Skip to content

Global-AI-Group-Inc/content-generation-skills / chatgpt/overrides/skills/mira-generate

mira-generate

Verified

Generate images and video on Mira AI: pick a model, attach the user's location or product, write the prompt, pass a playbook skill, wait for the result, extend or upscale a clip. Use it whenever the user wants to "generate", "make a video", "make an image", "animate this photo", "create an ad", "make a 3D model". Chain with mira-video-prompting and mira-image-prompting for what to write, and with the mira-* playbooks for the genre. NOT for prompting theory on its own, and not for voice, music, sound effects or captions (mira-audio).

Mira generate

Mira is a generation platform with a prompt pipeline behind every tool: what you state is kept word for word, and a set of specialists fills only the craft you left out. That changes how you write. You do not have to know lighting schemes to get good light. You do have to be exact about the facts: who, what, where, what happens, what is said, what must not change.

Connect

The Mira plugin connects the server for you. The user signs in with their Mira account the first time a tool needs it; there is no API key. If a call says the user is not signed in, ask them to connect Mira in the plugin settings and try again once.

Tools

ToolSpends creditsUse it for
list_modelsnoValid model ids, ratios, durations, resolutions. Call once per session.
list_skills / get_skillnoPlaybook ids for the skill parameter and the knowledge skills, bundled with the server.
list_brand_assetsnoThe user's saved locations and products with their ids.
upload_reference_imagenoAn image the user attached becomes a reference URL. A clip or a track they attach reaches the clip and audio tools as libraryItemId.
generate_imageyesStills. Returns a generation id at once.
generate_videoyesClips. Returns a generation id at once; video takes minutes. Takes up to 3 reference clips in referenceVideoUrls (motion, camera, pacing); kling-motion needs exactly one clip and one photo. draft=true on seedance-2-5 renders a cheap 480p take to try an idea; quality="hd" renders straight to 1080p.
list_generationsnoThe account's recent generations with ids, URLs and poster URLs: image / video / model3d by default, audio, transcript or analysis only when asked for by kind. To find an earlier clip to extend, a 3D model to rig, a source for a clip operation.
get_generation / wait_for_generationnoStatus and the result. wait_for_generation blocks up to ~50 s and may need several calls.
extend_videoyesNative continuation of a clip the account already made. direction="backward" (seedance-2-5) adds what happens before the first frame.
finalize_draftyesRe-renders a 480p draft as the same take in 1080p, within 7 days of the draft.
upscale_videoyesTopaz on a finished clip - a Mira clip (generationId) or an attached file (libraryItemId): target 1080p (default, 720p only), 4k (720p or 1080p, up to 30 s; fps up to 60 in the same pass) or fps (FPS Boost at the same resolution: fps 24-120, default 60; slowdown 2/4/8 for slow motion - longer clip, no sound; up to 60 s of output). credits_spent is the confirmed price. A result cannot be upscaled, boosted or extended again: for 4K at 60 fps run 4k + fps 60 on the original in one pass. Not for drafts: finalize them instead.
generate_3dyesOne object as a 3D model from text or up to four photos (hunyuan-3.1, meshy-7, trellis-2, rodin-fast). Result files (GLB plus other formats) come back in files. Playbook: 3d; read mira-3d.
rig_3dyesHumanoid skeleton with walk/run on a finished 3D generation (generationId), plus animation clips by preset id.
search_mira_guides / get_mira_guide / draft_mira_guide / list_my_mira_guides / update_mira_guide / publish_mira_guidenoCommunity guides at miralabs.art/guides: real results with the model and prompt behind them, written by users (not the playbooks). Search before an unfamiliar effect or look; draft a guide from a result the user likes; edit their own guides; publish only when they explicitly ask. references/guides.md.
list_direct_presets / direct_videodirect_video yesDirect presets: the user answers a short questionnaire (short film, music video, product or UGC ad, brand film, real-estate tour, micro drama, explainer, trailer, social post, ad remake) and the preset directs ONE clip up to 30 s - you write no prompt. Offer one when the user wants such a piece and brings no prompt. references/direct-presets.md.
swap_objectyesOne object in any clip (a product, a prop, a garment) replaced from 1-4 photos, everything else kept. Up to 10 s.
list_effectsnoReady-made effect presets (cakeify, figurine, age progression...) for the effect parameter of generate_image / generate_video, with what each needs attached. Only when the user names an effect.
modify_videoyesThe same clip with one thing changed: lighting, weather, time of day, backdrop, or a free restyle. Up to 15 s.
reframe_video / remove_backgroundyesA new aspect ratio with the new canvas filled (up to 15 s); a clean key without a green screen, with a luma matte or a ProRes alpha (up to 30 s).
marketing_briefyesScript, shot list, a ready generate_video prompt, caption, hashtags and voiceover text for a product clip, returned in the same call.
generate_audio / plan_musicyes / noSpeech, multi-voice dialogue, sound effects and music; plan_music drafts a free section plan for music timed to the cut.
list_voicesnoThe platform's library voices with their ids, for speech and voiceover.
add_soundtrack / voiceover_video / add_captionsyesOn a finished clip: music the model writes by watching it; a spoken script over the ducked original; burned subtitles plus an SRT.
isolate_voice / separate_stemsyesSpeech cleaned of noise, music and echo; a track or a clip's sound split into 2 or 6 stems.
transcribe_video / analyze_transcriptyes / partlyA word-timed transcript and SRT with speakers (up to 30 min); silence cuts and multicam splits (free), the best short clips, SEO, chapters.
estimate_costnoWhat a call would spend, before making it: image, video (duration, draft, quality="hd"), finalize, model3d (quality/pbr/rig/animations), rig, every clip operation (sourceSeconds) and every audio job (model, chars, durationSeconds). Quote it before a costly call.
film_list_projects / film_get_project / film_list_elements / film_list_assets / film_move_assets / film_update_brief / film_add_commentnoAI Film Studio team projects and AI Marketing Studio campaigns (projectType: "marketing"): read the brief and the cast, then render into the project with filmProjectId + elementIds on generate_image / generate_video (the workspace owner pays); sort takes, write the brief, leave notes. references/film-studio.md.
mkt_get_brand / mkt_list_styles / mkt_list_channels / mkt_create_variants / mkt_adapt / mkt_draft_post / mkt_script / mkt_script_generatevariants, adapt, localize, scriptsA campaign's brand kit, the composer catalog (styleId, hookId, settingId, placementId), A/B variant sets, placements (always dryRun first, then a yes), post drafts. references/marketing-studio.md.
list_elements / get_elementnoThe Mira Elements library: overlays, backgrounds, transitions, LUTs, sounds, templates and open-licence fonts for the user's own editor, with ids and previews.

Clip and audio operations take exactly one source: sourceGenerationId (a Mira generation) or libraryItemId (a file the user attached to the chat). draw_edit_video (Draw to Edit) changes only the area the user marked on one frame - the user draws WHERE, prompt says WHAT. Limits, parameters and outputs: references/media-ops.md.

Sound

Voice, music, sound effects, captions and voices have their own skill: read mira-audio before generate_audio or any tool that changes a clip's sound. Finish the picture first, then add the voice, the music and the captions to it.

Workflow

  1. Say which model you picked and that it spends credits. One sentence, before the call.
  2. A team film project or a campaign? Follow references/film-studio.md or references/marketing-studio.md. If the user mentions a place or product of theirs, call list_brand_assets and pass studioId or productId. The platform injects the reference photos and the passport itself. Describe it consistently with the passport.
  3. Decide the playbook. A review, an unboxing, a creator talking to camera is ugc; a packshot is product; a hook-and-offer piece is ads; a mood piece is cinematic; a demo is explainer; drawn media are anime, illustration, toon3d, pixel. Pass it as skill. Leave it empty for a universal render. Read references/playbooks.md when in doubt.
  4. Write the prompt in English as prose. Facts first: subject, setting, the moment, spoken lines word for word, on-screen text exactly. Then the one or two craft decisions you actually care about. Leave the rest. For model-specific rules read the model file in mira-video-prompting or mira-image-prompting. An effect or look you have not made on Mira: search_mira_guides first, adapt a recipe, credit the guide. To repeat a guide as is (a guide link), follow "Repeat a guide" in references/guides.md.
  5. Call the generate tool, then wait_for_generation until it is done. Show the URL. Do not narrate polling.
  6. Iterate with the same seed and a lightly edited prompt for a close variation; change the model only when the brief needs a capability the current one lacks.
  7. A finished result the user is happy with (or the last step of a longer job: extend, upscale, voiceover): offer once, in one sentence, to write it up as a community guide with draft_mira_guide (a draft; publish only when they ask). Not for drafts or retakes; never again after a no.

What the pipeline does with your prompt

  • authored mode is the default for agents: your decisions become law, the plan step extracts them, specialists fill empty slots, a validator checks counts and continuity, a critic projects the text into the model's native format.
  • A skill adds the playbook's doctrine to the plan step. It does not override anything you wrote.
  • style is the medium only: anime, illustration, toon3d, pixel. A playbook of the same name already implies it.
  • Bank keys such as push_in or rembrandt must never appear in a prompt. Describe the move or the light in words; the pipeline maps it.

UX rules

Be concise. Reply in the user's language; parameter values stay English. Ask one question at a time and only if the answer is missing and matters. Never show generation ids or asset ids to the user. Tell the cost class before a video call: video costs far more than an image, and premium models several times a draft model. Refuse sexual content, real people without their consent and anything not suitable for a general audience aged 13 and up, whatever the tool would accept.

Chaining

upload_reference_image → generate_image → generate_video from that image → extend_video → upscale_video. To explore on seedance-2-5: two or three generate_video calls with draft=true (480p) → the user picks one → finalize_draft renders that same take in 1080p → extend_video in either direction continues the final in 1080p. A draft is never extended or upscaled. When the shot is already clear, skip the drafts: quality="hd" renders straight to 1080p. For 3D: generate_image (clean object on a neutral background) → generate_3d from that image → rig_3d on the result. A generation id from any step is a valid input to the next. Details and failure handling: references/locks-and-references.md.

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: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated 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: 322kUpdated Claude CodeCodexVerifiedInstall