Global-AI-Group-Inc/content-generation-skills / skills/mira-video-prompting
mira-video-prompting
VerifiedWhich of Mira's video models to pick (the live list is in references/selection.md) and how each one treats a prompt: Seedance 2.5, Runway Gen-4.5, Kling 4.0, Kling 3.0 (and Turbo), Gemini Omni Flash, MiniMax H3, HappyHorse, Wan 3.0. Use it before any generate_video call on Mira, when the user asks "which model", "how long can it be", "does it have sound", "промпт для видео", or pastes a video prompt to improve. NOT for stills (mira-image-prompting) or for the camera and light banks themselves (mira-craft).
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 5.3 KB |
| references/action.md | markdown | 4.2 KB |
| references/models/capricornus-1.md | markdown | 9.9 KB |
| references/models/happyhorse.md | markdown | 5.8 KB |
| references/models/kling-4.md | markdown | 5.2 KB |
| references/models/kling-motion.md | markdown | 2.2 KB |
| references/models/kling-turbo.md | markdown | 2.0 KB |
| references/models/kling.md | markdown | 8.2 KB |
| references/models/minimax.md | markdown | 7.7 KB |
| references/models/mira-cinematic.md | markdown | 5.3 KB |
| references/models/omni.md | markdown | 5.7 KB |
| references/models/runway.md | markdown | 5.6 KB |
| references/models/seedance-2-5.md | markdown | 9.9 KB |
| references/models/veo-lite.md | markdown | 5.1 KB |
| references/models/veo-quality.md | markdown | 5.1 KB |
| references/models/veo.md | markdown | 5.0 KB |
| references/models/wan3-prime.md | markdown | 7.9 KB |
| references/models/wan3.md | markdown | 7.8 KB |
| references/selection.md | markdown | 6.3 KB |
| references/textures.md | markdown | 10.8 KB |
| references/universal-rules.md | markdown | 7.8 KB |
references/models/runway.md
View on GitHub<!-- generated from the registry by comfy_study/comfy_video/build_skills.py; edit the registry, not this file -->
# Runway Gen-4.5 (`runway`)
| | |
|---|---|
| seconds | 5 / 10 |
| output | provider default |
| reference images | 4 |
| audio | no |
| prompt budget | 1000 chars |
| extend | no |
| draft | no |
| last frame | no |
| reference clips | no |
| reference audio | no |
## When to pick it
Gen-4.5, text-to-video AND image-to-video: the sequencing specialist - executes several ordered camera moves and timed beats inside ONE prompt, and takes forceful camera work (sweeping arcs, crash zooms, whip pans) the others smear; weak on fine motor detail; NO audio; 720p; 5/10s
## How it reads a prompt
MODEL: Runway Gen-4.5. It handles BOTH text-to-video (no start frame) and image-to-video, and Runway's guides are now written for this model. WITH A START FRAME the frame already fixes subject, composition, colors and style, so the prompt describes ONLY motion and refers to the subject generically ('the subject', 'she', 'the bottle') - re-describing the picture reduces motion and destabilises the shot; describe visuals in that mode only to introduce an element absent from the frame, to specify a transformation, or to define an interaction between two elements. WITHOUT a start frame describe BOTH the visual scene (subject, setting, lighting, mood, lens) AND its motion, still leading with motion. Official template: 'The camera <motion description> as <the subject> <one concrete action>. <Additional descriptions>'. Motion slots: subject action, environmental motion, camera motion, motion style and timing, direction and speed. SEQUENCES ARE OFFICIALLY SUPPORTED on Gen-4.5 and are its strength - it executes complex, sequenced instructions, multiple camera moves and precise event timing inside ONE generation. Order the beats either in natural language ('X happens, then Y happens. Finally Z happens.') or with rough timestamps ('[00:01] X. [00:03] Y. [00:05] Z.'). Match the number of beats to the CLIP SPEC duration: two beats for a 5-second clip, three to four for a 10-second clip. Dynamic, forceful camera work is allowed and documented - sweeping arcs, crash zooms, whip pans, handheld shake - so pick the energy the idea actually needs instead of defaulting to slow. Name each camera move precisely ('dolly in', 'push in', 'truck left', 'pan', 'tilt', 'crane up', 'orbit around', 'tracking shot', 'whip pan', 'handheld') and pair it with speed ('slowly', 'rapidly', 'steadily'); prefer these over the ambiguous 'zoom'. CRITICAL - the final prompt must contain ZERO negation words (no, not, avoid, without, never, don't). Runway documents that negative phrasing 'may produce unpredictable or even opposite results'. Express every constraint POSITIVELY and NAME WHAT STAYS STILL: write 'Locked camera. The camera remains still.' rather than 'no camera movement'; write 'the label stays flat, sharp and rigid' rather than 'the text does not warp'; write 'sharp focus, high detail' rather than 'not blurry'. Keep hands calm and any readable text (labels, screens, packaging copy) flat, rigid and still while the camera carries the motion, and let hair, fabric, steam, liquid, light and reflections do the secondary motion - fine-motor beats (sipping, spraying, typing, finger gestures) remain the weakest spot of this family. Give no conflicting instructions ('dramatic shadows with soft even lighting'), no abstract or conceptual language ('embodies the essence of joy'), and no conversational requests ('can you please add...') - Runway documents all three as failure modes. GOLD EXAMPLES (verbatim from Runway's official guides, incl. Gen-4.5 - match this register: short, plain, motion-first): 'The camera slowly pushes in as the person scales the giant soda.' | 'The camera executes an aggressive, sweeping horizontal arc around the subject, followed by an extremely rapid, aggressive crash zoom that concludes with a sharp focus on the subject's eyes.' | 'Handheld camera: The man stands still as the crowd moves around him. He starts yelling as the camera slowly zooms out. Natural camera shake.' | 'Whip pan to painting of a fox. Whip pan back to the woman with a curious expression. Whip pan back to the fox painting, the fox is moving.' | 'The man extends his arm to shake hands, then nods politely.' Keep the final prompt under 1000 characters - Runway's hard cap, enforced by the API. If the beats will not fit, drop scene adjectives, never a beat. DOCS 2026-11 (Runway Gen-4.5 guide): several ordered camera moves inside one take are this model's strength - keep a sequence of moves as written; the general 'one dominant move per shot' rule does not apply to Runway.
## Doctrine
Runway Gen-4.5: the SEQUENCING model - it executes several ordered camera moves and timed beats inside one prompt, and it is the one place forceful camera work belongs (sweeping arcs, crash zooms, whip pans, handheld shake). Text-to-video and image-to-video both native, so it works with or without imageRefs. Silent - no audio track at all - and 720p, which is Gen-4.5's hard ceiling. Prompt under 1000 characters, positive phrasing with ZERO negation words; order the beats in natural language or with rough timestamps. Weak spot: fine motor detail.
## Format law the critic applies
FORMAT CHECK: positive phrasing only - zero negation words (no/not/without/never); name what stays still instead. With a start frame describe MOTION only; without one describe the scene AND its motion. Order the beats in sequence, matching their number to the duration - two for 5 seconds, three or four for 10. Every camera move named precisely and paired with a speed. Under 1000 characters.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |