 loop, while pyproject.toml currently declares package version 0.1.0; this discrepancy requires deliberate release-metadata review. No scheduled job or social/storefront publication was executed during this audit.

Document information

Author: Babuni.dev

View source