Skip to content

Global-AI-Group-Inc/content-generation-skills / skills/mira-video-prompting

mira-video-prompting

Verified

Which 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).

references/models/veo-quality.md

View on GitHub
<!-- generated from the registry by comfy_study/comfy_video/build_skills.py; edit the registry, not this file -->

# Google Veo 3.1 Quality (`veo-quality`)

| | |
|---|---|
| seconds | 8 |
| output | 1080p |
| reference images | 3 |
| audio | yes |
| prompt budget | 3000 chars |
| extend | forward |
| draft | no |
| last frame | yes |
| reference clips | no |
| reference audio | no |

## When to pick it

the flagship Veo for hero clips: top cinematic realism at ~3x the Fast price - finals only, never drafts. ONE UNBROKEN TAKE: it cannot cut, so a brief with cuts, a chase or a launch stride is not its brief even when it is the hero

## How it reads a prompt

MODEL: Veo 3.1 (Google, full quality tier - the same model family as Veo 3.1 Fast but with markedly better detail, motion coherence and audio fidelity, so it is worth the extra care in the prompt). Its differentiator is NATIVE SYNCHRONISED AUDIO plus strong cinematic realism and physics, so always design sound as a real scene layer rather than a 'with sound' afterthought - undefined audio makes Veo rush delivery, mismatch ambience, or paste distracting captions. GOOGLE'S DOCUMENTED FORMULA: cinematography + subject + action + context + style and ambiance. Front-load the shot type and camera work. DIALOGUE: introduce speech with a COLON and NO quotation marks - 'the man in the red hat says: Where is the rabbit?' is Google's own documented form, and it also avoids the burned-in captions that quotation marks tend to trigger. Put the spoken line early. Always add a vocal-tone cue plus a voice and accent descriptor, because Veo drifts accents between clips, and keep dialogue to ONE short sentence - a single natural breath is all that fits a 4-8 second clip. AUDIO: Google documents that audio should be described in SEPARATE SENTENCES from the visuals. Label them - 'SFX: the cap clicks shut.' and 'Ambient noise: quiet cafe murmur.' - plus a brief music mood. CAMERA: one clear move (push-in, dolly, tracking, crane, pan) with its pace named for the lens only - the subject keeps real-time speed and weight. Keep faces and motion natural and physically believable. Veo is weak at multi-subject coordination, so keep to one or two subjects. STORYBOARD FOR VEO: Veo officially supports timestamped storyboards, so ALWAYS write an explicit timeline in Google's bracket form - '[00:00-00:03] ... [00:03-00:06] ... [00:06-00:08] ...'. Cover the exact clip length given in the CLIP SPEC with no gap and no overrun: a 4-second clip gets two segments, 6 seconds gets two or three, 8 seconds gets three. Give every segment its own visual beat AND its own audio line, and keep one continuous camera move running across the segments so the clip still reads as a single unbroken take rather than a set of cuts. NEVER write the words 'no' or 'don't' anywhere in the prompt: Google documents that instructive negation is unreliable and that exclusions belong in the separate negative field as plain nouns. If the clip should be silent, fill the audio positively with ambience and music only and describe the subject as alone with lips closed. Do NOT add resolution jargon codes - those are set separately. GOLD EXAMPLES (verbatim, Google official docs): 'A medium shot in a dimly lit interrogation room. The seasoned detective says: Your story has holes. The nervous informant, sweating under a single bare bulb, replies: I'm telling you everything I know. The only other sounds are the slow, rhythmic ticking of a wall clock and the faint sound of rain against the window' | 'Medium shot, a tired corporate worker, rubbing his temples in exhaustion, in front of a bulky 1980s computer in a cluttered office late at night. The scene is lit by the harsh fluorescent overhead lights and the green glow of the monochrome monitor. Retro aesthetic, shot as if on 1980s color film, slightly grainy.' | 'A montage of pizza making: a chef tossing and flattening the floury dough, ladling rich red tomato sauce in a spiral, sprinkling mozzarella cheese and pepperoni, and a final shot of the bubbling golden-brown pizza, upbeat electronic music with a rhythmical beat is playing, high energy professional video.' Keep the final prompt to 4-6 vivid sentences plus the timestamp segments, and well under 1024 tokens. REFERENCE PHOTOS: when SEVERAL photos are attached they reach Veo natively as asset reference images that preserve each subject's appearance - NOT as a start frame, so describe the opening composition yourself. Name each photographed subject descriptively ('the red sneaker', 'the dark-haired woman') and keep its identity unchanged for the whole clip. References force the clip to exactly 8 seconds. A SINGLE photo still arrives as the literal opening frame - then describe only how it comes alive.

## Doctrine

Veo at full quality - markedly better detail, motion coherence and audio fidelity, and the best multi-subject coherence in the set: several people, a crowd or background life in one frame. Reserve for the flagship shot; ~3.5x veo.

## Format law the critic applies

FORMAT CHECK: introduce dialogue with a colon and NO quotation marks ('she says: ...'), audio as its own short sentences (SFX: ... / Ambient noise: ...). Zero negation words in the visual description.
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: 472kUpdated Claude CodeCodexVerifiedInstall
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: 323kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 323kUpdated Claude CodeCodexVerifiedInstall
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 availableStars: 323kUpdated Claude CodeCodexVerifiedInstall