A human-and-agent creative studio: films, animation, graphics, sound—and reusable production methods.
Explore the studio · FW identity · How the site stays current
Plan a film, keep sources and decisions together, review exact versions and reuse what worked. Framewright is an experimental methods/skills library: the agent runs selected tools in your environment. It is not a hosted renderer or an automatic publishing service.
| Your task | Begin here | What you get |
|---|---|---|
| Start a film | Brief and production proposal | A concept, available sources/gaps, feasible route, constraints and a small pilot |
| Continue a project | Recover the current state | Exact selected version, accepted work, waits/blockers and the next permitted step |
| Correct a fragment | Bounded change and A/B review | A precise request, preserved original and corrected version, comparison in context |
Current priorities · Strategic roadmap and release gates
From a local checkout with Python3, without additional Python packages:
python3 skills/video-production-router/scripts/plan.py examples/briefs/mixed-promo.json --format markdownThis prints a plan; it does not render, install, call a model or spend money. Example output · Agent setup.
| Scope | Evidence and boundary |
|---|---|
| Source version | Experimental0.8.0 / REL-C; release notes and publication audit. Verify actual publication at the canonical release page. |
| REL-C evidence | Completed character fiction, bounded continuity/contact/correction, two aspect layouts and independent same-case recovery; assessment and review. Agent-and-human-led manual workflow; no general rig or autonomous route claim. |
| REL-B evidence | Assessment: both formats accepted, scoped recovery/reuse/correction; local checks/review and PR50/merged-main CI pass. No labour-savings or whole-route-autonomy claim. |
| SSI handoff | 6 November cycle, owner report: three films, deployment and six Telegram confirmations. Partner creative/listening acceptance, public publication and the15November unattended outcome remain separate. |
| AzurMenton guide pilot | 40s guide film accepted; YouTube and guide embed owner-reported. A general automated guide template is not yet qualified. |
Private projects/media remain outside Git; follow the external-storage and cleanup policy. Technical QA, creative/listening acceptance, rights and publication are separate records. Labour savings, whole-route autonomy and audience performance are not established by these films.
No Further Claims character-film method now records the completed131.25s manga film and native aspect-layout workflow. REL-C assessment records the passed same-case cold-resume check and explicit limitations. This methods release adds no new renderer, mandatory dependency or automatic publishing service.
The accepted FW identity and English public landing are now recorded. Sting/watermark use in films remains an optional experiment. The live catalogue counts distinguish methods, detailed tool profiles and review entries.
| Film | What it demonstrates |
|---|---|
| The Last Seed | Original Blender short with targeted contact, mechanism, rain and soundtrack corrections; case |
| Beyond the Letters — internal test | Reused sources/recipe/score, new story/voice, accepted16:9 and9:16; repeat-production evidence. No publication planned. |
| Menton: market to sea | Illustrated guide, owner footage, stock cutaways, route ending and expressive English narration; case |
| Framewright workflow lesson | Two-host dialogue and real production examples; case and limits |
Four training films → coherent fiction: accepted AI-explainer, animated-poster, liquid-photo and transport-guide cases; reusable choices, remaining limits and three bounded next-production questions. These production results inform REL-C; they do not establish blanket tool qualification.
Publication links and owner acceptance are owner-reported unless a case identifies separate verification. More experiments · Prior detailed overview, dated history.
Use the experiment workbench for one bounded question with pinned inputs, an existing-method comparison and separate output/review evidence. Queue. Free/open-source/macOS tools are preferred for this workspace; other platforms and paid tools stay reserves. Existing subscriptions require actual access/terms checks. Selection policy.
Revisions retain the interval, defect, desired change and protected properties. Templates separate design from copy/media/voice, but variants need their own render and review. Caption-time preflight flags overly short English copy; it never silently retimes accepted films.
Source-first editing · OTIO handoff/return · Saved-shot execution · External-project handoff · Production guide.
python3 scripts/check.pyOffline checks and temporary package smoke; no model calls, installs or release. Optional dependency skips are reported. Checking policy. MIT for original repository materials; tools/models/media have separate terms.
Book methods and cases · English Studio Notes.
The executor draws on explicit task/checkpoint ideas also used by SET and behavior-first checks in ABVX agent skills. These are references, not runtime dependencies. AGENTS.md Generator is an optional instruction-practice reference. ID and Pomerado remain conditional candidates.