sextants 0.1.1__tar.gz

sextants-0.1.1/LICENSE

@@ -0,0 +1,30 @@
+MIT License
+
+sextants is a derivative work of "image2mode7" / "image2teletext" by
+Kieran Connell (https://github.com/kieranhj/image2mode7). The image-to-
+teletext conversion algorithm — including the per-row dynamic-programming
+solver, error model, and preprocessing pipeline — originates in that
+project and is used here under the terms of its MIT licence, reproduced
+below. Restructuring, packaging, and subsequent modifications are by the
+sextants authors.
+
+Copyright (c) 2020 Kieran Connell
+Copyright (c) 2026 Robert Smallshire
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

sextants-0.1.1/PKG-INFO

@@ -0,0 +1,109 @@
+Metadata-Version: 2.4
+Name: sextants
+Version: 0.1.1
+Summary: Convert raster images to teletext / viewdata block-mosaic graphics via an optimal per-row dynamic-programming solver
+Keywords: teletext,viewdata,videotex,mode7,saa5050,mosaic,sextants,bbc-micro
+Author: Robert Smallshire
+Author-email: Robert Smallshire <robert@smallshire.org.uk>
+License-Expression: MIT
+License-File: LICENSE
+Classifier: Development Status :: 3 - Alpha
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Multimedia :: Graphics
+Requires-Dist: numpy>=2.0
+Requires-Dist: pillow>=10.0
+Requires-Dist: click>=8.1
+Requires-Dist: tqdm>=4.66
+Requires-Dist: numba>=0.60 ; extra == 'fast'
+Requires-Python: >=3.11
+Project-URL: Homepage, https://github.com/rob-smallshire/sextants
+Provides-Extra: fast
+Description-Content-Type: text/markdown
+
+# sextants
+
+Convert raster images (JPEG, PNG, …) to **teletext / viewdata block-mosaic
+graphics** — the 40×25 character page rendered by the SAA5050 and its relatives
+(BBC Micro Mode 7, Ceefax/Oracle teletext, Prestel viewdata, CEPT videotex).
+
+The name comes from Unicode's own term for the 2×3 sub-pixel mosaic cell these
+displays draw — the *sextants* of the "Symbols for Legacy Computing" block
+(U+1FB00…). Each cell is six sub-pixels sharing a two-colour sub-palette, and a
+colour change is an in-band control code that costs a whole cell. Encoding an
+image well is therefore a constrained optimisation, not a simple resample —
+which is what this library does, row by row, with an optimal dynamic-programming
+solver.
+
+## Attribution
+
+**sextants is a derivative work of [image2mode7 / image2teletext] by
+Kieran Connell.** The conversion algorithm at its heart — the per-row
+dynamic-programming solver, the sub-pixel error model, and the preprocessing
+pipeline — is Kieran's work, used here under its MIT licence. This project
+repackages that code as an installable, `src`-layout Python library with a split
+module structure, a Click CLI, and a test suite, and is the basis for further
+development. Please see [`LICENSE`](LICENSE) for the full notice.
+
+[image2mode7 / image2teletext]: https://github.com/kieranhj/image2mode7
+
+## Install
+
+```sh
+uv add sextants           # or: pip install sextants
+```
+
+Optional numba acceleration for the DP solver:
+
+```sh
+uv add "sextants[fast]"
+```
+
+## Library usage
+
+```python
+from PIL import Image
+from sextants import convert, render_preview
+
+page = convert(Image.open("photo.jpg"), preset="photo")  # 1000-byte teletext page
+Path("photo.bin").write_bytes(page)                       # load at &7C00 on a BBC Micro
+
+render_preview(page).save("preview.png")                  # what it looks like on-screen
+```
+
+`convert(img, preset=None, **overrides)` is the preset-aware entry point;
+`convert_image(...)` exposes every knob directly. Presets bundle sensible
+preprocessing options for common sources:
+
+| Preset | For |
+| --- | --- |
+| `photo` | Portraits, landscapes, general photos |
+| `clean` | Safe universal default; light denoise + snap for unknown/BBS images |
+| `smooth` | Noisy JPEGs, soft gradients |
+| `vivid` / `graphic` / `flat` | Punchy colour, logos/cartoons, bold posterised |
+| `retro` / `art` | Authentic Ceefax look, hand-crafted teletext-art look |
+| `level1` | Source is already a Mode 7 / Level 1 teletext image |
+| `dark` / `tv` / `crt` | Exposure lift; LCD (PAR 1.2); CRT (PAR 1.22) |
+
+## Command line
+
+```sh
+sextants photo.jpg -o photo.bin --preset photo
+sextants photo.jpg --preview preview.png --url    # also print an edit.tf URL
+sextants --help
+```
+
+The output is a raw 1000-byte page (25 rows × 40 bytes of teletext character
+codes). It loads directly at `&7C00` on a BBC Micro, or paste the `--url` into
+[edit.tf](http://edit.tf).
+
+## Development
+
+```sh
+uv sync            # create the environment
+uv run pytest      # run the test suite
+```
+
+## Licence
+
+MIT — see [`LICENSE`](LICENSE). Original algorithm © 2020 Kieran Connell;
+packaging and modifications © 2026 Robert Smallshire.

sextants-0.1.1/README.md

@@ -0,0 +1,87 @@
+# sextants
+
+Convert raster images (JPEG, PNG, …) to **teletext / viewdata block-mosaic
+graphics** — the 40×25 character page rendered by the SAA5050 and its relatives
+(BBC Micro Mode 7, Ceefax/Oracle teletext, Prestel viewdata, CEPT videotex).
+
+The name comes from Unicode's own term for the 2×3 sub-pixel mosaic cell these
+displays draw — the *sextants* of the "Symbols for Legacy Computing" block
+(U+1FB00…). Each cell is six sub-pixels sharing a two-colour sub-palette, and a
+colour change is an in-band control code that costs a whole cell. Encoding an
+image well is therefore a constrained optimisation, not a simple resample —
+which is what this library does, row by row, with an optimal dynamic-programming
+solver.
+
+## Attribution
+
+**sextants is a derivative work of [image2mode7 / image2teletext] by
+Kieran Connell.** The conversion algorithm at its heart — the per-row
+dynamic-programming solver, the sub-pixel error model, and the preprocessing
+pipeline — is Kieran's work, used here under its MIT licence. This project
+repackages that code as an installable, `src`-layout Python library with a split
+module structure, a Click CLI, and a test suite, and is the basis for further
+development. Please see [`LICENSE`](LICENSE) for the full notice.
+
+[image2mode7 / image2teletext]: https://github.com/kieranhj/image2mode7
+
+## Install
+
+```sh
+uv add sextants           # or: pip install sextants
+```
+
+Optional numba acceleration for the DP solver:
+
+```sh
+uv add "sextants[fast]"
+```
+
+## Library usage
+
+```python
+from PIL import Image
+from sextants import convert, render_preview
+
+page = convert(Image.open("photo.jpg"), preset="photo")  # 1000-byte teletext page
+Path("photo.bin").write_bytes(page)                       # load at &7C00 on a BBC Micro
+
+render_preview(page).save("preview.png")                  # what it looks like on-screen
+```
+
+`convert(img, preset=None, **overrides)` is the preset-aware entry point;
+`convert_image(...)` exposes every knob directly. Presets bundle sensible
+preprocessing options for common sources:
+
+| Preset | For |
+| --- | --- |
+| `photo` | Portraits, landscapes, general photos |
+| `clean` | Safe universal default; light denoise + snap for unknown/BBS images |
+| `smooth` | Noisy JPEGs, soft gradients |
+| `vivid` / `graphic` / `flat` | Punchy colour, logos/cartoons, bold posterised |
+| `retro` / `art` | Authentic Ceefax look, hand-crafted teletext-art look |
+| `level1` | Source is already a Mode 7 / Level 1 teletext image |
+| `dark` / `tv` / `crt` | Exposure lift; LCD (PAR 1.2); CRT (PAR 1.22) |
+
+## Command line
+
+```sh
+sextants photo.jpg -o photo.bin --preset photo
+sextants photo.jpg --preview preview.png --url    # also print an edit.tf URL
+sextants --help
+```
+
+The output is a raw 1000-byte page (25 rows × 40 bytes of teletext character
+codes). It loads directly at `&7C00` on a BBC Micro, or paste the `--url` into
+[edit.tf](http://edit.tf).
+
+## Development
+
+```sh
+uv sync            # create the environment
+uv run pytest      # run the test suite
+```
+
+## Licence
+
+MIT — see [`LICENSE`](LICENSE). Original algorithm © 2020 Kieran Connell;
+packaging and modifications © 2026 Robert Smallshire.

sextants-0.1.1/pyproject.toml

@@ -0,0 +1,70 @@
+[project]
+name = "sextants"
+version = "0.1.1"
+description = "Convert raster images to teletext / viewdata block-mosaic graphics via an optimal per-row dynamic-programming solver"
+readme = "README.md"
+license = "MIT"
+license-files = ["LICENSE"]
+requires-python = ">=3.11"
+keywords = [
+    "teletext",
+    "viewdata",
+    "videotex",
+    "mode7",
+    "saa5050",
+    "mosaic",
+    "sextants",
+    "bbc-micro",
+]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "Programming Language :: Python :: 3",
+    "Topic :: Multimedia :: Graphics",
+]
+dependencies = [
+    "numpy>=2.0",
+    "pillow>=10.0",
+    "click>=8.1",
+    "tqdm>=4.66",
+]
+
+[[project.authors]]
+name = "Robert Smallshire"
+email = "robert@smallshire.org.uk"
+
+[project.optional-dependencies]
+fast = ["numba>=0.60"]
+
+[project.scripts]
+sextants = "sextants.cli:main"
+
+[project.urls]
+Homepage = "https://github.com/rob-smallshire/sextants"
+
+[build-system]
+requires = ["uv_build>=0.12.3,<0.13.0"]
+build-backend = "uv_build"
+
+[dependency-groups]
+dev = [
+    "pytest>=8.0",
+    "bump-my-version>=0.28",
+]
+
+[tool.bumpversion]
+current_version = "0.1.1"
+commit = true
+tag = true
+tag_name = "v{new_version}"
+message = "Bump sextants to {new_version}"
+
+[[tool.bumpversion.files]]
+filename = "pyproject.toml"
+regex = true
+search = '^version = "{current_version}"'
+replace = 'version = "{new_version}"'
+
+[[tool.bumpversion.files]]
+filename = "src/sextants/__init__.py"
+search = '__version__ = "{current_version}"'
+replace = '__version__ = "{new_version}"'

sextants-0.1.1/pyproject.toml.orig

@@ -0,0 +1,64 @@
+[project]
+name = "sextants"
+version = "0.1.1"
+description = "Convert raster images to teletext / viewdata block-mosaic graphics via an optimal per-row dynamic-programming solver"
+readme = "README.md"
+authors = [
+    { name = "Robert Smallshire", email = "robert@smallshire.org.uk" }
+]
+license = "MIT"
+license-files = ["LICENSE"]
+requires-python = ">=3.11"
+keywords = ["teletext", "viewdata", "videotex", "mode7", "saa5050", "mosaic", "sextants", "bbc-micro"]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "Programming Language :: Python :: 3",
+    "Topic :: Multimedia :: Graphics",
+]
+dependencies = [
+    "numpy>=2.0",
+    "pillow>=10.0",
+    "click>=8.1",
+    "tqdm>=4.66",
+]
+
+[project.optional-dependencies]
+fast = ["numba>=0.60"]
+
+[project.scripts]
+sextants = "sextants.cli:main"
+
+[project.urls]
+Homepage = "https://github.com/rob-smallshire/sextants"
+
+[build-system]
+requires = ["uv_build>=0.12.3,<0.13.0"]
+build-backend = "uv_build"
+
+[dependency-groups]
+dev = [
+    "pytest>=8.0",
+    "bump-my-version>=0.28",
+]
+
+#  Releasing: `uv run bump-my-version bump patch|minor|major` edits the version
+#  in both places it lives, commits and tags `v<version>`.
+[tool.bumpversion]
+current_version = "0.1.1"
+commit = true
+tag = true
+tag_name = "v{new_version}"
+message = "Bump sextants to {new_version}"
+
+[[tool.bumpversion.files]]
+filename = "pyproject.toml"
+#  Anchored to the line start so it matches the [project] version only, not the
+#  current_version key in this [tool.bumpversion] block (which bumps itself).
+regex = true
+search = '^version = "{current_version}"'
+replace = 'version = "{new_version}"'
+
+[[tool.bumpversion.files]]
+filename = "src/sextants/__init__.py"
+search = '__version__ = "{current_version}"'
+replace = '__version__ = "{new_version}"'

sextants-0.1.1/src/sextants/__init__.py

@@ -0,0 +1,32 @@
+"""sextants — convert raster images to teletext / viewdata block-mosaic graphics.
+
+The public API mirrors the pipeline: :func:`convert` (preset-aware, the usual
+entry point), :func:`convert_image` (full keyword control), :func:`preprocess_image`
+(the preprocessing stage alone), and :func:`render_preview` (render a solved page
+back to an image). ``PRESETS`` holds the named option bundles.
+
+This is a derivative of Kieran Connell's *image2mode7* / *image2teletext*
+(MIT). See the README and LICENSE for attribution.
+"""
+
+from .constants import COLOR_RGB, MODE7_WIDTH, MODE7_HEIGHT
+from .presets import PRESETS
+from .preprocess import preprocess_image
+from .convert import convert, convert_image, to_edittf_url, to_zxnet_url
+from .render import render_preview
+
+__version__ = "0.1.1"
+
+__all__ = [
+    "convert",
+    "convert_image",
+    "preprocess_image",
+    "render_preview",
+    "to_edittf_url",
+    "to_zxnet_url",
+    "PRESETS",
+    "COLOR_RGB",
+    "MODE7_WIDTH",
+    "MODE7_HEIGHT",
+    "__version__",
+]

sextants-0.1.1/src/sextants/cli.py

@@ -0,0 +1,130 @@
+"""Command-line interface for sextants.
+
+A thin Click wrapper over :func:`sextants.convert.convert_image`, writing the
+1000-byte page to a ``.bin`` and optionally a preview PNG / editor URL.
+"""
+
+import sys
+from pathlib import Path
+
+import click
+
+from .constants import MODE7_BLACK_BG
+from .convert import convert_image, to_edittf_url, to_zxnet_url
+from .presets import PRESETS
+from .render import render_preview
+
+
+@click.command(context_settings={"help_option_names": ["-h", "--help"]})
+@click.argument("input_image", type=click.Path(exists=True, dir_okay=False, path_type=Path))
+@click.option("-o", "--output", type=click.Path(dir_okay=False, path_type=Path),
+              help="Output .bin file (default: <input>.bin).")
+@click.option("--preview", "preview_path", type=click.Path(dir_okay=False, path_type=Path),
+              metavar="PNG", help="Save a preview PNG of the rendered output.")
+@click.option("--url", is_flag=True, help="Print an edit.tf URL for the output.")
+@click.option("--zxnet", is_flag=True, help="Print a ZXNet teletext editor URL for the output.")
+@click.option("--preset", type=click.Choice(sorted(PRESETS)),
+              help="Named bundle of options for common sources. Explicit flags override it.")
+@click.option("--nohold", is_flag=True, help="Disable Hold Graphics optimisation.")
+@click.option("--nofill", is_flag=True, help="Disable New Background optimisation.")
+@click.option("--sep", is_flag=True, help="Enable Separated Graphics mode (experimental).")
+@click.option("--greedy", is_flag=True,
+              help="Use the fast greedy solver + refinement instead of the full DP.")
+@click.option("--luma", is_flag=True,
+              help="Use perceptual luminance weighting (ITU-R BT.601) for the error metric.")
+@click.option("--linear", is_flag=True,
+              help="Linearise source pixels from sRGB to linear light before squaring error.")
+@click.option("--dither", is_flag=True,
+              help="Apply Floyd-Steinberg dithering at sub-pixel level after resize.")
+@click.option("--filter", "filter_", default="bilinear",
+              type=click.Choice(["bilinear", "lanczos", "bicubic", "nearest", "cimg"]),
+              help="Resampling filter for image resize (default: bilinear).")
+@click.option("--par", type=float, default=1.2, metavar="RATIO",
+              help="Pixel aspect ratio of target display (1.0 square, 1.2 LCD, 1.22 CRT).")
+@click.option("--sharpen-radius", type=float, default=1.0, metavar="R",
+              help="Unsharp-mask blur radius on the resized sub-pixel image (default: 1.0).")
+@click.option("--sharpen-amount", type=int, default=0, metavar="PCT",
+              help="Unsharp-mask strength percent (default: 0 = off).")
+@click.option("--sharpen-threshold", type=int, default=0, metavar="T",
+              help="Unsharp-mask threshold: min per-channel difference before sharpening.")
+@click.option("--gamma", type=float, default=1.0, metavar="G",
+              help="Power-law tone adjustment before resize (>1 brightens, <1 darkens).")
+@click.option("--contrast", type=float, default=1.0, metavar="C",
+              help="Contrast enhancement factor before resize (1.0 = off).")
+@click.option("--saturation", type=float, default=1.0, metavar="S",
+              help="Colour saturation factor before resize (1.0 = off).")
+@click.option("--posterize", type=int, default=0, metavar="BITS",
+              help="Posterise to BITS bits/channel before resize (0 = off, 1-7).")
+@click.option("--smooth", type=int, default=0, metavar="N",
+              help="Merge colour runs shorter than N cells after solving (0 = off).")
+@click.option("--snap", type=int, default=0, metavar="T",
+              help="Snap pixels within RGB distance T of a palette colour before dithering.")
+@click.option("--snap-palette", is_flag=True,
+              help="Snap every (quantised) colour region unconditionally to the nearest palette colour.")
+@click.option("--bg-flatten", type=int, default=0, metavar="T",
+              help="Flatten a border-connected background within distance T before resize (0 = off).")
+@click.option("--median", type=int, default=0, metavar="RADIUS",
+              help="Median filter of (2*RADIUS+1)^2 pixels before resize (0 = off).")
+@click.option("--quant", type=int, default=0, metavar="N",
+              help="Pre-quantise to N diverse colours after resize (0 = off).")
+@click.option("--direct-sample", is_flag=True,
+              help="Bypass resize: quantise at full resolution and point-sample sub-pixel centres.")
+@click.option("--edge-weight", type=float, default=1.0, metavar="W",
+              help="Silhouette-edge saliency weight (1.0 = off; try 2.0-5.0).")
+def main(input_image, output, preview_path, url, zxnet, preset,
+         nohold, nofill, sep, greedy, luma, linear, dither, filter_, par,
+         sharpen_radius, sharpen_amount, sharpen_threshold, gamma, contrast,
+         saturation, posterize, smooth, snap, snap_palette, bg_flatten,
+         median, quant, direct_sample, edge_weight):
+    """Convert INPUT_IMAGE to a 40x25 teletext page (1000-byte .bin)."""
+    # Start from the preset (if any), then let explicitly-passed flags win.
+    # Click can't tell a default from an explicit value, so we only override a
+    # preset key when the corresponding flag differs from its documented default.
+    kwargs = dict(PRESETS.get(preset, {}))
+
+    explicit = {
+        "use_hold": not nohold if nohold else None,
+        "use_fill": not nofill if nofill else None,
+        "use_sep": sep or None,
+        "greedy": greedy or None,
+        "luma": luma or None,
+        "linear": linear or None,
+        "dither": dither or None,
+        "filter": filter_ if filter_ != "bilinear" else None,
+        "par": par if par != 1.2 else None,
+        "sharpen_radius": sharpen_radius if sharpen_radius != 1.0 else None,
+        "sharpen_amount": sharpen_amount or None,
+        "sharpen_threshold": sharpen_threshold or None,
+        "gamma": gamma if gamma != 1.0 else None,
+        "contrast": contrast if contrast != 1.0 else None,
+        "saturation": saturation if saturation != 1.0 else None,
+        "posterize": posterize or None,
+        "smooth": smooth or None,
+        "snap": snap or None,
+        "snap_palette": snap_palette or None,
+        "bg_flatten": bg_flatten or None,
+        "median": median or None,
+        "quant_colors": quant or None,
+        "direct_sample": direct_sample or None,
+        "edge_weight": edge_weight if edge_weight != 1.0 else None,
+    }
+    kwargs.update({k: v for k, v in explicit.items() if v is not None})
+
+    output_path = output or input_image.with_suffix(".bin")
+    page = convert_image(input_image, **kwargs)
+
+    output_path.write_bytes(page)
+    click.echo(f"Written {len(page)} bytes -> {output_path}")
+
+    if preview_path:
+        render_preview(page).save(preview_path)
+        click.echo(f"Preview saved -> {preview_path}")
+    if url:
+        click.echo(to_edittf_url(page))
+    if zxnet:
+        click.echo(to_zxnet_url(page))
+
+
+if __name__ == "__main__":
+    sys.setrecursionlimit(10000)
+    main()

sextants-0.1.1/src/sextants/constants.py

@@ -0,0 +1,70 @@
+"""Teletext / Mode 7 constants: palette, control codes, sub-pixel geometry.
+
+These describe the SAA5050 display model shared by the whole teletext /
+viewdata / videotex family, not any single machine.
+"""
+
+import numpy as np
+
+
+MODE7_WIDTH  = 40
+MODE7_HEIGHT = 25
+MODE7_MAX_SIZE = MODE7_WIDTH * MODE7_HEIGHT
+
+# Maximum pixel dimensions of the image area
+MODE7_PIXEL_W = 78   # (MODE7_WIDTH - 1) * 2
+MODE7_PIXEL_H = 75   # MODE7_HEIGHT * 3
+
+# Column 0 of each row is used for the initial graphics colour code;
+# image content starts at column 1.
+FRAME_FIRST_COLUMN = 1
+
+# Control-code byte values (as stored in the raw .bin)
+MODE7_BLANK       = 32   # 0x20 - space / all-off graphics char
+MODE7_BLACK_BG    = 156  # 0x9C
+MODE7_NEW_BG      = 157  # 0x9D
+MODE7_HOLD_GFX    = 158  # 0x9E
+MODE7_RELEASE_GFX = 159  # 0x9F
+MODE7_GFX_COLOUR  = 144  # 0x90  (add 1-7 for colours red..white)
+MODE7_CONTIG_GFX  = 153  # 0x99
+MODE7_SEP_GFX     = 154  # 0x9A
+
+SEP_FG_FACTOR = 128  # blending factor for separated graphics (0-255)
+
+# Perceptual luminance weights for RGB error (ITU-R BT.601)
+# Human vision is ~6× more sensitive to green than blue; weighting the error
+# metric accordingly makes the DP prioritise brightness accuracy over hue.
+LUMA_WEIGHTS = np.array([0.299, 0.587, 0.114], dtype=np.float32)
+
+# Lookup table: sRGB byte value (0-255) → linearised value scaled back to 0-255.
+# The Teletext palette contains only 0 and 255, which map to 0 and 255 unchanged,
+# so only source pixel values need this correction.
+# Formula: c = v/255; linear = c/12.92 if c<=0.04045 else ((c+0.055)/1.055)^2.4
+_c = np.arange(256, dtype=np.float64) / 255.0
+_SRGB_LUT = np.where(_c <= 0.04045, _c / 12.92,
+                     ((_c + 0.055) / 1.055) ** 2.4).astype(np.float32) * 255.0
+del _c
+
+# State is a 15-bit integer: (sep:1)(last_gfx:7)(hold:1)(bg:3)(fg:3)
+MAX_STATE = 1 << 15  # 32768
+
+# Teletext colour palette: index → (R, G, B)
+COLOR_RGB = [
+    (  0,   0,   0),  # 0 black
+    (255,   0,   0),  # 1 red
+    (  0, 255,   0),  # 2 green
+    (255, 255,   0),  # 3 yellow
+    (  0,   0, 255),  # 4 blue
+    (255,   0, 255),  # 5 magenta
+    (  0, 255, 255),  # 6 cyan
+    (255, 255, 255),  # 7 white
+]
+
+# Bit-positions within a graphics character for the 6 sub-pixels:
+#   [TL TR]   bit 0 (1)   bit 1 (2)
+#   [ML MR]   bit 2 (4)   bit 3 (8)
+#   [BL BR]   bit 4 (16)  bit 6 (64)   ← bit 5 is always 1 (0x20 base)
+GFX_PIXEL_BITS = [1, 2, 4, 8, 16, 64]
+
+
+_PALETTE_NP = np.array(COLOR_RGB, dtype=np.float32)  # (8, 3)