Mira Generate

---
name: mira-generate
description: >-
  Generate images and video on Mira AI through its MCP server: connect, pick a model, attach the
  user's AI Blogger, 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", "сгенерируй", "сделай видео", "сделай картинку" and
  a Mira MCP server is connected or can be. 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, dubbing or captions (mira-audio).
license: MIT
metadata:
  version: "0.1.0"
---

# 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

Server: `https://mcp.mybots.pro/mcp` (streamable HTTP). It signs in with the user's Mira account
through OAuth on the first call that spends credits; there is no API key. The setup page for each
client is [mira.mybots.pro/mcp](https://mira.mybots.pro/mcp). If `tools/list` works but a call
returns 401, the user has to finish the sign-in in the browser window the client opened.

## Tools

| Tool | Spends credits | Use it for |
|---|---|---|
| `list_models` | no | Valid model ids, ratios, durations, resolutions. Call once per session. |
| `list_skills` / `get_skill` | no | Playbook ids for the `skill` parameter and the knowledge skills, bundled with the server. |
| `list_brand_assets` | no | The user's AI Bloggers, locations and products with their ids. |
| `upload_reference_image` | no | A local image becomes a reference URL (base64, up to 8 MB). Only clients with file access can do this. |
| `create_upload` | no | A one-time PUT URL for files that do not fit base64: a clip for `referenceVideoUrls` or a GLB/FBX mesh for `rig_3d` (up to 200 MB). PUT the raw bytes, then use `value.url` or `value.id`. |
| `generate_image` | yes | Stills. Returns a generation id at once. |
| `generate_video` | yes | Clips. 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_generations` | no | The 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_generation` | no | Status and the result. `wait_for_generation` blocks up to ~50 s and may need several calls. |
| `extend_video` | yes | Native continuation of a clip the account already made. `direction="backward"` (seedance-2-5) adds what happens before the first frame. |
| `finalize_draft` | yes | Re-renders a 480p draft as the same take in 1080p, within 7 days of the draft. |
| `upscale_video` | yes | 720p to 1080p on a finished clip (Topaz). Not for drafts: finalize them instead. |
| `generate_3d` | yes | One object as a 3D model from text or up to four photos (fal.ai: `hunyuan-3.1`, `meshy-7`, `trellis-2`, `rodin-fast`). Result files (GLB plus other formats) come back in `files`. Playbook: `3d`. |
| `rig_3d` | yes | Humanoid skeleton with walk/run on a finished 3D generation (`generationId`) or a mesh uploaded through `create_upload` (`libraryItemId`), plus animation clips by preset id (Meshy). |
| `search_mira_guides` / `get_mira_guide` / `draft_mira_guide` / `list_my_mira_guides` / `update_mira_guide` / `publish_mira_guide` | no | Community guides at [mira.mybots.pro/guides](https://mira.mybots.pro/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 (media, galleries, comparisons, uploads, video embeds, the tools they used); edit their own guides; publish only when they explicitly ask. [references/guides.md](references/guides.md). |
| `list_direct_presets` / `direct_video` | `direct_video` yes | Direct 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](references/direct-presets.md). |
| `list_blogger_traits` / `create_blogger_draft` / `save_blogger` / `list_blogger_presets` / `blogger_motion` / `blogger_speak` / `set_blogger_voice` / `swap_object` | draft, save, motion, speak, swap | AI Bloggers, the user's persistent brand characters: build one from traits or words (a draft 8, the save 45 after a clear yes, with a voice), make it repeat a clip, take a person's place in one or speak a script in its voice; `swap_object` replaces one element of any clip from photos. `bloggerId` works on the generate, clip and audio tools. [references/ai-blogger.md](references/ai-blogger.md). |
| `list_miracle_presets` / `miracle_analyze` / `miracle` | `miracle` only | Miracle: in a clip of 3-30 s, replace the people (an AI Blogger or a photo), an outfit, a product, the place or a sign while every move, the camera and the sound stay. Analyse first (free: who is where, what can be swapped, the price), confirm, then run (one pass up to 30 s, or in parts of up to 10 s; 6-10 credits per second by engine, about 5-8 min). [references/miracle.md](references/miracle.md). |
| `list_effects` | no | Ready-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_video` | yes | The same clip with one thing changed: lighting, weather, time of day, backdrop, or a free restyle. Up to 15 s. |
| `reframe_video` / `remove_background` | yes | A 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). |
| `motion_control` / `recast_video` | yes | The character in a photo performs the motion of a clip; the person in a clip swapped for the character in a photo, performance and camera kept. Up to 30 s. |
| `lipsync_video` | yes | The mouth in a clip re-animated to a new track from `generate_audio` or an upload. Up to 60 s. |
| `marketing_brief` | yes | Script, shot list, a ready `generate_video` prompt, caption, hashtags and voiceover text for a product clip, returned in the same call. |
| `generate_audio` / `plan_music` | yes / no | Speech, multi-voice dialogue, sound effects and music; `plan_music` drafts a free section plan for music timed to the cut. |
| `list_voices` / `design_voice` / `save_voice` / `delete_voice` | design only | The platform's voices and the user's own; a new voice from a description, kept or removed. |
| `add_soundtrack` / `voiceover_video` / `add_captions` | yes | On a finished clip: music the model writes by watching it; a spoken script over the ducked original; burned subtitles plus an SRT. |
| `translate_video` / `change_voice` | yes | The clip's speech dubbed into another language, lip sync optional (up to 180 s); the same performance in another voice (up to 120 s). |
| `isolate_voice` / `separate_stems` | yes | Speech cleaned of noise, music and echo; a track or a clip's sound split into 2 or 6 stems. |
| `transcribe_video` / `analyze_transcript` | yes / partly | A word-timed transcript and SRT with speakers (up to 30 min); silence cuts and multicam splits (free), the best short clips, SEO, chapters. |
| `estimate_cost` | no | Credits 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. |
| `get_credit_balance` | no | What the user can afford. |
| `blender_status` | no | Whether the user's Blender (Mira add-on, "Connect agents" on) is reachable; version, file, frame range, selection. Call it before any other `blender_*` tool. |
| `blender_get_scene` / `blender_screenshot` | no | See what the user sees: objects with transforms and dimensions, cameras, lights; a viewport or camera capture returned inline. |
| `blender_execute_python` | no | Run bpy code in the user's Blender and get its output back: build blockouts, place lights and cameras, keyframe. Small steps, check with a screenshot. |
| `blender_import_generation` | no | Put a finished `generate_3d` / `rig_3d` result into the user's scene at the 3D cursor. |
| `ask_mira` / `mira_history` / `list_mira_dialogs` | yes (2 per reply) | Talk to Mira, the platform's own creative agent, in the user's dialogs: a brief, a look, a scene idea, or the whole task — she starts generations herself and they come back as cards. |
| `blender_playblast` | no | Render the user's viewport animation to a 720p clip, uploaded to their library; its URL goes into `referenceVideoUrls`. Bind it to camera, framing and rough placement, not the figures' exact gestures; Clay and frozen gestures by default (`look`, `freezeFigures`); build people with `stand_in()` (mira-blender-scene). Keep ranges under 30 s. |
| `adobe_status` | no | Whether the user's After Effects or Premiere Pro (Mira panel, bridge on) is reachable; app, version, open project, active composition, what the panel is busy with, running renders. Call it before any other `adobe_*` tool. |
| `adobe_get_project` | no | Compositions or sequences with size, fps and duration, the active one with its layers, the playhead. |
| `adobe_execute_script` | no | Run ExtendScript in the user's app (up to 40 000 characters, 180 s): build and edit compositions, layers, keyframes, effects, imports. Returns the `result` variable and what the script printed (`$.writeln`, `print()`). |
| `adobe_export_frame` / `adobe_contact_sheet` | no | One frame (at the playhead or at `time`) as a PNG, or up to 12 frames at given times on one image with timecodes: check shots and transitions, or use a frame as a reference image. |
| `adobe_beat_markers` | no | Tempo, beat grid, downbeats, drops and sections of the music layer, written as `mira:*` markers; the grid comes back as `grid_offset + k * beat_seconds`. |
| `adobe_render` / `adobe_render_status` / `adobe_render_cancel` | no | An MP4 of a composition rendered on the user's machine (background aerender for a saved project), with a platform preset check; optionally uploaded to the Mira library. |
| `adobe_import_generation` | no | A finished Mira generation placed into the project or the active composition. |
| `list_elements` / `get_element` | no | The Mira Elements library for After Effects and Premiere: overlays, backgrounds, transitions, LUTs, sounds, templates, open-licence fonts, with ids, previews and apply recipes. |
| `adobe_place_element` / `adobe_install_fonts` | no | An element placed into the active composition at a time (the panel downloads it and installs the fonts a template needs; templates take texts, controls and `media` for their placeholders; luma wipes reveal the incoming layer), or fonts installed on their own. |
| `adobe_command` | no | Typed commands when one fits better than raw script: list_layers, set_keyframes with ease, add/read markers, precompose, set_parent, set_time, set_work_area, open_comp, create_composition, add text or solid layer, apply effect, set track matte. |

Clip and audio operations take exactly one source: `sourceGenerationId` (a Mira generation) or `libraryItemId` (a file sent through `create_upload`). Limits, parameters and outputs: [references/media-ops.md](references/media-ops.md).

## Sound

Voice, music, sound effects, dubbing, 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.

## Blender

When `blender_status` says connected, the user's own Blender is your canvas: read the scene, change it with `blender_execute_python` in small steps, look at `blender_screenshot` after each, import generated assets with `blender_import_generation`, and finish a shot with `blender_playblast` → `generate_video`. Code runs on the user's machine: stay inside Blender's data, never touch files outside it, and say what each step does before running it. Not connected → tell the user how to switch on "Connect agents" in the Mira tab and stop; do not retry in a loop.

## After Effects

When `adobe_status` says connected, the user's After Effects (or Premiere Pro) is your compositor: read the project with `adobe_get_project`, change it with `adobe_execute_script` in small steps, look at `adobe_contact_sheet` after each, place finished generations with `adobe_import_generation`, and render with `adobe_render`. Building a whole motion piece (plates, overlays, templates, type, transitions, sound, grade, render) is the job of mira-after-effects; read it first. Not connected → tell the user how to switch the bridge on in the Mira panel and stop; do not retry in a loop.

## Workflow

1. Say which model you picked and that it spends credits. One sentence, before the call.
2. If the user mentions a person, place or product of theirs, call `list_brand_assets` and pass
   `bloggerId` (their AI Blogger), `studioId` or `productId`. The platform injects the reference photos and the
   passport itself. Describe the character consistently with the passport; never redraw the face.
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](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, the `repeat_guide` prompt), follow "Repeat a guide" in [references/guides.md](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, blockout to clip): 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 price class before a video call: video costs far more than an image, and premium
models several times a draft model.

## 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, about 18 credits per 5 s) → 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 (about 100 credits per 5 s instead of 41 for 720p). 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.
From a DCC tool such as Blender: export the selection as GLB → `create_upload` (kind `model3d`) → `rig_3d`
with `libraryItemId`; a viewport playblast → `create_upload` (kind `video`) → `generate_video` with that URL
in `referenceVideoUrls`. Details and failure handling:
[references/locks-and-references.md](references/locks-and-references.md).