Whalechan Image Comic

Whale-chan Image Comic

Turn one recognizable fact into five concrete proposals, then generate the user's confirmed selection. Preserve the source's actual comic target, turn and tone. Strategic reinterpretation is one possible mechanism; language, reaction, timing, status exposure and bittersweet recognition can also carry the joke. Choice letters and image numbers are identifiers, never creative instructions.

Workflow: analyze → explore and review five proposals → Gate 1: user selects → Gate 2: user confirms proposal/image totals → expand, validate and freeze → generate, review and deliver. Stop and wait at both gates; selection alone never authorizes generation.

Load the guidance

Read these before every run:

• references/proposal-selection.md for the mandatory seven-column table, both user gates, reply handling and records.
• references/comedy-engine.md for source analysis, premise exploration, selection, and B/C intensity.
• references/character-personality.md for Whale-chan, form, outfit, and abstract-supporting-character locks.
• references/form-profiles.md for the five recommended presets, custom-ratio rules, measurement, and panel adaptation.
• references/panel-grammar.md for 1/2/4-panel structure and retry limits.
• references/expression-presets.json for per-panel visible facial acting and performance levels.
• references/text-style-routing.md to select one primary text treatment per image and review variety across the set.
• references/asset-index.md to select bundled references, references/form-authority.json for form targets, and references/asset-catalog.json for paths and hashes.
• references/provider-routing.md before any provider call.
• references/qa-rubric.md before reviewing any candidate.

Apply defaults and user overrides

• Offer exactly five proposals A–E. Default to five final PNGs per selected proposal; honor explicit per-proposal quantities, including one each. Default to automatic provider-native 1:1 output with 1024×1024 as a recommendation, not a required pixel size.
• Accept an explicit aspect ratio, an explicit pixel resolution, or both. Infer the ratio from an explicit resolution when omitted; reject conflicting values. Automatic mode validates the ratio, while explicit mode validates exact width and height.
• Explore distinct premises before allocating images. Keep only actual evaluations and comparisons, not a fixed eight-entry pool or prewritten winners.
• Make the five proposals materially different. Within each selected proposal, develop distinct executions without replacing its core scene, comic turn or user-locked wording. There is no rank allocation or B/C intensity quota.
• Choose 1, 2, or 4 panels by narrative need, not image index or a diversity quota.
• Select typography, framing, cast staging and text placement independently for each joke. Repetition is allowed when justified by the source; random rotation is not creative diversity. A consistent character identity does not require consistent layouts.
• For dialogue-driven material, default to bodily present abstract indigo partners. Record a narrative reason for an offscreen, avatar-only, or omitted role; preserve who speaks and who reacts.
• Default to semi-chibi, the canonical clean rounded cel-shaded style, the canonical navy-and-white maid outfit, and a pure white #FFFFFF background.
• Follow the input's main language; prefer Simplified Chinese for Chinese or mixed Chinese input.
• Treat the five bundled forms as recommended presets. Accept any finite measurable custom head ratio greater than 1.0; custom ratios have no bundled proportion reference.
• Allow any physically depictable action and any user-resolved background, art style, or outfit, subject to provider safety and technical limits. A custom field changes only that field.
• Default to sequential execution. Accept requested parallelism from 1 through 5 and record the lower effective runtime capacity when necessary.
• Start planning immediately, then always wait at Gate 1 and Gate 2. Never infer approval from silence, recommendations or a preselected answer. A planning-only request never authorizes image calls.
• Give each task an independent maximum of three image-producing calls. Never transfer unused calls.

Prepare proposals and obtain both user decisions

1. Record input.source_analysis: the source event, audience expectation, actual turn, comic target, tone, meaningful language differences and user corrections. Distinguish observed facts from your interpretation; preserve uncertainty instead of inventing a hidden meaning. Extract the fact anchor and source participants with evidence. Use input.participants: [] only for genuinely solo material. Images supply semantics unless the user requests their visuals; preserve speaker roles without copying real faces or avatars.
2. Explore premises from this analysis before deciding layouts. Record each premise's expectation, reversal, mechanism, personality, scene and actual gate reason. Preserve a reaction or language joke when that is the source's engine; do not automatically replace it with food, laziness, machinery or self-serving wordplay.
3. Reject flat retellings, generic reactions without a source-specific reveal, random metaphors, unclear anchors, decorative scenes and jokes that need explanation. A specific, timed reaction can itself reveal the contradiction.
4. Prepare five passing proposals with concrete scenes, turns, staging and key lines; review them against the source before showing them. Record actual comparisons only. Do not prefill PASS, rank by candidate number, or backfill a candidate pool from finished image plans.
5. Follow proposal-selection.md: render the seven-column Markdown table, ask for selection below it and explain the generation strategy. Stop for Gate 1. Record the user's chosen proposals, quantities and changes.
6. Show every selected proposal's title and image count, plus proposal and image totals. Include explicit changes. Ask whether to begin and stop for Gate 2. A changed selection or quantity requires an updated summary and fresh confirmation. If proposals are redesigned, return to Gate 1.

Expand and freeze the confirmed assignment

After Gate 2, continue through internal preparation and generation without a third approval:

1. Expand exactly the confirmed quantities. Set ranked_ideas to the selected idea ids and explain the selection in selection_reason. Link each image by idea_id and describe its distinct payoff in execution_note. Pose, font and background swaps alone are insufficient.

2. Resolve each field from explicit user instructions, declared reference roles, source semantics, then defaults. Freeze composition, proportion_check, output, style, costume, background, exact core_text, text_style and reason, dialogue_plan, cast_plan, proportion, action_plan and expressions. Assign image numbers last. Intentional close-ups use visible-only; measurable full-body shots use measured.

3. Use a canonical identity reference first and the selected bundled text-style reference as the sole typography reference. Declare each reference's roles: identity, style, pose_action, composition, costume, background, typography, or proportion. Load the abstract-user pose sheet only when a supporting character appears. Use at most five effective references; never silently drop required typography.

4. Review each execution against its approved proposal and changes, then check typography, speaker ownership and cast staging against the source. Reject random template rotation, generic reasons and offscreen choices made just to simplify drawing. Fix internal execution details without changing the approved core. Write schema v10 assignment.json including the proposal and both user decisions, then validate and initialize; initialization writes creative-record.md:

   python3 scripts/manage-run.py validate-assignment --assignment <assignment.json>
   python3 scripts/manage-run.py init --assignment <assignment.json> \
     --effective-parallelism <current-capacity>

For multiple inputs, finish all assignments, then run python3 scripts/manage-run.py validate-batch --assignment <first.json> --assignment <second.json> with every assignment before the first generation call. Review its case-by-position matrix and warnings, not just aggregate counts. Compare mechanisms, narrative beats, shots, cast positions and text placement across cases and across positions; changing order must not hide a repeated skeleton. Ask whether another case's dialogue could replace this one's without changing the drawing. Save a batch-review.md beside the assignments naming the compared cases, warning dispositions, source-specific reasons for retained similarities, and revisions. Do not generate until this review is actually performed. Structural validity is not creative approval; no warnings is not approval either. Repeat this comparison on final images. See references/batch-review.md.

Use the current schema in references/run-schema.md. All assignment, generation and QA operations require that contract. The initialized run belongs under artifacts/whalechan-image-comic/<run-name>/ unless the user gives another destination. External references are frozen into the run with hashes.