← WidemileCONTENT HISTORY
WHAT CHANGED
Update to Widemile
Snapshot Oct 24, 2026 · 13:07 UTC · version 1.1.0
Collection source: downloaded plugin package.
First saved snapshot
No earlier snapshot is available to establish a change.
Save Widemile and check future changes in Following.
Up to 3 plugins free · no card required.
Compare saved observations
Download comparison JSONFull technical diff · 0 changed fields
Full snapshot data
{
"description": "Create a cinematic scale-shifting VFX shot where a person or subject appears as a towering giant integrated into a city; use for giant, colossal subject, urban giant, diventare gigante, persona gigante, colosso in città, hacerse gigante, or escala colosal, not for ordinary resizing, cropping, or unrelated miniature effects.",
"included_files": [],
"name": "vfx-giant",
"skill_md_contents": "---\nname: vfx-giant\ndescription: Create a cinematic scale-shifting VFX shot where a person or subject appears as a towering giant integrated into a city; use for giant, colossal subject, urban giant, diventare gigante, persona gigante, colosso in città, hacerse gigante, or escala colosal, not for ordinary resizing, cropping, or unrelated miniature effects.\n---\n\n# Giant-scale VFX\n\n## Tool requirements\n\n- **Core evidence and bindings:** `prostudio_brief_get`, `prostudio_workspace_get`, `prostudio_references_list`, `prostudio_references_resolve`, `get_prostudio_media_catalog`, and `inspect_prostudio_media`.\n- **Generative giant shot:** `prostudio_models_list`, `prostudio_generation_prepare`, `prostudio_video_generate`, `prostudio_job_status`, and `prostudio_generation_get`.\n- **Conditional support asset:** `prostudio_image_generate` only when an approved still support asset must be created before the final video generation. The final giant result must still be one video clip.\n- This declaration does not activate tools. Use only the tools required by the selected production method. If a required tool is absent from the current scoped catalog and `load_prostudio_tools` is available, call it with the missing outcome before claiming the capability is unavailable. If retrieval still exposes no matching tool, report that limitation. A model-catalog lookup alone is never completion for a requested giant shot.\n\nCreate a cinematic shot in which the intended subject reads as physically enormous inside an urban environment. Preserve identity while making scale, contact, lighting, depth, motion, and the city's response feel coherent. The city is a narrative participant: in a realistic giant shot, people, traffic, and nearby urban elements must react to the presence of the giant. Never let the city remain indifferent, with normal pedestrians, routine traffic, or background extras ignoring the event, unless the user explicitly requests an intentionally surreal counterpoint.\n\nChoose the smallest viable workflow from the user's intent and the live workspace evidence. Do not add deterministic routing, keyword matching, catalog changes, or UI changes as part of this skill.\n\n## Choose the production method\n\n- Use a **generative giant shot** when the user has an identity reference, wants ProStudio to create or reinterpret the city, wants a moving camera, or wants the giant to interact with buildings, streets, traffic, weather, or crowds. Unless the user explicitly requests a continuous take, design the finished clip as an edited multi-cut sequence rather than one uninterrupted angle.\n- Use a **scale transformation shot** only when the user explicitly wants to see the subject grow from human scale to giant scale. Default to a matched multi-cut progression inside the same generated clip—human-scale setup, growth or reaction insert, colossal reveal—when the duration supports it. Use one continuous generative take only when the user requests it or when cutting would make the transformation less legible. If the selected model cannot preserve the progression within one clip, simplify the cut plan or select another compatible live model; never split the result into separately generated shots.\n- Use an **existing city plate integration** only when the supplied material is suitable for the tools currently available. The current Editor does not provide automatic rotoscoping, animated masks, depth reconstruction, or a deterministic clean subject cutout. Never claim a clean composite into arbitrary moving footage. Offer a generative reinterpretation or request compatible prepared transparent or masked media when needed.\n- If the user only asks for an already-giant result, do not invent a transformation beat.\n\n## Establish the shot\n\n- Resolve from the request, the canonical Brief, and current workspace media: the exact subject, whether the result begins human-sized or already giant, target scale or a clear urban scale anchor, action, city and time of day, intended duration, aspect ratio, camera movement, realism or stylization, environmental damage, dialogue, audio, and the intended reaction beat. Ask only for choices that materially change the method or result.\n- Classify every shot before prompting it:\n 1. **Reveal:** people begin to perceive the giant; use scattered eyelines, turns, stops, pointing, or phones, with fear beginning to spread rather than an instant synchronized stampede.\n 2. **Street-level or close:** fear is dominant; show nearby people backing away, running, ducking, taking cover, dropping belongings, or searching for an escape path. Some may look up, point, or film, but they must read as endangered rather than casually curious.\n 3. **Wide:** show a distributed evacuation wave and the city's response: people split into different escape paths, traffic brakes or diverts, vehicles create congestion, and some vehicles may be abandoned when the blocking supports it.\n 4. **No visible people:** communicate scale through traffic, architecture, shadows, occlusion, dust, vibration, displaced objects, glass, signage, haze, and other physically coherent environmental cues; do not fill the shot with invented indifferent extras.\n- Treat the reaction as a causal sequence, not a static mood: baseline city motion before perception, nearest observers notice first, fear spreads with distance and attention, then flight, traffic disruption, and environmental aftermath. Vary reaction latency, direction, intensity, and individual choice so the crowd never behaves as one synchronized animation.\n- Load `media-inspection` and inspect every selected identity reference, city reference, or source clip before planning. Prompts, filenames, thumbnails, and summaries are not evidence that identity, framing, lighting, or perspective is suitable.\n- Reuse user-designated media through `reference-management`. Do not generate a replacement identity, outfit, or city reference unless the user asks for one or chooses to create a missing reference.\n- Identify missing support assets before generation. Depending on the shot, these can include a city plate, street-level or aerial city reference, crowd/traffic reference or plate, clean background, and environmental-reaction reference. Ask for the missing asset or, when generative assets are allowed by the request, generate the smallest compatible support asset; do not silently assume a crowd, traffic pattern, or destruction level, and never replace the selected identity or city without the user's choice.\n- Treat the selected subject as the identity reference. Preserve face, body proportions, hair, wardrobe, accessories, and defining features unless the user requests a transformation beyond scale.\n\n## Design the multi-cut sequence\n\n- A standard giant clip should contain multiple deliberate editorial cuts unless the user explicitly asks for a oner, continuous take, or surveillance-style locked shot. Use two or three readable cuts for clips up to 5 seconds, three or four cuts for 6–10 seconds, and four or five cuts for 11–15 seconds. Prefer fewer clear cuts over a frantic montage; each cut needs enough screen time for its scale cue and reaction beat to register.\n- Give every cut a distinct narrative and visual purpose. A strong default progression is: **CUT 1 — reveal or establishing scale**, **CUT 2 — street-level fear and first evasive movement**, **CUT 3 — low-angle giant action, contact, or environmental impact**, and **CUT 4 — wide evacuation, traffic disruption, or aftermath**. Adapt or omit beats to fit duration and intent, but do not repeat the same composition merely to increase the cut count.\n- Vary shot scale and viewpoint across the sequence: combine an architectural wide, street-level or over-the-shoulder reaction, low-angle medium or close detail, aerial/crane geography, or a no-people environmental insert. Change focal length and movement intentionally while preserving the same lens family, capture intent, time of day, city geography, screen direction, wardrobe, and giant scale.\n- Make cuts editorially explicit. Prefer motivated direct cuts, reaction cuts, impact cuts, eyeline matches, foreground wipes, or sound bridges. Do not substitute dissolves, random whip pans, speed ramps, camera teleports, morphing viewpoints, or a continuous zoom for actual cuts unless the user requests that language.\n- Preserve causality across the edit: the reveal must precede fear, fear must precede flight or traffic disruption, and aftermath must follow a visible or audible cause. Reaction timing may overlap across cuts, but a later cut must advance the event rather than reset the city to an indifferent baseline.\n- Preserve spatial continuity. Maintain the giant's orientation, contact plane, landmark relationships, sun and shadow direction, traffic direction, crowd escape vectors, and screen geography across adjacent cuts. Deliberately cross the axis only when the new geography is re-established.\n- The entire result must be one video generation plan, one `prostudio_video_generate` execution, and one durable video clip. Write every cut into the same prompt with explicit time ranges. Never generate cuts as separate shots, never create multiple video jobs for one giant clip, and never use the Editor to assemble the cut sequence. When the model cannot support the desired number of cuts reliably, reduce the cut count while keeping at least two distinct setups, simplify the action, or select another compatible live model.\n\n## Plan the scale illusion\n\n- Load `production-planning`, `reference-management`, and `generation-planning`. Read or create the smallest shot plan needed for the effect and bind each exact reference with an explicit identity, environment, or style purpose.\n- Define scale with at least one visible relationship rather than adjectives alone: for example, shoulders level with a twenty-story roofline, feet spanning a street intersection, or a hand wrapping around the edge of a recognizable building. Keep the relationship plausible throughout the shot.\n- Establish a clear contact plane. Describe where both feet, hands, or body weight meet the environment; how nearby architecture occludes the subject; where cast shadows fall; and how haze, atmospheric perspective, contrast, and detail diminish across depth.\n- Favor slow, weighty acceleration and readable body mechanics. A colossal subject should not move with the light timing of an ordinary person unless the user intentionally wants a surreal result.\n\n## Direct the cinematic capture\n\n- Specify camera language, not only \"cinematic\": camera height and angle, lens focal length and character, sensor or format intent, focus distance and depth of field, foreground occlusion, horizon and vanishing lines, framing, aspect ratio, frame rate, shutter or motion-blur feel, exposure, color temperature, contrast, and light direction. Keep perspective, verticals, parallax, and screen geography consistent with the city reference.\n- When the user requests an ARRI ALEXA look, express it as capture intent: **ARRI ALEXA 35, LF, or 65-style image** as appropriate, natural skin tones, controlled highlight roll-off, detailed but non-digital shadows, restrained film grain, realistic motion blur, and a coherent LogC-style cinematic grade. Do not claim that a specific camera body, lens, codec, or color pipeline was actually used unless the available tool reports it; describe it as a visual target.\n- Choose movement that reinforces the story and shot class. Use a slow dolly or push-in, a controlled tilt/crane reveal, or a measured lateral track for discovery; a low-angle retreat, shoulder-mounted handheld, stabilized tracking, or rack focus for street-level fear; a crane, aerial pullback, orbit, or wide lateral move for collective evacuation; and a restrained push, tilt, parallax move, or locked-off frame for no-people scale cues. Avoid arbitrary orbiting, weightless drone motion, or random shake.\n- Give adjacent cuts genuinely different coverage: change shot scale, camera height, lens, foreground relationship, or subject emphasis while keeping the package coherent. Do not repeat the same camera move in every cut, and do not treat a digital punch-in as a new setup.\n- Let camera behavior respond to the event without stealing causality: the camera may lag, reframe, shake from a grounded impact, lose and recover focus, or pull back as the city reacts, but the motion must have a physical or editorial reason. Keep handheld instability controlled and readable; do not use identical shake on every shot.\n- Shape light and atmosphere as part of scale: motivate key and fill from the actual sun, sky, practicals, or street sources; preserve exposure continuity; use backlight, haze, volumetric depth, shadow travel, reflections, wet-surface response, and localized dust only when the location and impact support them. Keep faces, signage, and the giant readable without flattening depth or clipping highlights.\n- Across the cuts inside the single clip, lock the camera package intent: format, lens family, color and contrast, time of day, shutter/motion blur, camera axis, and light direction. Change focal length or movement only when the cut calls for a deliberate escalation or reveal.\n\n## Plan the city's reaction\n\n- Build the prompt around five readable beats: **before** (ordinary motion only until the giant is perceived), **perception** (eyelines, turns, stops, pointing, phones, or a first vehicle brake), **fear** (staggered shock and self-protective behavior), **flight or disruption** (running, dispersal, cover, evasive traffic, congestion, or abandonment), and **aftermath** (dust, alarms, displaced objects, shadows, vibration, or settling debris when supported by contact and scale).\n- Make the nearest people react first and the farther groups later. Mix freezing, backing away, falling, helping someone, taking cover, running in different directions, pointing upward, and filming as secondary behaviors. The dominant read must remain danger and self-preservation; bystanders must not continue shopping, walking, driving, or chatting normally while looking at the giant.\n- Give traffic a physical cause and a visible consequence. Drivers should brake, stop short, swerve, divert, block lanes, or abandon vehicles only when the giant, shadow, debris, crowd, or blocked road explains it. Preserve lane direction and screen geography; do not make every vehicle move at once or turn traffic into weightless chaos.\n- Scale environmental reactions by proximity. Nearby objects can shake, lift dust, rattle glass, flex signs, move loose debris, and receive heavy occlusion or shadow; midground elements can carry haze, vibration, alarms, and changing shadows; distant architecture should mainly provide silhouette, occlusion, perspective, and atmospheric scale. Do not add arbitrary destruction when the shot does not support impact.\n- Keep reactions asynchronous and legible: stagger people and vehicles across the frame, use different escape vectors and response times, and preserve plausible gaps between cause and effect. Avoid a crowd that turns, runs, falls, or raises phones on the same frame.\n\n## Generate the shot\n\n- Read the live video-model catalog and choose a currently available reference-compatible mode. Do not hardcode a provider or promise deterministic identity, exact physics, or perfect scale continuity.\n- Prepare exactly one video generation plan and execute exactly one `prostudio_video_generate` action for the finished result. All planned cuts must be authored inside that one clip; do not dispatch one job per cut.\n- Write one spatially explicit prompt in English. State the multi-cut structure; exact identity reference; scale anchor; city geography; continuity locks; camera package or ARRI ALEXA visual intent when requested; aspect ratio, frame rate, shutter/motion-blur feel, exposure and color intent; and the timing relationship between perception, fear, flight, and aftermath. For every cut, state its exact time range, shot class and causal beat, framing, camera height and angle, focal length and lens character, movement and motivation, focus and depth of field, giant action and contact, foreground and background occlusion, people and traffic reaction, environmental reaction, lighting, cast shadows, and atmospheric depth. Preserve dialogue and visible text in the user's requested language.\n- Use this compact structure for a multi-cut prompt and replace every bracketed field with concrete shot information:\n\n ```text\n EDITED MULTI-CUT CINEMATIC SEQUENCE, not one continuous take.\n CONTINUITY LOCK: [identity, wardrobe, giant scale, city landmarks, screen direction, time of day, light and shadow direction, camera package, color intent].\n CUT 1 [start–end]: [reveal/establishing purpose]; [shot scale, height, angle, lens]; [motivated camera movement]; [giant action and scale anchor]; [first perception cues]. DIRECT CUT.\n CUT 2 [start–end]: [street-level fear purpose]; [distinct framing and lens]; [motivated movement]; [distributed human reaction and traffic cause]. DIRECT OR REACTION CUT.\n CUT 3 [start–end]: [giant action/contact purpose]; [distinct low-angle or detail coverage]; [impact, shadow, occlusion, dust or vibration]; [reaction escalation]. IMPACT OR EYELINE CUT.\n CUT 4 [start–end, when duration supports it]: [wide evacuation/aftermath purpose]; [geographic coverage and lens]; [escape vectors, traffic disruption, environmental consequence]. DIRECT CUT OR SOUND BRIDGE.\n AUDIO CONTINUITY: [continuous city bed, alarms, crowd perspective, cut-specific impacts, dialogue placement].\n ```\n- If duration supports only two or three cuts, remove unused lines and redistribute the causal beats; never leave placeholder text or compress four unreadable setups into a short clip.\n- Include positive behavior, not only negative constraints: specify who notices first, how reactions spread, which people flee or seek cover, which secondary actions appear, how traffic responds, and which nearby objects communicate impact. For a no-people shot, explicitly substitute non-human scale cues.\n- Explicitly forbid an indifferent city: no pedestrians calmly continuing routine activity, no regular traffic flow after the giant is visible, no extras staring without danger behavior, no synchronized panic cycle, and no background group ignoring the giant. Also exclude a normal-sized duplicate of the subject, floating or sliding feet, toy-city or miniature-diorama appearance unless requested, scale drift, merged anatomy, extra limbs, extra people resembling the subject, identity drift, abrupt wardrobe changes, and buildings intersecting the body without intended occlusion.\n- For a growth transformation, specify the start scale, end scale, duration of the growth beat, fixed environmental scale anchors, body posture, camera response, and how contact with the ground changes. Avoid uncontrolled morphing by keeping action and camera movement simple during the scale transition.\n- Prepare the exact generation, report the returned model, duration, aspect ratio, audio behavior, and credit charge, then stop for the host's required confirmation. After approval, execute the prepared action without changing its arguments, poll the durable job, and inspect the completed video before calling the effect successful.\n\n## Preflight and quality controls\n\nBefore generation, verify that:\n\n- the shot has one of the four reaction classes and a visible scale anchor;\n- the clip has a duration-appropriate multi-cut plan with explicit time ranges, at least two genuinely different setups unless the user requested a continuous take, and no repeated composition disguised as coverage;\n- adjacent cuts preserve identity, giant scale, city geography, screen direction, light, and reaction state while changing viewpoint for a clear editorial reason;\n- the prompt contains the causal order perception → fear → flight or disruption, with no crowd-wide synchronization;\n- the necessary identity, city, crowd/traffic, and environmental references exist or have been requested/generated as support assets;\n- the camera plan states movement, motivation, height/angle, lens, focus, format, frame cadence, light, exposure, and color intent rather than relying on the word \"cinematic\" alone;\n- people, traffic, and environmental cues are specified positively and match the shot's framing, contact, and intended damage level;\n- the prompt explicitly rejects an indifferent city and does not promise a deterministic composite.\n\nAfter generation, inspect every cut and every transition, then inspect the sequence as a whole. Verify that the cut count and timing match the plan, each setup contributes different scale or reaction information, identity and scale remain stable, geography does not reset, direct cuts read as edits rather than morphs, and audio perspective remains coherent. Inspect the first perception moment, the fear transition, the escape/disruption beat, and the final environmental state. Check foreground, midground, and background separately, then check camera movement, focus pulls, motion blur, exposure/highlight roll-off, color continuity, lens distortion, parallax, and whether the camera's response has a physical or editorial motivation. Fail the shot if a requested multi-cut clip collapses into one uninterrupted angle, cuts are replaced by zooms or morphing camera teleports, consecutive setups are materially identical, continuity resets between cuts, the giant is visible but pedestrians remain routine, traffic stays regular without a stated reason, reactions happen before anyone could perceive the giant, the whole crowd reacts in unison, the camera movement feels arbitrary or weightless, or the environment has no scale consequence. For no-people shots, fail it if architecture, traffic, shadows, occlusion, or physical atmosphere do not carry the scale read.\n\nWhen revising, isolate the failing cut and layer in the diagnosis, then revise the one-clip prompt while preserving every approved element: add delayed and distributed human reactions without changing identity or scale; repair traffic causality without turning every vehicle into chaos; strengthen contact, shadow, dust, vibration, or occlusion without inventing unsupported destruction; repair one transition or continuity mismatch; or replace missing support assets. Because the output must remain one generated clip, regenerate the complete clip with the localized correction rather than generating or replacing an isolated cut.\n\n## Revisions and completion\n\n- Evaluate the result on exact identity, readable colossal scale, stable scale through time, body integrity, ground contact, shadow direction, depth and occlusion, atmospheric perspective, weight of motion, camera coherence, motivated camera movement, lens and perspective integrity, focus/depth-of-field continuity, motion-blur cadence, exposure and highlight handling, color/light continuity, city consistency, and absence of an unintended normal-sized duplicate.\n- For reveal shots, confirm that the first observers notice before the wider reaction; for street-level shots, confirm that fear and self-preservation dominate; for wide shots, confirm that escape paths, traffic disruption, and urban impact read across the frame; for no-people shots, confirm that non-human cues carry the entire scale and causal read.\n- Reject any result that depicts an indifferent city, routine pedestrians, regular traffic, or a synchronized/artificial crowd as the default reality of the shot. A few people filming or looking upward are acceptable only when they remain embedded in a dangerous, distributed response.\n- Preserve approved identity, framing, scale, timing, environment, cut order, and unaffected cut instructions. Apply localized prompt corrections, but regenerate the complete single clip unless the user asks for a new direction.\n- Summarize which method was used, which source assets or references were used, what single clip was created, and any remaining limitation. Never describe a generative reinterpretation as a deterministic composite of the user's original performance.\n"
}SHA-256 of public snapshot: 6895864e4f94c9b50251c9fa63ecf31fda40b147f178e99b2f305a700ed8b8f2