pruna-generative-pipeline

Use when the user is unsure which production workflow fits, needs a recipe menu for multi-step media projects, or wants chained image, video, and audio with approval gates.

Pruna generative pipeline (scenario hub)

When NOT to use

• Single-shot generation with minimal intake → pruna-run
• One talking-head clip only → avatar-single-scene

This skill is the menu of proven chains over the atomic tools in tools/image and tools/video. For HTTP details see pruna-api.md.

Marketing and demo sites change layout and messaging often; this repo is the source of truth for what agents should run. When humans find a compelling flow elsewhere, encode it here (extend a recipe, add a row to the idea map, or add a dedicated workflow skill) after checking it against docs.api.pruna.ai—do not send agents to browse live demos as part of execution.

Tone: Explain plans to the user like a producer—clear scene order, what generates first, and what they should hear or see—especially whenever voice_script or narration exists. Avoid stiff jargon in client-facing script drafts unless they ask for it.

Quick reference

Resource	Path
Full recipe steps (A–R)	recipe-catalog.md
Feedback gates	staged-generation-gate.md · requesting-generation-feedback
Parallel execution	parallel-execution.md
Pruna HTTP	pruna-api.md

Intake: pick a scenario first

Ask the user which recipe fits (or hybrid). Capture answers before any POST /v1/predictions:

Topic	Questions
Recipe	Which row in the routing table below (or combination)?
Deliverable	Stills only, video clip(s), avatars, or mixed?
Aspect / resolution	e.g. 9:16 vs 16:9; video default 720p + 24 fps unless user wants 1080p/48 or draft preview?
References	Files to upload to /v1/files (URLs only in predictions)?
Style lock	One style bible sentence for technical consistency (aspect, no text)? Per-scene visual_style_tag for deliberate variety (anime, clay, Disney 3D, cyberpunk, etc.) — visual-variety-bible.md
Character / scenes	Character sheet (age, look, realism, personality)? Per-scene angle, setting, emotion, lighting deltas? Locked seed? Cast diversity on launch reels?
Voice / cast	For talking heads: one Pruna voice per character; natural human voice_script; realistic conversational voice_prompt; unique video_prompt per scene.
Scenes	If multiple clips: scene table with distinct backgrounds/angles (or defer to narrated-multi-scene / avatar-multi-scene)?

Confirmation and execution (all recipes)

0. Generation diversity — ritual seed + diversify axes before the first generation in a session.

1. Draft any scripts, beats, or prompts that the user must approve in natural, human language (spoken wording for VO; clear intent per scene otherwise).

2. Pause and obtain explicit confirmation (“approve”, “go”, “run it”) before the first upload or POST /v1/predictions. If the user revises copy or cast, re-confirm when the change matters for cost or outcome.

3. Staged execution — after plan approval: stills only → user approves → audio prep (TTS / song) when applicable → user approves → video → user approves clips → assembly and final audio (bed). See staged-generation-gate.md and requesting-generation-feedback.

4. Write a runnable generation package—script or phased API steps—that matches the approved plan exactly. Use async parallel fan-out within each phase (see parallel-execution.md); avoid serial scene-by-scene execution when lanes are independent.

5. Execute that package when possible (PRUNA_API_KEY, network). Prefer subagents per scene lane (still pipeline or avatar job) when the host supports parallel agents. If execution is not possible here, deliver the same artifact for local runs.

Deep avatar workflows already spell out cast ledgers, hero reuse, and read-throughs in avatar-multi-scene and avatar-single-scene; defer to those when recipe G or H involves p-video-avatar.

Routing table

Recipe	You get	Primary models	Deep workflow (if any)
A — Style-locked mood board	N stills, same world	p-image → optional p-image-edit	Stay in this skill
B — Hero + variants	One anchor + edits	p-image → p-image-edit	—
C — Print / pixel rescue	Higher-res master	p-image-upscale → optional p-image-edit	—
D — Animate a plate	One motion clip from a still	p-image or upload → p-video (I2V)	image-to-video
E — Audio-led cut	Video length follows VO/music	upload audio → p-video	image-to-video
F — Draft → final video	Cheap preview then hi-fi	p-video (draft: true) then rerun draft: false with locked seed/prompt	narrated-multi-scene for many beats
G — Talking head	Portrait + speech	p-image / p-image-edit → p-video-avatar	avatar-single-scene or avatar-multi-scene
H — Social hook stack	Short vertical beats	Several p-video and/or avatars	Multi-scene skills above; batch avatar/product beats via tool skills
M — Motion-transfer showcase	Same motion, new subject + slider before/after	motion .mp4 + still → p-video-animate → slider compare MP4	avatar-multi-scene (animate rows)
N — In-video replacement showcase	Swap people/products in footage + slider before/after	dynamic p-image refs → optional p-video-avatar source → p-video-replace → slider compare MP4	p-video-replace · avatar-multi-scene
O — AI music video	Full song + lyric-synced video	lyrics → music-2.5 → cut map → p-video-avatar + p-video → assembly; hero + p-image-edit when one singer throughout	music-video
P — Narrated story film	Multi-scene B-roll + VO (+ optional bed)	hero → p-image-edit start/end stills → p-video with image + last_frame_image chain → gemini-3.1-flash-tts → concat + mux ± stable-audio-2.5	narrated-multi-scene · audio-post-production.md
Q — Visual transition reel	Multi-scene motion between two stills per beat (no VO)	p-image hero → p-image-edit start/end stills → p-video pair (duration) → selective frame chain → concat	visual-transition-reel · scene-anchor-pair.md
R — Educational explainer	Narrator VO + expert/character dialogue	p-image hero → p-image-edit stills → narrator triple + character avatar → concat ± bed	interactive-explainer · interactive-explainer-scenes.md
S — Illustrated story reel	Still-image story with VO or music (no video API); 9:16, 16:9, or 1:1	p-image hero → p-image-edit beats → Gemini TTS or Stable Audio → ffmpeg Ken Burns + mux	illustrated-story-reel
T — Virtual try-on launch	Fashion vertical showcase reel	p-image person + garment → p-image-try-on → p-video-avatar / p-video / still slider → Gemini TTS → concat + Stable Audio bed	p-image-try-on · recipe-catalog.md (Recipe T)

Handoff rules (all recipes)

• Use Pruna file URLs only (POST /v1/files → urls.get) in images, image, audio.
• Async + parallel by default for p-video, p-video-avatar, and batch stills—create all jobs in the current phase together, poll all get_url until done (parallel-execution.md). Use subagents for independent scene lanes after confirmation.
• Do not chain the next step until the previous URL is valid and (for portraits) checklist-approved when using recipe G.
• Character continuity: One approved source / hero URL per recurring subject; branch new looks and styles with p-image-edit from that URL (plus style bible), not unrelated fresh p-image identity pulls—unless the user resets the character.
• Dynamic scenes: Multi-scene avatar pieces must vary camera angle, background/setting, lighting, and visual_style_tag per beat—no repetitive office-only stacks unless requested. Run visual-variety-bible.md checklist on launch reels.
• Photoreal stills: Hero via p-image-ideogram (high-quality fast) or p-image (fastest drafts) with documentary/photoreal prompt → slop gate; every p-image-edit still → slop gate before p-video-avatar. Persona bar: realistic-persona-showcase.md.
• Dynamic avatar motion: unique video_prompt per scene — not duplicated medium-CU dolly on every row.
• Identity lock: Record approved hero plate URL at hero generation; branch with p-image-edit for same character. Log ritual_seed (SSoT) for prompt planning — do not pass to API unless user sets api_seed.
• Same voice for the same role: When a character speaks in more than one clip, keep voice (and usually voice_language) identical across those clips.
• Human delivery: voice_script = speakable dialogue; voice_prompt = realistic performance direction (never marketing copy).

Recipes (one-liners)

Full intake, steps, and shine lines for every letter: recipe-catalog.md.

Recipe	One-liner
A	N style-locked p-image panels ± p-image-edit drift fix
B	One hero → p-image-edit variants
C	p-image-upscale → p-image-edit
D	Still → p-video I2V (image-to-video)
E	Upload audio → p-video (duration follows audio)
F	draft: true previews → locked draft: false finals
G	Hero + edits → p-video-avatar (avatar skills)
H	2–5 vertical avatar beats, distinct worlds per scene
M	Motion template + still → p-video-animate + slider
N	Replace in footage + slider (p-video-replace)
O	Lyrics → song → synced clips (music-video)
P	Scene anchor triple + VO (narrated-multi-scene)
Q	Start/end still pairs, no VO (visual-transition-reel)
R	Narrator + character dialogue (interactive-explainer)
S	Still story + VO or music (illustrated-story-reel)
T	Try-on launch reel — six fashion verticals (p-image-try-on)

Atomic tool index

Model	Skill
p-image	p-image — good quality, extremely fast (Pruna P-API)
p-image-ideogram	p-image-ideogram — high-quality fast photoreal (Replicate)
p-image-edit	p-image-edit
p-image-upscale	p-image-upscale
p-image-try-on	p-image-try-on
p-video	p-video
p-video-avatar	p-video-avatar
p-video-animate	p-video-animate
p-video-replace	p-video-replace
stable-audio-2.5 (Replicate)	stable-audio-2.5 — launch reel bed under VO
gemini-3.1-flash-tts (Replicate)	gemini-3.1-flash-tts — narration / voiceover
music-2.5 (Replicate)	music-2.5 — full songs with vocals

Audio layering guide: audio-post-production.md · Scene anchor triple: scene-anchor-triple.md

Product-positioning idea map: recipe-catalog.md More ideas.