Transform any Amplifier session into a polished, multi-page comic strip — with consistent characters, dramatic storytelling, and AmpliVerse publisher branding.
Sessions are thousands of lines of structured events. Useful for machines, invisible to humans. Nobody re-reads them.
Great work happens in sessions — debugging breakthroughs, architectural pivots, creative leaps — but there's no artifact that captures the story.
You can't email a JSONL file to your team. There's no shareable, self-contained way to show what you built — and how.
Your best sessions deserve more than a log file. They deserve a story — one with characters, drama, and a cover page.
A composable agent pipeline transforms raw session data into a self-contained HTML comic book.
Defines the visual identity — prompt template, color palette, panel conventions, and text treatment from 29 predefined styles or any custom description.
Analyzes the session, selects characters, and creates a multi-issue saga plan with per-panel dialogue, camera angles, and page layout structures.
Generates visual reference sheets for each character so they look consistent across every panel and page. Reuses characters across projects via embedding similarity.
Generate panel and cover images in parallel with vision-based self-review (up to 3 attempts per image). Cross-provider fallback across OpenAI and Gemini.
Assembles everything into a navigable multi-page HTML file with SVG clip-path panels, speech bubbles, and AmpliVerse publisher branding.
The session-to-comic recipe is a thin orchestrator calling three composable sub-recipes in sequence — each independently invocable.
Text-only · low cost. Research, style, storyboard, character roster. Ends with an approval gate.
Image generation · project-scoped. Reference sheets for each character, parallel execution.
Image generation · high cost. Panels, cover, composition — per-issue with quality review.
After saga-plan completes, review the character roster and narrative arc before committing to image generation costs.
issue-retry for surgical single-issue re-generation. issue-compose for text-only reassembly with zero image gen.
All assets tracked via comic:// URIs. Project-scoped for characters and styles; issue-scoped for panels and covers.
From Frank Miller noir to Studio Ghibli watercolors. Pick a predefined style or describe any aesthetic — the style-curator agent interprets it into a full visual guide.
The image generation bridge maintains a capability registry across OpenAI and Gemini. It picks the optimal model per task — and falls back across providers on moderation blocks.
If OpenAI blocks a prompt for content policy, the same prompt is automatically retried on Gemini. If all providers hit moderation, structured guidance is returned for the agent to rewrite the scene.
Gemini Embedding 2 computes multimodal embeddings for duplicate prevention, drift detection, and cross-project reuse.
Panel and cover artists use review_asset for vision-based QA — up to 3 regeneration attempts per image.
Each agent declares a model_role fallback chain. Mechanical steps use fast; creative steps use creative.
One sentence in, a multi-page comic out:
The output is a single self-contained HTML file: all images base64-embedded, SVG speech bubbles, keyboard/touch/click navigation, AmpliVerse branding. Open in any browser. No server needed.
Support for multi-issue story arcs with shared character rosters, per-issue evolution maps, and saga continuity tracking.
Sin City, Jujutsu Kaisen, Watchmen, Ghibli, Naruto, and a 3-issue Transformers saga — all shipped with exact generation prompts.
All images base64-embedded. Works offline, no external resources, no server. Email it, Slack it, open it anywhere.
Install the bundle, pick a style, point it at a session. Your comic book is one sentence away.
Primary source: colombod/amplifier-bundle-comic-strips repository on GitHub, cloned and analyzed locally.
git log, git shortlog -sn --all: 283 commits total; Diego Colombo 281, Brian Krabach 1, Marc Goodner 1; date range 2026-02-27 to 2026-05-20find + wc -l: ~28,560 lines Python total (~9,434 source, ~19,126 test); 193 files; 86 test files; 17 source modules; 6 agents; 8 recipes; 3 skills; 3 tool modules; 5 interactive modesteam_knowledge(operation="search"): 17 capability YAML files for the bundle covering all agents, recipes, modules, and the bundle itself; confirmed owner colombodmodel_role chains and inter-agent dependenciesStyle count (29) derived from the README style gallery table. Model count (12) from the model capability registry tables in README (5 OpenAI + 4 Gemini generateContent + 3 Imagen). No data was fabricated.
Presentation template: amplifier-bundle-gitea-story.html from the amplifier-stories repository, adapted with accent color #FF4081.