ComicFrame Studio

ComicFrame Studio is a local desktop app that turns ordinary video into aggressively stylized animation through an AUTOMATIC1111/Forge-compatible Stable Diffusion WebUI API.

Current product runtime: v3.5 · Simple Product Consolidation

The public workflow is intentionally small:

choose video
→ choose look
→ optional ControlNet structure rail
→ choose steps
→ PROCESS VIDEO
→ open result

Everything else—shot detection, reference locking, subject continuity, temporal transport, adaptive rendering, cache safety, VFR reconstruction, audio restoration, and backend compatibility—is engine-owned.

What v3.5 changes

v3.5 consolidates the product around one canonical shell instead of stacking a new UI class for every release.

Production now enters through:

app.py
  → comicframe_product.py
      → comicframe_simple.py
          → mature engine stack

comicframe_product.py owns only the public experience and render-policy bridge. comicframe_style_library.py owns public style registration. Engine modules stay independent.

The old interface → presence → aggro → style_overhaul application inheritance ladder is retired.

Public controls

Normal use exposes only:

• Look — searchable style/process library
• ControlNet — on/off structural rail
• Steps — 12–36, default 24
• Process Video

Aggressive redraw is the baseline policy. There is no AGGRO toggle.

The app remembers the last selected Look, ControlNet choice, and Steps between launches.

Functional behavior

The simple product shell includes:

• source/result preview
• searchable style browser
• automatic shot-aware sequence treatments
• live rendered-frame preview during processing
• elapsed-time processing heartbeat
• compact cancel action
• result actions for open, show in folder, save copy, and copy path
• primary action gating until a real source file exists
• non-destructive output naming beside the source video

The source video is never overwritten.

Production pipeline

source video
    ↓
source fingerprint + project ownership validation
    ↓
ffmpeg frame extraction + source timing capture
    ↓
shot analysis / treatment
    ↓
source frame
    + Shot Memory
    + reference / subject conditioning
    ↓
Stable Diffusion img2img
    + optional source Canny ControlNet
    ↓
deterministic style finishing
    ↓
optical-flow temporal stabilization
    ↓
validated styled PNG cache
    ↓
VFR-aware video reconstruction
    ↓
source audio restoration
    ↓
validated final MP4

ControlNet always uses the current source frame for geometry. Turning ControlNet off removes that structural unit; continuity systems remain engine-owned.

Quick start

1. Start Stable Diffusion WebUI

Use Forge or AUTOMATIC1111 with API access enabled.

For ControlNet mode, install sd-webui-controlnet and a compatible Canny model. ComicFrame capability-detects the local backend rather than assuming one fixed model name.

2. Launch ComicFrame

Windows:

run_comicframe_studio.bat

Or:

python app.py

Python dependencies are listed in requirements.txt. ffmpeg and ffprobe must be available on PATH.

3. Process

1. CHOOSE VIDEO
2. choose a Look
3. leave ControlNet on for a loose structural rail, or turn it off for freer redraw
4. choose Steps
5. PROCESS VIDEO
6. open the result

Project/cache storage is derived from the source automatically.

Style library

The public library combines shot-aware sequences with aggressive single-style passes. Current examples include:

• Graphic Shock
• Cyberpunk Print
• Toxic Xerox
• Photocopier Riot
• Newsprint Panic
• Bootleg Anime Print
• Street Poster Melt
• Chrome Nightmare
• Dead Channel
• Acid Cathedral
• Synthetic Fever
• Neon Ruin
• Memory Burn
• Paranoid Broadcast
• Heavy Gouache
• Ink Brutalism
• Pastel Nightmare
• Pulp Oil
• Storybook Ruin
• Charred Sketch
• Neo-Noir
• Manga Motion
• Risograph Zine
• VHS Horror
• Signal Rupture
• Watercolor Wash
• Clean Graphic Novel

The browser is searchable because the library is intentionally large.

Project layout

A typical generated project contains:

<source>_comicframe/
  .comicframe_project.json
  source_info.json
  comicframe_timeline.json
  comicframe_profile.json
  frames/
  styled_frames/
  subjects/
  shot_memory/
  previews/
  cache/
  styled_silent.mp4
  FINAL_STYLED.mp4

Generated state is contained inside the owned project directory.

Architecture

Product

• app.py — stable launcher
• comicframe_product.py — single public UI + public render-policy bridge
• comicframe_style_library.py — curated style registration and aggressive redraw baseline
• comicframe_simple.py — one-button video processing boundary over the engine

Engine

• comicframe_director.py — shot detection and treatments
• comicframe_reference_lock.py — shot-local reference conditioning
• comicframe_subjects.py — recurring subjects across cuts
• comicframe_shot_memory.py — pre-diffusion temporal/style memory
• comicframe_optical_flow.py — post-diffusion temporal transport
• comicframe_efficiency.py — adaptive render intelligence
• comicframe_autopilot.py — automatic orchestration
• comicframe_webui_contract.py — Forge/A1111 capability contract
• comicframe_media.py — project/media integrity
• comicframe_manifest_safety.py — persisted-path confinement
• comicframe_runtime_v28.py, comicframe_runtime_v29.py, comicframe_stability.py, comicframe_usability.py — hardened engine compatibility/audit boundaries

The versioned engine compatibility modules are intentionally retained because they carry audited media/resume behavior. Product-shell version layers are not.

CI

The repository uses one GitHub Actions workflow.

It:

• compiles the repository
• imports the canonical product and major engine modules
• runs the full pytest regression suite
• smoke-checks ControlNet payload compatibility
• verifies the public style registry

Workflow concurrency cancels superseded runs so a multi-file change does not create another queue storm.

Safety / resume guarantees

ComicFrame retains the hardened engine behavior from the v2.8–v2.9 audit series:

• full-file source fingerprints
• strict generated-project ownership
• corrupt cached PNG rejection
• symlink/path-confinement checks
• bounded transient WebUI retries
• adaptive OOM handling
• atomic validated final-media replacement
• source mutation checks
• VFR timing preservation
• selective compatible-frame reuse

See KNOWN_ISSUES.md for current non-blocking debt.

Engine documentation

• PROJECT_WORKSPACE.md
• AUTOPILOT.md
• DIRECTOR.md
• REFERENCE_LOCK.md
• SUBJECT_LIBRARY.md
• SHOT_MEMORY.md
• OPTICAL_FLOW.md
• EFFICIENCY.md
• CONTROLNET.md
• WEBUI_CONTRACT.md
• TROUBLESHOOTING.md
• AUDIT.md
• ROADMAP.md
• CHANGELOG.md