ToBeWin/mg-motion-production

MG: From Product Proposition to Reviewable Videos

The public version retains general methods, components, and case source code; historical media and large native projects are not distributed with the repository. Old cases requiring these assets cannot be directly rebuilt; licensed assets and sound are provided for new projects.

The core task is to make the audience understand and remember a supported proposition. Composition, materials, typography, framing, motion, and sound all serve it. Adding effects, using AE-like software, or successful export does not independently prove sophistication.

Free exploration is also a formal production mode. When the user has no preset purpose and requests imaginative output, start from a form, material, space, or rhythm proposition, design a mother form capable of continuous change, create a complete audiovisual work, and let the user judge the application. No need to pre-select a scene, product, or style number; existing recipes are references, not creative boundaries. Specific methods and this round's FORM project are in Free Exploration & Design Assets.

Treat design assets as an independent production stage. Establish key frames and reusable components in available design tools like FigCraft as needed, complete paths and vector shapes in VectorCraft, process layered bitmaps, textures, and lighting in PhotoCraft; iterate alternately with motion previz, then pass to Blender / EffectCraft. First confirm the actual appearance and editable layers of assets in the target tool; do not equate successful import with lossless transfer. Read Design-to-Tool Handoff. Prioritize the validated local design chain locally; Figma and Pen.dev can be used per project, but their accounts, quotas, or connectivity should not block design asset creation. SoundCraft can handle stems and mixing; DesignCraft / DeckCraft can produce reports and presentations within the video.

When a software video lacks design sense, first complement asset outcomes and component states, then increase motion complexity. Read Design-Driven Software Launch Video: extract action-result relationships from reference videos, independently design multiple layouts for the same content, and maintain continuity of object identity, fonts, and anchor points during transformation. FOLD is an original concept case and should not be treated as a real AI functionality demo or a fixed template for all projects.

The user has approved retaining FOLD as a style: FOLD | Design Unfolding, ID fold-design. Organize the creative process with original works, bright environments, component states, and continuous layout changes; suitable for design software, creative AI, content tools, and multi-size brand visuals. Call scripts/choose_style.py --show fold-design, or scripts/plan_scene.py --scenario software-launch --style fold-design --out /absolute/new-project/plan. Color, mother form, content, and duration are designed per project, not fixed to petals, paper colors, or 32 seconds.

New user-approved independent style: Bold Brand Pop, ID bold-brand-pop. Start from original packaging, wordmarks, and graphic systems, organizing oversized typography, strong color fields, physical products or illustrations, continuous mother forms, and music into brand advertisements. Call scripts/choose_style.py --show bold-brand-pop, or use --style bold-brand-pop with the planner. SUNPOP's beverage can, sun, orange-blue color scheme, and 24-second duration are only examples; redesign per industry, brand personality, and assets.

New user-requested independent style: Sculptural Type + Dynamic Typography, ID sculptural-type. Continuously transform the same letterform across 3D solids, 2D text carriers, and negative space; suitable for brand refresh, creative tools, tech hardware, and keynote visuals. Call scripts/choose_style.py --show sculptural-type, or use --style sculptural-type with the planner. The guide includes shared paths, 3D/2D handoff, and music selection based on brand personality; AUREL's letter A, color scheme, 16 seconds, and piano are only examples.

Previous feedback: TRACE v10 continuous motion segment received "this effect is decent" evaluation, and a request to expand to other scenes and styles. Retain this directional approval, select new motion logic per Motion-First Narrative and Cross-Scene Continuous Motion; branding, hardware, and collaboration cannot just swap colors and titles of the same UI. The old nine scenarios' use case classification and unapproved scope remain preserved.

Select light/dark, color, and environment based on brand, scene, and audience; simultaneously check text, foreground, and caption contrast.

Captions and overall art must be designed together. Caption colors must coordinate with background and subject; determine five color roles (environment, subject, emphasis, secondary info, captions) for each route and validate in real shots. Do not treat desaturation or gray-white gradients as quality guarantees. See Color, Environment, and Captions.

Current production focus: concentrate on tech, first deepen software/AI, agents, developer tools, and data products, then expand to multimodal, knowledge assistants, cloud services, and tech hardware as needed. Read Tech Direction & Production Priority to select entry points; 24 scenes remain as a usage directory, non-tech topics used on demand. Increasing recipe or sample count does not replace core work refinement.

When commissioned for tech product promotion, software product launch is the primary design direction. Maintain product content credibility, while clarifying whether the request is for an operation demo or an MG launch video. MG launch organizes tasks and results through graphics, typography, shape changes, and editing; a full workspace establishes location and proves functionality, but the whole video must not remain on the same interface. Select visible inputs, key actions, results, and sources, then design how objects transform continuously. Software, brand, and hardware each have multiple routes; not every time can just swap colors and titles. Read Style Routes for Three Domains, use scripts/choose_style.py --domain software to select; track completion levels separately for three main videos and six 8-second motion studies.

For real product promotion, first collect product evidence. When the user provides local products and authoritative business videos, first verify real tasks, visible states, and actual results; self-designed cards cannot replace product pages. When the presentation targets a broad audience, organize real interfaces with a few value propositions, preserving MG's scale changes and object handoffs. See Public Value & Real Product Footage: distinguish between captured pages, completed task replays, continuous screen recordings, and conceptual renderings; maintain correspondence between files, data, and tasks.

When producing multiple style versions of the same real product, first lock in business facts, original requests, results, and narration, then design space and attention separately. luminous-workspace uses persistent task anchors and window yielding to organize results; companion-ui-explainer can use a continuous pointer guide per brand requirements, without adding cartoon characters. Both share evidence, not the same set of actions; specific adaptations and boundaries are documented in their respective guides.

For sound-enabled promotional videos, prioritize using the user-specified or configured voiceover and music; if not specified, read Voiceover & Music Options. First check installed corresponding skills and models; do not quietly replace final music with simple synthesized melodies. When voiceover is present, provide synced subtitles, SRT, and non-burned-in subtitle versions per language and usage; avoid button-style rounded caption backplates, give subtitles space that doesn't obscure the subject; silent loops, UI feedback, etc., are still decided per use. Specific workflow see Voiceover, Music, and Captions. Full listening review, no-script transcription, sound-image cross-check, and revision rules see Full-Length Sound Review.

Curated Key Styles

• Graphic Component Product Pop: design components, dynamic typography, file and result handoffs.
• Luminous Workspace: persistent task anchors, window yielding, local operations, and results.
• Character/Pointer Guide: positioning, operation, feedback, reading, and review.
• Editorial Product Narrative: organize product value with big type, interface close-ups, and continuous editing.
• Evidence-Led Showcase: task proof, capability tour, results collection, and brand closure.
• Kinetic Process: continuously transform goals into structures, stages, and results.

These methods are redesigned per new project brand, assets, and rhythm; do not treat other products' screen recordings or sounds as generic assets.

First Distinguish Four Dimensions

Application Scene determines what to prove and how to conclude; Visual Style determines typography, colors, materials, and frame density; Space/Motion Grammar determines how attention moves; Delivery Medium determines aspect ratio, viewing distance, sound conditions, and playback method. Do not treat "geometric" and "product UI" as an exhaustive binary choice for all scenes.

When purpose is clear, read 24 Scenarios & Narrative Recipes, select the route corresponding to the usage, then use scripts/plan_scene.py --show <id> as needed to read asset lists, shot chains, and checklists. Recipes are not completed videos; produced and approved scope is in Capability Evidence. Prioritize solid software and AI launches: provide inputs, key actions, visible results, and value evidence. Brand ads, keynote openers, social media ads, website loops, data stories, and physical product displays have different acceptance criteria; cannot just swap copy on the same timeline.

Only read Visual Routes & Assets when style needs decisions; read Space & Motion Grammar when involving framing, depth, or continuous transitions.

Use the planner to generate storyboards, asset lists, and sound markers for the project, continue actual production:

python3 scripts/plan_scene.py --list
python3 scripts/plan_scene.py --scenario developer-api --title "This Launch" \
  --duration 25 --style technical --out /absolute/new-project/plan

The generated brief.md, production-plan.json, and cues.json are all editable; when user has authorized, proceed directly to key frames and animation after planning, without adding approval gates. Fixed duration is distributed in whole frames; if reading time is too short, prompt rearrangement, do not use speed-up instead of cuts.

Extract Methods from Reference Videos

Evident Showcase Depth for this route supplements object continuity, reading rhythm, relationship of the last result back to the matrix, and the norm of shared identity across task/result/capability components. Research on evidence and native component implementation, sound listening review, and new video acceptance remain separately documented. Executable components, 6/8 result layout, real task migration, and corrected issues are in Native Production & Reuse; entry point is assets/evidence-led-showcase/rebuild.py, only copy licensed assets referenced by the configuration.

When the user provides references, actually watch the video. If none specified, select 2–4 official cases close to the task, covering different solutions; separate "recent launches" from "most worth learning"; do not present old videos as latest. Research method see Launch Video Research.

Record each segment as: timecode/version → visible change → attention in frame → before/after state → spatial relationship → sound range (if actually heard) → transferable use. Distinguish between frame facts, visual inferences, and unverified content. Sparse frame sampling supports composition and sequence relationships but cannot prove precise easing, motion blur parameters, or sound/picture sync points.

Do not infer from style that the author must have used AE. Learn object identity, spatial continuity, and information organization; do not treat brand logos, soundtracks, or ready-made shots as your own assets.

Production Workflow per Project Scale

1. Clarify proposition and evidence. Write one sentence the audience should ultimately understand about the value, then list one piece of visible result. Mark unimplemented or unsubstantiated content as conceptual. Identify usage, audience, duration, aspect ratio, silent/sound, delivery project; do not re-ask what can be inferred from existing context.
2. Complete visual direction and design assets. Provide key frames, font family and actual glyphs, color roles, materials, camera distance, frame density; produce splittable paths, icons, illustrations, glyphs, product components, or material sources; retain editable files, object names, and control points needed for deformation. For high-requirement projects, first check if key frames and core assets can support the goal, iterating with motion previz. When user has authorized, proceed directly.
3. Establish narrative and space. Shot duration is determined by action and reading. Mark subject, camera, occlusion, world coordinates and screen coordinates, inherited state, and carried items during transitions. The more specific the proposition, the more avoidable are meaningless floating cards, node spheres, and full-screen color changes.
4. Create continuous motion preview. Iterate with key frame design; first complete the hardest connections and information chains; do not polish all static frames before adding movement. Check if pauses have purpose, if camera movement reveals information, and if the subject of the previous shot has a clear destination in the next. Refer to Production Depth & Revision.
5. Complete visuals and sound. Select assets/tools matching the direction; arrange sound per common markers. Use Timing & Audio; use Tooling & Engineering for engine, fonts, rendering, and native delivery.
6. Review, revise, deliver. Watch actual MP4, dense motion segments, reading frames, and ending; separate full playback/review from structural check. When user criticizes quality, return to the corresponding level for revision; do not just increase sharpness, resolution, or effect count. See Quality & Delivery.

Refine Existing Works

Retain styles, product identities, and content still approved by the user; save old versions separately. When user rejects current quality, "retain style" does not mean retain rough assets, incorrect glyphs, average rhythm, or content-less UI. First list specific gaps: unclear proposition, insufficient product evidence, material issues, font/layout, spatial inconsistency, too-short reading, motion losing focus, sound-picture mismatch. Each change addresses a gap and is rechecked.

For details on existing directions, read Brand Geometry/Dynamic Type, Product UI, or Extended Style Adaptation. These are local production references, not fixed video structures.

When software feature launches need further polish quality, read Continuous Production of Assets, Operations, and Results. Brand Color Sync v3 provides a complete natively editable layout, operation close-ups, state propagation, independent horizontal/vertical layout, and original music project. First design the demonstrated results, then arrange motion effects; preserving object identity and input traces is more useful than swapping in more effects.

Specific production and revision for the three main lines (software launch, precision hardware, brand typography) are in Production Records for Three Flagship Directions: TRACE v2 workspace and citation backtracking, ORBIT v4 bright metallic photography, SHIFT v2 glyph and mother-form transitions. Maintain credibility of normal software interfaces; native compositing, 3D photography, and sound tools each handle their appropriate work.

When agent collaboration or data analysis videos need to strengthen result credibility, read Two Evidence Chains from Task to Result. Yinian Collaboration v3 includes browseable web pages and actual checks; DataTalk v3 uses the same CSV to drive graphics and downloadable briefs. Learn object continuity, same-scale comparison, result delivery, and shared sound markers; check if the space curve of pause segments causes charts to drift.

For knowledge assistant, research Q&A, or sourced AI feature videos, read Continuous Presentation of Sources, Answers, and Citations. TRACE provides continuous excerpt objects, citations returning to the same original text, bright document space, accompanying answer HTML/JSON, and rebuildable project; separate actual retrieval/inference from conceptual illustration.

When backgrounds look cheap, mismatch the subject, or lack texture, read Background, Lighting, and Subject Layering. First adjust color roles, main light direction, card edges, and shadows, then decide if texture is needed; use independent pre-compositions to protect text and data encoding. Yinian Collaboration / DataTalk v3.1 provide two rebuildable cases: neutral graphite and warm/cool white.

When functional videos are tidy but still lack persuasiveness, read Making Key Shots Carry Visible Proof: before/after comparison of the same object, visible generation and object inspection, and countable ratios—three methods addressing different evidence needs, with rebuildable v4 cases. Do not solidify the three methods as mandatory structure for every scene.

Cross-Scene Reuse

22 Style Routes describe differences in assets, motion, and review; 18 Motion Modules describe when to use mask reveals, match cuts, data transformations, task handoffs, disassembly, tracking, etc. Combine methods per communication task, not by stacking effect names.

Re-arrange and validate horizontal/vertical screens, public displays, transparent overlays, silent loops, and runtime animations per Frame & Delivery Adaptations. MP4 mockups are not delivered interactive components.

6 Native Components provide executable JSX and editable EffectCraft projects: focus highlight, task handoff, same-scale comparison, path drawing, name/topic bar, countdown. They are motion building blocks; actual projects still need their own visuals and assets.

When developer launches, brand ads, or service explanations require standalone video examples, read API, Brand Manifesto, and Service Ads: fields spanning from input to task, mother form extended to occlusion and in-letter patterns, exception loops for the same request. Includes three bright-direction, original sound, rebuildable native projects.

Thirteen new use cases have independent executable examples; see 13 Scenario Implementations & Boundaries: onboarding, case study, logo, hardware function and disassembly, map, editorial timeline, keynote, live stream, exhibition hall, UI feedback, live-action tracking, and opening titles. Includes real native tracking, transparent assets, split models, and runnable state logic; read corresponding projects per scenario; do not treat use case count as equal commercial maturity.

Native EffectCraft Production

Four software styles share enhancements but are individually orchestrated. Graphic Component Pop, Luminous Workspace, Character/Pointer Guide, and Evidence-Led Showcase all read Input, Guide, Mouse, and Cross-Style MG Enhancements. Input states, original glyphs, hotspot clicks, reading guides, typography, masks, page space, results collections, and sound markers are governed by common specifications, then arranged per style. The selector and planner have integrated assets/native-interaction/style-profiles.json; this round's rules being synchronized does not imply automatic re-render of historical videos.

Full video motion design cannot be reduced to interaction guides. Dynamic typography, form succession, foreground occlusion, native 3D pages, camera and staggered arrays, results aggregation, and brand closure should also be arranged per subject. See MG Motion Methods for Software Launch Videos; implemented scope and research-pending capabilities are separately retained.

Software interactions must have visible cause and effect. Several routes have been criticized by users for missing mouse clicks, insufficient detail, and inadequate MG guidance. First read Native Interaction & Attention Guidance: real input fields and original glyphs, focus/insertion cursor/submit, real entry points and after-states, hotspot coordinates, pointer arrival/press, file mask expansion, fixed viewport scrolling, same-source local zoom, reading strokes and sound linking. EffectCraft native components are in assets/native-interaction/; Evidence-Led, Graphic Component, Luminous Workspace, and Pointer Guide each organize these actions. Preserve old versions; distinguish real captured state reconstruction from continuous screen recording; do not refer to unreviewed effects in new samples as "flagship level."

When EffectCraft is available locally, prioritize its validated native capabilities for path trimming, repeaters, track mattes, expression linking, pre-compositions, 3D cameras, materials, and rendering. Read Native Capabilities & Rebuildable Examples, query actual properties and create mini samples as needed; do not default to treating it only as a video exporter, nor stack unrelated effects just to showcase the tool.

When transparent glass, precision metal, or macro product photography is needed, read Optical & Material Cinematography. First check solid thickness, normals, light reflections, and continuous shots, then select native 3D or Blender Cycles plate plus EffectCraft compositing. Using more native effects is not the quality goal for such visuals.

Existing seven-scenario research records are stored in Seven Scenario Cases; new recipes continue to expand applications. Form/ORBIT v3 glass and metal rework received positive user feedback; retain this route; this does not mean all 24 scenarios and 22 styles have passed equivalent quality review.

Practical Resources

• assets/design-exploration/FORM/: Original "Form Formation," including Figma design link, VectorCraft / PhotoCraft / Blender / EffectCraft source files, 18-second three versions, original sound and stems, FilmCraft / FCPXML handoff, and actual checks. Used for free exploration and design asset handoff, not as a mother form or use-case limitation for all new videos.

• assets/design-exploration/FOLD/: fold-design complete case with positive user style feedback, including components and event visual design, topic and layout variations for the same object, 32-second three versions and native project. Migrate per Design Unfolding; latest feedback in case.json, historical sound and check records kept as-is.

• assets/design-exploration/SUNPOP/: bold-brand-pop user-agreed 24-second brand ad, including original packaging and graphics, continuous 3D can, native compositing, three sound versions with stems, subtitles, and checks. Migrate design and motion methods per Bold Brand Pop; parent-child coordinates, asset in/out points, transparent plate cropping, and sound ending corrections are all preserved.

• assets/design-exploration/AUREL/: sculptural-type user-requested sculptural type case, including 16 seconds four sound choices, silent visuals, shared vector paths, continuous 3D plate, native compositing, and sound stems. Migrate per Sculptural Type + Dynamic Typography; corrections for 3D occlusion to flat outline and revision experience for mismatched music timbre are preserved.

• assets/multistyle-motion/: SHIFT brand typography, ORBIT function feedback, Yinian Collaboration product assembly—three v11 continuous motion videos, including native projects, 3D rotation data, audio and rebuild methods; completion and review scope in accompanying records.

• assets/motion-correction/: TRACE v10 16-second continuous graphic narrative correction segment, native project, audio editing plan and actual checks; user subsequent evaluation "this effect is decent" applies only to this version, not extended to old nine scenarios or all use cases.

• assets/scenario-catalog.json / scripts/plan_scene.py: 24 scenario recipes, 22 style routes, asset and shot planning. Prioritize reading individual items by ID to avoid irrelevant context.

• assets/domain-style-routes.json / scripts/choose_style.py: 11 software, 5 brand, 3 hardware specific routes, cross-combined with existing usage directory; route count does not indicate equal maturity.

• assets/flagship-studies/: TRACE v2 / ORBIT v4 / SHIFT v2 and six style short samples' editable native projects, 3D sources, original sound, stems, subtitles, and rebuild entry points. Copy to new directory and execute; main video build.py, short samples build_styles.py, mix audio/mix.py. Full usage and check scope in accompanying production_instructions.md.

• assets/scenario-components/: 6 native components, rebuild scripts and actual export checks; usage see component reference page.

• assets/feature-release/: Brand Color Sync v3 24-second horizontal and 18-second vertical projects, original footage, note and sound effect sources, stems, and rebuildable scripts. Use python3 assets/feature-release/rebuild.py --out /absolute/new-project/source --render to copy to new directory and produce. It is a specific feature video case; do not treat it as a commercial template for all scenes just by swapping titles.

• assets/task-results/: Yinian Collaboration v3 / DataTalk v3 native projects, actual web pages and briefs, shared data and cues, original sound, and rebuild scripts. Use python3 assets/task-results/rebuild.py --out /absolute/new-project/source --render to produce new copies; scope and limitations in task-to-result reference page.

• assets/background-staging/: v3.1 background and lighting overlay, reusing assets/task-results/ assets; use this directory's rebuild.py --out /absolute/new-project/source --render to generate independent complete projects.

• assets/scene-expansion/: PULSE API 24s, SHIFT brand manifesto 18s, and LOOP service journey 26s vertical—three independent narratives and visuals. Use this directory's rebuild.py --out /absolute/new-project/source --render to produce new copies; includes story data validation and sound cue consistency checks.

• assets/coverage-expansion/: 13 1080p30 tech-use-case sample videos' native projects, split models, map/tracking data, UI components, and original sound; rebuild.py --out /absolute/new-project/source --only <id> --render allows single-item rebuild. All 13 projects built in new directories and passed per-item native single-frame consistency checks.

• assets/knowledge-assistant/: TRACE 32-second knowledge assistant concept video native project, same-source materials and answers, original sound, and check records; use rebuild.py --out /absolute/new-project/source --render to produce new copies.

• assets/release-polish/: Brand Color, Yinian Collaboration, and DataTalk v4 overlays, sound, and rebuild entry; reuse historical assets, generate independent projects in new directories.

• assets/effectcraft-native/: Four fully EffectCraft editable cases, including actually executed layer/effect/camera scripts and sound generation sources; copy to project directory, then rebuild with rebuild_v2.py --new-scenes-only --render.

• assets/optical-materials/: Form glass controls and ORBIT metal controller Blender, EffectCraft, original sound source code. Use for solid material and macro production research; specific results and boundaries in optical reference page; not a pre-reviewed commercial template.

• scripts/compile_cues.py: Generate sound effect cues from named visual markers in the project, check time ranges and shot intervals; avoid using old hardcoded audio after changing frame timing. Format see assets/production-plan.example.json.

• scripts/qa_media.py: Fully decode actual video, measure final audio track, extract frames from review segments; does not score aesthetics.

• scripts/create_example.py / assets/canvas-example/: Old 32-second geometric and UI technical exercise, used for learning deterministic rendering and regression checking. User has pointed out its quality falls short of high-end launch video goals; do not default to copying it into new commercial videos, nor use it as an aesthetic acceptance baseline.

• assets/focus-study/: This round's continuous 3D product space research generation source code and plan. Used for expressing material → association → brief → source backtracking; actual quality and validation scope per its case records.

python3 scripts/compile_cues.py /absolute/path/to/production-plan.json --out /absolute/path/to/cues.json
python3 scripts/qa_media.py /absolute/path/to/film.mp4 --out /absolute/path/to/qa --times 3.8,4,4.2,9.9,10.1

Delivery scale matches task: final video, editable sources, actually used assets, necessary stems/dependencies, and rebuild method. Produce other aspect ratios, silent loops, subtitle/localized versions as needed; do not add external publishing or software installation without authorization. Report verified capabilities; do not claim all styles are mastered or "one-click commercial grade."

The public version retains local, procedural, and online options per Voiceover & Music Options; prioritize respecting the user's existing configuration.