Low-fi sketch wireframes
This skill produces mobile screens that look drawn on paper or in FigJam. They exist to explore structure and flow, not to present a finished UI. Everything is drawn in code (Python + Pillow): no HTML screenshots and no external fonts.
Files
| File | Purpose |
|---|---|
scripts/lowfi_kit.py | Main library. Screen lays a screen out top to bottom with a cursor (no coordinates). render_project() writes the PNGs, the comparison sheets and runs the checks. |
scripts/template_project.py | Starting point. Fictional example with 2 proposals × 3 screens. Copy it and replace the screens. |
scripts/sketch.py | Drawing base: wobbly lines, rectangles, circles, crossed boxes, icons. |
scripts/hand.py | Uppercase hand lettering drawn stroke by stroke, hatching (hatch) and hatched placeholder bars (bar). |
examples/template_output.png | Output of the template, as a visual reference. |
Requirement: Python 3 + Pillow (pip install pillow).
cp scripts/template_project.py my_flow.py # edit the screens
OUT=./MyProject_Wireframes_v1 python3 my_flow.py
(Keep my_flow.py next to lowfi_kit.py, sketch.py and hand.py, or add scripts/ to sys.path.)
1. Workflow
- Understand the request: product, user, flow, number of screens and device size (default 375 × 812). Ask how many proposals to create before drawing (see «Number of proposals» below). Default: 2.
- Write the minimum content of each screen before drawing: title, actions, sample data and states. Use the same data in every proposal.
- Define the structural difference between proposals (section 4) before composing.
- Compose each screen as a function
def screen(s): ...using theScreenmethods. - Run
render_project({...}, out, "Title"). - Validate (section 5) and fix.
- Deliver in a versioned folder (
<Project>_Wireframes_vN/) without overwriting earlier versions. If asked, add aUX_ANALYSIS.md(section 6).
Number of proposals (N)
The number of proposals is a variable N ≥ 1, with a default of 2.
- Always ask before composing, unless the user already gave a number. Use a multiple-choice question when available, for example: «How many alternative proposals do you want for this flow?» →
2 (Recommended)·1·3· Other. - If nobody answers (unattended run), use 2 and say at the top of the delivery that you assumed the default.
- N = 1: skip the comparison rules in section 4; only
Proposal_1_Full.pngis produced. - N ≥ 3: every proposal needs its own dominant structure. The abstraction test applies to each pair of proposals, not only to the first two. If you cannot make a proposal genuinely different, say so and suggest a smaller N instead of producing near-duplicates.
render_projectaccepts any number of entries in its dictionary. It writes oneProposal_<i>_Full.pngper proposal and stacks all of them inAll_Proposals.png.
2. Screen API (lowfi_kit)
Everything stacks vertically from y≈98. A leading "*" marks an option as selected.
| Area | Methods |
|---|---|
| Header | header(title, left='back'│'x'│'menu'│text, right=text, chip=text, link=text) |
| Text | title(s) · text(s, cap, gray, bold) · label(s) · hint(s) · link(s) · lines(n) · placeholder_bar(frac) |
| Media | image(h) (crossed box) · avatar(cx, cy, initials) |
| Inputs | field(label, value│[lines], multiline) · select(label, value) · search(text) · radios([...]) · checkboxes([...]) · toggle(label, on) · chips([...]) · segmented([...], active) · tabs([...], active) · stepper(label, value, unit) |
| Progress | progress(n, total, name) (step bar) · steps([names], current) (1 → 2 → 3) |
| Content | list_item(title, sub, avatar, right, chevron, selected) · card_grid([labels], cols) · kv([(key, value)]) · row_edit(label) · check_line(s) · info(s, sub) |
| Grouping | begin_box(title, kind='normal'│'strong'│'dashed', state='open') … end_box() · accordion(title, 'done'│'closed', right) |
| Validation | alert(title, sub, chips) (dashed box with «!») · note([lines]) (design annotation, not UI) |
| Buttons | button(label, 'pri'│'sec'│'off') · buttons([(label, kind), ...]) |
| Bottom area | footer([(label, kind), ...]) (pinned CTAs) · nav([(icon, label)], active) · fab(label, sub, hint) |
| Variants | sheet(title, dim_title) (bottom sheet over a dimmed background) |
- Available icons:
home search list chat user bell plus back x chev down. - For anything not covered, draw directly on
s.d(Sketch) and withhand(s.d, x, y, text, cap). Screen(w=390, h=844)supports other sizes;render_project(..., w=390, h=844)passes them through.
3. Visual rules
- Black on white only. Gray is for secondary text and notes. No colors, gradients or shadows.
- Imperfect strokes everywhere: gentle wobble, overshooting line ends, overlapping closures.
- Uppercase hand lettering inside screens. Cap sizes: titles 12–15, body 8–10, notes 7, tab bar 6 minimum.
lowfi_kitshrinks text automatically when it does not fit. - No solid-filled buttons: primary = double outline; secondary = single outline; disabled = gray dashed.
- Selection = hatching (bottom stripe or fill), never a black block.
- Schematic filler: hatched bars for names and secondary titles, lines for paragraphs, crossed boxes for images.
- Minimal text: only what is needed to understand each decision.
- Nothing decorative: every element must communicate a function.
- Balanced negative space: if more than a third of the screen is empty, move the main action up or tighten the layout. Never fill it with decoration.
4. UX rules for comparing proposals
- Same scenario: same user, same data, same features, same number of screens, same end result. Only the information architecture, navigation, hierarchy and arrangement of actions change.
- Abstraction test: reduced to boxes and arrows, every proposal must still look different from every other one. Each needs its own dominant shape, for example:
- guided: a column of steps, progress bar, one CTA per screen, progress blocked while a required field is missing;
- modular: tabs or a card grid, collapsible sections in any order, review with in-place fixes, undo;
- other options: search-first, conversational, card-based, single long page.
- One state per screen. Confirmations, later states and rules are documented with
note()or in a clearly separate extra screen. - Actions that look alike must be told apart with microcopy (for example, "the category is preselected" vs "no category").
- Error prevention and feedback stay visible: required fields with
*, warnings withalert(), disabled buttons with'off'. - Do not declare a winning proposal without evidence from user testing.
5. Validation before delivery
render_project already warns about three things:
⚠ … content reaches y=…: the screen overflows. Remove content or reduce heights.⚠ Characters without a glyph: some characters were drawn as blanks. Replace them or add them tohand.py(section 7).- If any single screen is not exactly
w × h, the script stops on anassert.
Also check the Comparison/All_Proposals.png sheet by eye:
- no overlapping text and nothing outside the frame;
- visual consistency across all screens;
- the proposals are genuinely different (abstraction test).
6. Delivery structure
<Project>_Wireframes_vN/
├── Proposal_1_<Name>/01_<Screen>.png …
├── Proposal_2_<Name>/…
├── … (one folder per proposal, N in total)
├── Comparison/Proposal_1_Full.png … Proposal_N_Full.png · All_Proposals.png
└── UX_ANALYSIS.md (if requested)
The comparison sheets paste screens unscaled, aligned horizontally, 40 px apart.
Suggested UX_ANALYSIS.md:
- Problems in the previous version.
- Changes and their UX rationale (cite Nielsen's heuristics where relevant).
- Table of differences between proposals.
- Strengths and limitations of each flow.
- Questions to validate with users.
7. Extending
- New character: add it to
Ginhand.pyas'X': (width, [[(x, y), …], …]). The grid is 6 units tall, with y = 0 at the top. It already includes A–Z, Ñ, Ü, accented vowels, 0–9 and+ - = % # $ & @ ' " , . : · ! ? ¿ ( ) [ ] / * < > → ← —. - New icon: add it to
Sketch.icon()using lines or circles relative to(cx, cy)and sizes. - New component: add a method to
Screenthat draws fromself.yand advancesself.ywhen done. - More or less wobble: tune
ampin_wobble(boxes ≈ 0.9, letters ≈ 0.25).