ascii-art-hd

O que ascii-art-hd faz?

Detailed ASCII art comes from high resolution plus a real source, not from clever character choice. Recognisable subjects (a horse's mane, armour, fur) need roughly 160–260 columns, per-character colour and a real picture or mesh underneath. A 50×24 hand-assembled blob of ellipsoids will never read as an animal; do not try to fix that with more shading characters.

scripts/ascii_hd.py does the conversion. It needs only Python 3.9+, Pillow and NumPy (pip install pillow numpy).

Workflow

1. Find out what the source is. The script converts; it cannot invent a picture.
   • User supplied an image/GIF/frames/OBJ → use it.
   • No source → ask for one, or generate one (draw it with Pillow/NumPy, build a parametric mesh and write an .obj, or use an image-generation tool if the session has one). Say clearly which route you took.
   • Copyright: do not convert someone else's artwork into something the user will publish unless they own it or it is CC0/permissively licensed. For public repos prefer own/CC0 sources.
2. Pick the command.
   • image – one still picture.
   • frames – animated GIF or a folder of numbered frames.
   • obj – rotating turntable of a mesh (--up z for Blender exports, --pitch for camera tilt, --frames per full turn).
3. Start with a preview, not the final run.

   python scripts/ascii_hd.py image input.png --cols 200 --edges 0.35 --formats png,txt --out-dir out
   

   Then look at the PNG (view the image file). Do not judge the result from the text alone; text dumps lie about proportions and colour.
4. Tune, then re-render. See the table below. Change one thing at a time.
5. Deliver the formats the user needs and report what was produced, its size and the character count per frame.

Tuning guide

Symptom	Change
Subject mushy / details lost	raise --cols (200–260), add --edges 0.3–0.6, lower --gamma slightly
Too noisy / busy	lower --cols, set --edges 0, raise --quant
Dark subject disappears	lower --bg-threshold (e.g. 0.02), raise --min-brightness
Background shows up as characters	raise --bg-threshold (0.08–0.15)
Washed out	--black-point 5 --white-point 98, --gamma 1.1
Colours dull	--saturation 1.4
Looks stretched	--aspect must match the font (default 0.5 fits Consolas/Cascadia)
Dark text on light paper	--invert --bg "#ffffff"
Rotating mesh looks flat	--pitch 20–40, make sure normals are sane; model Z-up → --up z

Tone mapping uses measured ink density of the chosen font, so brightness steps stay even. Contrast levels are computed once across all frames, so animations do not flicker.

Output formats (--formats a,b,c)

• txt – plain text, frames separated by ===FRAME===.
• ansi – first frame with 24-bit terminal colour codes.
• html – self-contained page; animated if there are several frames.
• png – preview of one frame (--png-frame N).
• json – ASCII Mate .ascii.json (schema 2.0: frames, palette_map, color_frames). Colours are reduced to 62 palette entries, so photos lose some colour fidelity there; use html/png for the best look.

Using the result in ASCII Mate

Drop the .ascii.json onto a running widget, or copy it to assets/animations/community/. Keep it modest: 100–140 columns and ≤ 36 frames keeps the widget light. Larger "HD" assets need the HD mode of ASCII Mate (separate feature) to stay cheap on the CPU.

Limits to state honestly

• Quality is bounded by the source. A blurry or low-contrast image gives blurry ASCII.
• The OBJ rasteriser is pure NumPy: fine up to ~10k triangles; slower above 30k. It uses one flat colour (--mesh-color), no textures or MTL.
• No tool here judges aesthetics for you; always inspect the PNG before saying it looks good.