Creative Studio Playbooks

Product Teaser Video

• Good for: social media clips, pitch visuals, concept demos
• Best setup: text description or reference image
• Availability: Video Lab — available on every plan; billed per clip by duration from your monthly allowance
• Example prompt:

Product teaser

Create a 5-second product teaser video. Cinematic lighting, smooth camera movement, modern tech aesthetic. The scene should convey innovation and speed without any text overlay.

Animate a Still Image

• Good for: turning mockups, photos, or illustrations into motion
• Best setup: upload the image first, then describe the animation
• Availability: Video Lab — available on every plan; billed per clip by duration from your monthly allowance
• Example prompt:

Animate image

Animate this product photo: slow zoom in, subtle parallax, soft ambient lighting shift from cool to warm. 5 seconds, cinematic quality.

Cinematic Multi-Provider Video — Seedance + xAI + MiniMax + Runway

• Good for: native-audio dialogue/SFX clips, keyframe morphs, reference-to-video character consistency, slow-orbit portraits, 20–30s Seedance shots
• Best setup: to animate a still, upload the exact opening shot and name the camera motion in plain English. To put a character into a new scene, upload their photo as a reference instead and describe the scene — ask for "reference images", not "animate this image". Anchoring a portrait as the first frame forces the camera to wander away from it.
• Availability: Video Lab — available on every plan; billed per clip by duration from your monthly allowance
• Pick a provider — name it in your prompt, or say nothing and let Alfrada OS choose:
  • Let Alfrada OS choose (default): it tries Seedance 2.5 first, then xAI's Grok Imagine, then MiniMax Hailuo 3, then Runway — one attempt each — and tells you which one made the clip. If you attached reference photos, MiniMax goes first instead (Seedance declines reference stills of real faces).
  • Seedance 2.5 for the longest shots — 4–30 seconds with native audio at 480p or 720p, and zero prompt retention.
  • Grok Imagine (xAI) for cheaper, faster native-audio drafts — 1–15 seconds at 480p or 720p; prompts are retained for 30 days.
  • MiniMax Hailuo 3 for native-audio clips up to 2K — from text, an image, or up to 9 reference photos.
  • Runway for text-to-video, first-to-last keyframe morphs, and explicit control over which model runs.
• Audio rule: Alfrada OS reports whether the finished MP4 carries real audio — if it does, have the old audio stripped (or explicitly confirm the overlay) before layering narration on top. Care about prompt or image retention? Ask which provider made the clip and what it keeps; every result carries its retention policy.
• Example prompt:

Seedance 2.5 orbit

Animate this portrait with a slow orbit around the subject, cinematic depth of field, golden-hour rim light. Use Seedance 2.5 with native audio. 12 seconds.

Grok Imagine native-audio scene

Create a 6-second café scene with a barista calling out 'Latte for Maya!', natural café ambience, and cup sounds. Use Grok Imagine and tell me whether the final MP4 has real audio.

Architecture Diagram

• Good for: system design, process flows, org charts
• Best setup: describe the structure in plain language
• Availability: Mermaid Diagrams — free on all plans
• Example prompt:

Architecture diagram

Draw a system architecture diagram showing our frontend (React), API gateway, three microservices (auth, billing, notifications), a PostgreSQL database, and a Redis cache. Show the data flow direction between components.

Onboarding Funnel Flowchart

• Good for: product flows, decision trees, process documentation
• Best setup: describe the steps and decision points
• Availability: Mermaid Diagrams — free on all plans
• Example prompt:

Funnel flowchart

Draw a flowchart of our user onboarding funnel: signup → email verification → plan selection → first session → activation milestone. Include a decision branch for users who don't verify within 48 hours.

Marketing Hero Image

• Good for: landing pages, blog headers, social cards
• Best setup: describe the visual, mood, and colour palette
• Availability: Image Lab — available on every plan; each image draws from your monthly allowance
• Pick a provider — name it in your prompt if you have a preference:
  • The default (Seedream 5.0 Pro) handles general image generation, lifelike commercial scenes, precise product edits, and compositions that combine up to 14 reference images. It returns one 1K/2K image per call.
  • Ask for Gemini when you want 4K or several variations from a single request — those route to it automatically anyway.
  • Ask for Grok Imagine when the image needs clean, readable text — posters, ads, packaging, menus — or brand variants, named locations, and public-figure creative.
  • Grok Imagine also steps in automatically when the default renderer declines a request.
• Copyright/IP rule: xAI can render named entities, but that is not legal clearance. For third-party franchises or characters, use descriptive homage language unless you own or have rights.
• Example prompt:

Hero image

Generate a hero image for our landing page. Minimal, modern design with a teal and white palette. Abstract geometric shapes suggesting connection and intelligence. No text. 1200×630 aspect ratio.

Grok Imagine launch poster

Create a 16:9 2K premium launch poster for Alfrada OS with Grok Imagine: dark graphite background, platinum text reading 'Alfrada Creative Stack', subtle soft-gold accents, cinematic lighting, clean typography.

Seedream product campaign edit

Use Seedream to edit out/product.jpg. Preserve the exact product shape, label, reflections, and camera angle. Change only the environment to a premium midnight-blue campaign set with soft brass accents, at 2K.

Need the finished image dropped into a branded template or resized for every channel? Hand it to Canva — see the Canva playbook.

Custom Meme

• Good for: internal comms, social posts, team culture
• Best setup: pick the format and write the captions
• Availability: Meme Generator — free on all plans
• Example prompt:

Custom meme

Make a Drake meme. Top panel: 'Spending 3 hours formatting the report'. Bottom panel: 'Asking Alfred to do it in 90 seconds'.

Background Music for Video

• Good for: product demos, presentations with audio, podcast intros
• Best setup: describe the genre, mood, and duration
• Availability: Music Lab — available on every plan; billed per track from your monthly allowance
• Example prompt:

Background music

Create a 30-second upbeat instrumental track for a product demo video. Modern electronic with light piano, building energy. No vocals. Suitable for corporate but not boring.

Video Editing — Highlight Reel

• Good for: trimming recordings, adding subtitles, compilations
• Best setup: upload your video file, describe the edits
• Availability: Video Editor — available on every plan; rates per call are published separately
• Example prompt:

Highlight reel

Ingest this 20-minute meeting recording. Find the three most important segments based on the transcript, cut the dead air out of each, and build a 3-minute highlight reel with crossfade transitions and burned-in subtitles. Reframe it to 9:16 around the speaker — don't letterbox.

Restyle a Video

• Good for: turning real footage into an animated or stylized version while keeping the soundtrack
• Best setup: upload the clip first, then name the style in plain words
• Availability: Video Editor — available on every plan; rates per call are published separately
• How it works: Alfrada OS breaks the clip into its key frames, restyles each frame in the style you asked for, then reassembles the video with the original soundtrack laid back over it. Every restyle also attaches an automatic frame-drift consistency verdict — ask "what did the consistency check say?" when it finishes (see Character Consistency QA below).
• If one frame comes out wrong, you don't redo the whole clip — ask Alfrada OS to redo just that frame with a tweaked style note.
• Example prompt:

Watercolour restyle

Restyle my clip as watercolour animation and keep the original soundtrack.

Work on Your Own Footage

• Good for: highlight clips from YouTube talks, relighting or cleaning up real footage, 4K upscales, making a character act out your performance
• Best setup: upload the footage — or give Alfrada OS a YouTube link and let it fetch just the part you need
• Availability: Video Lab — available on every plan; billed per clip by duration from your monthly allowance

Video Lab is not generate-only — point it at real footage. Four things people do with it:

Pull in a YouTube source. Alfrada OS downloads the video into your session and ingests only the window you name, so you never pay to analyse a 40-minute talk to work on 90 seconds of it. Sources up to 20 minutes long by default (if you only want a summary, a transcript is faster and cheaper); YouTube downloading can be switched off per deployment.

YouTube rip to vertical clip

Download this YouTube talk at 720p, ingest just 12:00–14:30, and cut the three strongest moments into a vertical clip cropped around the speaker, with word-by-word captions and a lower-third with their name for the first five seconds.

Prompt-edit real footage (Runway Aleph 2). Describe the change and the clip is re-rendered with it — relighting, removing objects, changing weather — keeping the original framing and length. Works on clips up to 30 seconds; for longer footage, extract the segment first.

Aleph 2 relight

Make clip.mp4 golden hour and remove the parked cars.

Upscale to 4K (Magnific). A creative upscaler that lifts a clip as high as 4K.

Magnific 4K upscale

Upscale out/final.mp4 to 4K.

Transfer a performance (Runway Act-Two). Film yourself acting the lines, and a character still delivers your performance — expressions, timing, and optionally body gestures.

Act-Two performance transfer

Animate character.png with the performance in take3.mp4 — expressive, keep body gestures.

Footage you upload or have Alfrada OS fetch is yours to work with — it transforms it as directed. If a provider's moderation rejects a request, Alfrada OS reports it rather than working around it.

Stock Footage & Photos — Pexels Stock

• Good for: b-roll, intro clips, hero photos, filling gaps a generator shouldn't
• Best setup: describe the subject, orientation, and mood; the files download straight into your session
• Availability: Pexels Stock — free on all plans
• Everything is royalty-free with photographer attribution included, and downloaded files land in your session by name — so the Video Editor, Presentation Create, and Image Lab can use them immediately.
• Example prompt:

Stock clips for an intro

Find three licensed stock clips of a city skyline at dusk on Pexels for the intro — landscape orientation, moody lighting.

Character Consistency QA — Generated Images & Restyled Video

• Good for: verifying a generated character stayed the same across images, catching style drift in restyled video, same-person / same-object checks
• Best setup: have the images in your session (generated by Image Lab, restyled video frames, or uploads); say what should match — the person, the object, or the overall style
• Availability: Image Consistency — free on all plans, runs locally (images never leave the deployment)
• Recipe:
  1. Generate or upload the images (Image Lab, a Video Editor restyle, or a file upload).
  2. Ask for a consistency check across 2–4 of them, naming what should match — you get a verdict, a confidence level, and the specific differences, plus a saved report in your session files.
  3. If the verdict flags drift, have Image Lab (or the restyle) redo the flagged image, using the reported differences as the correction.
• Video restyles check themselves: every restyle automatically attaches an advisory frame-drift verdict — ask "what did the consistency check say?" afterwards.
• Example prompt:

Character consistency check

I generated four shots of our mascot in out/. Compare them — same character in all four? Focus on hair, outfit, and facial features. If any drifted, fix the odd one out with Image Lab.

Restyle with QA

Restyle my clip as watercolour animation, then tell me what the consistency check found — if the palette drifted anywhere, re-run just those frames.

Under the hood

• Video Lab providers — the provider parameter accepts auto (default chain Seedance 2.5 → xAI → MiniMax → Runway; MiniMax leads when reference_images are present), seedance (ByteDance Seedance 2.5 via OpenRouter — distinct from Runway's seedance2 model, which is reached via runway), xai (Grok Imagine), minimax (Hailuo 3), and runway (Gen-4 / Veo / seedance2). Reference-image caps: Seedance 30, MiniMax 9, xAI 7. Every result carries has_audio and a data_policy retention stamp.
• Own-footage operations (Video Lab): edit_video (Runway Aleph 2 prompt edits), upscale_video (Magnific creative upscaler; upscale_resolution, auto = 2K), and character_performance (Runway Act-Two). All three accept inputs up to 30 seconds / ~16 MB encoded.
• YouTube sourcing: the youtube tool's download action (480p/720p/1080p, or audio_only for m4a), capped at 20 minutes per rip. Windowed ingest is video_editor ingest with time_range.
• Restyle pipeline: restyle_video = restyle_frames → compose_from_frames → original audio overlaid; fix a single frame via restyle_frames with frames=[t]. The automatic drift verdict comes from image_consistency (subjects: person, object, style), which also runs standalone.
• Image Lab providers: default is ByteDance Seedream 5.0 Pro via OpenRouter (one 1K/2K output, up to 14 references); gemini for true 4K or several images per call (4K and n > 1 route there on their own); xai routes to Grok Imagine Image Quality via OpenRouter, which is also the automatic retry when the default renderer refuses.
• Stock: pexels_media (search_photos, search_videos, curated_photos, popular_videos, plus per-file download operations).
• Billing: Video Lab bills per clip by duration tier per provider; Image Lab and Music Lab bill per call.