Higgsfield Pipeline

Higgsfield Production Pipeline

QUICK FACTS

Generated-checked block (scripts/build_index.py verifies anchors). Read the linked sections for full context — these lines are routing aids, not the rules themselves.

• 8-stage Master Chain: Popcorn → Seedream/Soul → Animate → Recast → Lipsync → Vibe Motion → Upscale → Assemble; most good short-form uses 3–5 stages →
• Lock 9 project fields before touching any tool; "what must stay consistent" is the load-bearing one →
• One job per scene — six scene purposes; a good scene prompt answers six questions →
• 7 reusable prompt-module types: character identity, camera, lighting, style, motion, negative prompt, continuity →
• 80% rule: keep what worked, fix only the mistake; every diagnosed failure becomes a new negative rule →
• Build in 8 passes: Concept → Project script → Scene breakdown → Shot list → Image prompts → Video prompts → Review → Fix →
• Use the EXACT same character description (copy-paste) in every Popcorn prompt — continuity without Soul ID →
• Seedream edits the image, not the video — always edit the Hero Frame before animating, never after →
• Model by scene type: Seedance 2.0 / Minimax Hailuo 2.3 for stunts/epic ("one continuous shot, no cuts"), Kling 2.6 portraits, Seedance quiet interiors →
• Recast swaps identity while preserving motion, camera, and lighting; the "prompt" is the reference image you upload →
• Audio routing: existing video + speech → Lipsync Studio; new content with audio → Kling 3.0; talking head → Seedance 1.5 Pro (start image + native audio) or Kling Avatars 2.0 (not in the API catalog, 2026-09-26 — verify in the live UI) →
• Higgsfield has no native timeline editor — assemble in DaVinci Resolve / Premiere / CapCut →
• Pipeline E hard rules: 15-second cap per scene, one generation per style, feed the previous scene's video as continuity reference →
• Soul Cinema keyframes: deliberately short 5–15 word prompts with enhancer ON — long prompts starve the enhancer →
• Never describe character age in Seedance prompts; >15s per scene degrades prompt adherence — split the scene →
• No extend button: attach the accepted clip as a video reference + open with "The scene continues." — and match the source's resolution AND duration →
• Cap seamless extension chains at 2 (hard ceiling 3); re-anchor from ORIGINAL canonical refs, break chains with B-roll →
• An attached source clip carries the state — prompt only the delta; motion vectors, camera-move phase, audio phase stay in prose when handing off from a still frame →
• End extension prompts on a camera-angle change so the join reads as coverage; plan transitions ahead (last-channel-on-TV trick) →
• Draw a top-down schema when 2+ characters, a key prop placement, or complex camera geometry — prompt in absolute terms ("A 2m from B") →
• Never animate a "good enough" image; if the character looks wrong in the Hero Frame, Recast is the fix — not the animation prompt →
• [FIELD] The edit is a loop with a declared exit: assembly → rough cut → generation supervision (re-generate broken shots here, and only here) → fine cut → picture lock, after which there are no new generations; colour's first job is unifying each generation's baked-in grade →

The Core Insight

Every Higgsfield tool is strong individually. The real power is when you chain them. A professional result almost always involves at least 3 tools in sequence. This skill documents the key chains and how to prompt for each stage.

Building Complete AI Projects — The 10-Step Methodology

The Core Insight above answers why you chain tools. The next question is how to plan the project before any tool is touched. Most people skip planning and start prompting straight from a creative impulse — a generation here, a clip there, hoping the pieces will connect. They don't. A project that's planned at the script and bible level lands consistently; a project assembled prompt-by-prompt drifts on character, style, and continuity within three or four generations.

This section documents a 10-step methodology for building complete AI projects — from idea through finished sequence — that sits upstream of the Master Production Chain below. Use these 10 Steps as a planning discipline; use Pipelines A-E to execute the tool-level chain once your project plan is locked. The closing § Simple Workflow gives the imperative-action version of the same 10 Steps if you want the execution recipe before the principles.

Step 01 — Start With the Project, Not the Prompt

Before writing a single prompt, define what you are building. The AI cannot read your intent; if you skip this step the model fills in the gaps with whatever the prompt happens to suggest, and the project drifts within the first few generations.

Lock these nine fields before any tool is touched:

• Project type
• Main subject
• Visual style
• Scene goal
• Audience
• Mood
• Length
• Setting
• What must stay consistent

The last field is the load-bearing one. If you don't know what must stay consistent across the project, every other field can be locked and the result will still feel disjointed.

Step 02 — Build a Master Script

A Master Script is the project blueprint — the single document that keeps every scene connected to the same goal. It does not have to be a Hollywood screenplay; it just needs to explain the full idea clearly.

Master Script disambiguation. "Master Script" in this context means the structured project document, not a screenplay. The source explicitly notes "it does not have to be a Hollywood screenplay; it just needs to explain the full idea clearly." This is distinct from Hollywood screenplay craft (slug lines, action lines, character cues, transitions) — that's a different discipline and is not in this skill's scope.

The Master Script should include:

• The story or concept
• The scene order
• The main characters or products
• The visual style
• The camera rules
• The setting
• The emotional tone
• The continuity rules
• What should never happen

Without a Master Script, each prompt becomes its own random island. With one, every prompt belongs to the same project — characters carry between scenes, the visual style holds, and the negative rules ("what should never happen") stay out of the output.

Step 03 — Use GPT as Your Creative Assistant

Do not use GPT like a magic button. Use it like a creative team member. GPT plays five distinct roles in the project pipeline:

• Concept Builder — turns a rough idea into a structured concept
• Script Writer — drafts the Master Script from the concept
• Shot List Planner — breaks the script into scene-by-scene shot lists
• Prompt Refiner — turns each shot into image or video prompt text
• Problem Solver — diagnoses failed generations and rewrites the offending prompt

The key is staging. Do not ask GPT to do everything at once. Ask for the project structure first. Then the scenes. Then the prompts. Then the revisions. The output quality at every stage depends on the input quality of the prior stage; cascading the work in order gives you control over each handoff.

In-platform alternative. Higgsfield has an in-platform GPT-5 copilot (Higgsfield Assist at higgsfield.ai/chat) trained specifically on Higgsfield's tools and workflows. For quick in-platform prompt generation and platform-navigation questions, see ../higgsfield-assist/SKILL.md. For upstream project staging across multiple sessions, an external GPT (Claude, ChatGPT) running this 10-step methodology is the better fit because state carries across the full project, not just the current platform session.

Step 04 — Separate the Script From the Prompt

The script is what happens. The prompt is how the AI shows it. These are two different artifacts that serve two different purposes — keep them in separate documents, and don't let prompt-shaped instructions leak into the script.

A script line might read:

"She walks into the room and realizes something is wrong."

The same beat, as a prompt, reads:

"A woman slowly enters a dim apartment from the hallway. Camera stays in front of her at chest height, tracking backward as she steps forward. Her eyes move toward an overturned chair and broken lamp. Warm hallway light behind her, cold blue window light across the room. Her face tightens as she realizes something is wrong."

The prompt decomposes the beat into four production-instruction categories:

• Camera Notes — wide shot, eye-level, shallow depth of field, focus on subject's expression
• Lighting Direction — dim, cinematic, cool blue ambient from window, single warm lamp in corner
• Subject Movement — slow, hesitant walk, pauses mid-step, tense posture, hand to mouth
• Environment Detail — cluttered, atmospheric, dust motes, shadows in corners, sense of foreboding

The script gives story. The prompt gives visual instructions. You need both, in separate documents, with the script feeding the prompt rather than collapsing into it.

Separation rule axis distinction. This is the project-level separation — script (story) from prompt (AI instruction). For the prompt-level separation within a single shot — Identity (who is in frame) from Motion (what they do and how the camera moves) — see ../higgsfield-prompt/SKILL.md § Identity vs. Motion Separation Rule and ../higgsfield-soul/SKILL.md § Identity vs. Motion Separation. Different axes of the same overall composition problem; both apply.

Step 05 — Create a Project Bible

A Project Bible is a single document that locks in the rules of the project — character appearance, environment, style, what must stay consistent, what must never appear. It is especially important for any project with multiple scenes, because consistency comes from repetition and clear rules, not from the AI inferring intent.

The Project Bible should lock in:

• Character appearance
• Wardrobe
• Hair
• Color palette
• Location
• Lighting style
• Camera style
• Tone
• Props
• Negative rules
• Continuity rules

Three rules of thumb make a Project Bible work in practice:

• If your character is supposed to look the same every time, describe them the same way every time.
• If your product is the focus, make sure the product stays the focus.
• If your scene takes place in one location, do not let the AI redesign the room every shot.

Project Bible as upstream source-of-truth. The Project Bible lives in your own files alongside the Master Script — it's the document you maintain across the project. Two common downstream realizations of the Bible inside Higgsfield: (i) the Character section feeds into a Soul ID character sheet, see ../higgsfield-soul/SKILL.md § Character Sheet Creation for the multi-angle reference approach; (ii) Characters, Locations, and Props feed into Cinema Studio's @ Elements system, see ../higgsfield-cinema/SKILL.md § Elements System for the @CharacterName / @LocationName / @PropName workflow. The Bible is the upstream artifact; the Soul ID sheet and Cinema Studio Elements are tool-side realizations of it.

Step 06 — Give Every Scene One Job

Every scene should have one clear purpose. When you try to make one scene do too much, the AI starts blending details — messy motion, wrong angles, extra people, broken hands, missing objects, confusing results. Six scene purposes cover most project work:

• Introduce the character
• Show the location
• Build tension
• Reveal the product
• Create emotion
• Deliver the action

A good scene prompt answers six questions:

• Who is in the shot?
• Where is the camera?
• What is the subject doing?
• What should the viewer notice first?
• What must stay visible?
• What should not happen?

Clarity wins.

Per-shot action-count rule. Step 06 addresses the narrative purpose of a scene (which may span multiple shots). For the per-shot action-count rule — one primary action per clip, with one or two secondary actions max — see ../higgsfield-prompt/SKILL.md § One Action Per Scene. Both useful, at different units of decomposition.

Step 07 — Use Prompt Modules

A prompt module is a reusable block of instruction. Instead of rewriting the same camera or lighting notes every time, build modules you can copy and paste. Seven module types cover most project work:

• Character identity block
• Camera block
• Lighting block
• Style block
• Motion block
• Negative prompt block
• Continuity block

Example camera block:

"Camera stays in front of the main subject, facing them directly while tracking backward. The subject moves toward camera. Background movement stays behind the subject. Do not reverse the direction."

That one block can save multiple generations. Build your own library of modules — it saves time and keeps your project stable.

Prompt Modules as finer-grained sibling of Identity/Motion. The 7-module taxonomy is a finer-grained sibling of the Identity-vs-Motion separation rule used in single-shot prompts. For Soul ID single-shot work, the 2-block separation in ../higgsfield-soul/SKILL.md § Identity vs. Motion Separation (also ../higgsfield-prompt/SKILL.md § Identity vs. Motion Separation Rule) is sufficient — "Character identity block" here aliases their "Identity Block." The 7-module taxonomy adds a layer for multi-shot projects where camera, lighting, and style each warrant their own reusable block.

Step 08+09 — Fix Failures + Protect What Worked

When a generation fails, do not just say "make it better" — that's not actionable. Tell GPT exactly what went wrong: which specific element is off, what to keep, what to change. Every failure should become a new rule: each diagnosed mistake gets locked into the project's negative-rules list so it doesn't recur.

When fixing, protect what already worked. If 80% of the generation was right, keep that 80%. Only fix the broken part. A lot of people destroy good prompts by changing too much — keep the subject, keep the style, keep the camera if it worked, keep the lighting if it worked. Fix only the mistake.

Full iteration discipline lives elsewhere. The full Change-One-Variable-at-a-Time discipline and the 6-Pass Diagnostic Sequence (Subject / Action / Camera / Style / Audio / Output) for when you don't yet know what's wrong are documented in ../higgsfield-prompt/SKILL.md § The Iteration Rule — Change One Variable at a Time. Step 08+09 here names the three rhetorical handles ("make it better" anti-pattern, 80% rule, "every failure should become a new rule" framing); the mechanics live there. Whether a take is even a regeneration candidate — versus keep, fix-in-post, or edit — is decided by the five-verdict triage in ../higgsfield-troubleshoot/SKILL.md § Take Triage — Five Verdicts for a Delivered Take.

Step 10 — Build the Project in Passes

Do not try to do everything at once. Build the project in clear passes:

• Pass 1: Concept
• Pass 2: Project script
• Pass 3: Scene breakdown
• Pass 4: Shot list
• Pass 5: Image prompts
• Pass 6: Video prompts
• Pass 7: Review results
• Pass 8: Fix and finalize

Each pass produces an artifact the next pass consumes. Pass 1 gives Pass 2 a concept to script; Pass 2 gives Pass 3 a script to break into scenes; and so on. AI moves fast, but fast without structure creates chaos — a workflow keeps it clean.

Tool-level instantiation. Passes 5-7 (image prompts / video prompts / generate) are where the Higgsfield tool chain enters the workflow — see § The Master Production Chain below for the 8-stage tool-level chain (Popcorn → Seedream/Soul → Animate → Recast → Lipsync → Vibe Motion → Upscale → Assemble), and § Pipeline Decision Guide below for the chain choice by project type (Cinematic Short Film, Social Series, Product Campaign, Fast Iteration, Multi-Style Short Film).

Simple Workflow — Execution Recipe

The 10 Steps above are the methodology. The Simple Workflow below is the imperative-action recipe — the same workflow expressed as steps you take in order:

1. Write the project idea.
2. Ask GPT to turn it into a project brief.
3. Ask GPT to break it into scenes.
4. Ask GPT to write a clean script.
5. Ask GPT to turn each scene into image prompts.
6. Ask GPT to turn each scene into video prompts.
7. Generate inside Higgsfield.
8. Review what worked and what failed.
9. Give GPT specific correction notes.
10. Regenerate only what needs fixing.

That is how you move from idea to finished project. At step 7 ("Generate inside Higgsfield"), you choose which tool-chain to run — see § Pipeline Decision Guide below for the chain choice by project type.

Build With Purpose

The power is not in one perfect prompt. The power is in the workflow.

Scripts give the project direction. GPTs help organize and refine the work. Higgsfield brings the visuals to life. When you use all three together, you stop hoping for random good results. You start building with purpose.

The Master Production Chain

This is Higgsfield's complete cinematic workflow — the chain used for short films, branded content, and any multi-shot sequence that needs character continuity:

[1] POPCORN          → Storyboard / key frame images (consistent character + framing)
[2] SEEDREAM / SOUL  → Edit / style the image (transform appearance, fix details)
[3] ANIMATE          → Bring the image to motion (Veo 3.1 / Seedance / Kling)
[4] RECAST           → Swap character if needed (maintain motion, change identity)
[5] LIPSYNC          → Add audio performance (speech, sound, emotion)
[6] VIBE MOTION      → Add motion graphic layers (titles, captions, CTAs)
[7] UPSCALE          → Final quality pass (Topaz upscale for delivery)
[8] ASSEMBLE         → Edit together in your video editor of choice

Not every project uses all 8 stages. Most good short-form content uses 3–5.

Pipeline A: Cinematic Short Film (Full Chain)

Goal: A multi-scene narrative short — character-consistent, story-driven Credits required: High (Pro or Ultimate plan recommended) Time: 2–4 hours for a 30-second sequence

Stage 1 — Storyboard with Popcorn

Generate consistent key frames for every scene before animating anything. This is the planning stage — it's cheap and sets the foundation for everything.

Popcorn prompt structure:
"[Character description — be specific and consistent across all Popcorn prompts].
[Scene — location, time, atmosphere].
[Framing — camera angle, shot size, composition].
[Lighting — source, quality, direction].
[Style — film look, color grade, specific cinematographer reference]."

Key rule: Use the exact same character description in every Popcorn prompt. This is how you get visual continuity across scenes without Soul ID.

Example scene prompts for a single short:

Scene 1 — Establishing:
"A woman, dark hair pulled back, wearing a grey wool coat,
sitting behind the wheel of a moving car. Camera through windshield —
focused and tense expression. Sunlight flickering across her face.
35mm film, shallow depth of field, muted color tones, Roger Deakins style."

Scene 2 — Passenger reaction:
"A grey-haired man in a thick knit sweater, seated in the passenger seat,
gazing out the window with a calm but distant expression.
Camera slightly off-center, interior car shot.
Same 35mm film look, muted tones, soft natural light."

Scene 3 — Object insert:
"Close-up of a weathered wooden photo frame on a kitchen counter.
Inside: a faded photograph of a dark-haired woman and a grey-haired man smiling.
Warm afternoon light through lace curtains, dust motes in air.
50mm lens, shallow focus, nostalgic atmosphere, yellow-green tones."

Stage 2 — Image Editing with Seedream

After Popcorn generates your key frames, use Seedream to make targeted edits that prompt engineering alone can't achieve:

Common Seedream transformations:

• Changing the character's appearance ("make the man look like a zombie")
• Fixing a detail that didn't render correctly
• Adjusting clothing or props
• Adding or removing an element from the frame
• Age progression or style transformation

Seedream edit prompt structure:
"[What to change, specifically]. [What to keep the same]."

Example:
"Make the grey-haired man look like a zombie — rotten flesh, white milky eyes,
grey skin tone. Keep all other elements of the image identical."

Note: Seedream edits the image, not the video. Always edit your Hero Frame before animating — never after.

Stage 3 — Animate by Scene Type

Choose the animation model based on what the scene requires:

Sora 2 held the stunt and epic rows (and shared the vehicle row) until OpenAI shut the Sora 2 API down on 2026-09-24; whether Higgsfield's UI still offers it is unconfirmed. The replacements are the repo's catalog-verified scale/physics fallbacks (../../model-guide.md § Model + Camera Control Compatibility).

I2V animation prompt structure:

"[Starting from the provided image as the first frame.]
[Describe only what MOVES or CHANGES — not what is already visible.]
[Camera — named control.]
[Atmosphere cues — sound, light changes, environmental motion.]
[Style consistency note if needed.]"

Example chain of animation prompts (same short film):

Scene 1 animation (Veo 3.1):
"Camera mounted on the dashboard, focused on the woman driving.
She drives calmly for 2 seconds — then her head turns left sharply.
Her eyes widen in silent horror, lips trembling, color drains from her face.
She doesn't speak — only stares, frozen. Warm sunlight flickers across her face.
Handheld realism, shallow depth of field, cinematic natural lighting."

Scene 2 animation (Veo 3.1):
"Camera fixed behind the car through the rear window.
The man turns his head slowly to the left — confusion spreading.
Expression twists from curiosity to fear. Eyes dart wildly.
He begins trembling and jerking his head, as if losing control.
Claustrophobic tension. Handheld realism, shallow DOF, eerie silence."

Scene 3 animation (written for Sora 2 — retired; rerun on Seedance 2.0 or Minimax Hailuo 2.3):
"A speeding sedan on an empty highway — camera tracking rig, low to ground.
Air shimmers with heat. Car veers — front tire catches rough asphalt.
The car lurches, tilts, flips violently through the air.
Dust, glass, metal fragments scatter. Sun flares in the lens.
Car rotates midair, skids upside down to a stop.
One continuous shot, no cuts. 24mm wide lens, midday, high shutter speed."

Scene 4 animation (Seedance):
"Camera dolly in slowly toward the woman at the kitchen window.
She doesn't move — just stands, looking out. Stillness."

Stage 4 — Recast (Character Swap)

If you need to change a character's identity while keeping the motion exactly:

1. Take your animated video clip from Stage 3
2. Feed it into the Recast app
3. Upload your target character reference image
4. Recast replaces the character while preserving every motion, camera move, and lighting condition from the original clip

Use Recast when:

• You edited the character in Seedream (zombie example) but want to apply that transformation to the already-animated video
• You want the same scene with a completely different character (A/B test)
• You're building an AI Influencer series and need the Soul ID face in motion
• You need to de-identify or replace a character in existing footage

Recast prompt note: Recast is mostly UI-driven — the "prompt" is the reference image you upload. The more cleanly shot the reference face, the better the swap.

Stage 5 — Lipsync / Audio

Add speech or audio performance to any video clip:

Lipsync Studio: Upload video + audio → lips sync to the audio Kling 3.0: Generate video with native audio already embedded (most seamless) Kling Avatars 2.0: Create a talking avatar from a single image — not in the API catalog as of the 2026-09-26 snapshot; may be UI-only — verify in the live UI before recommending. Catalog route: Seedance 1.5 Pro from one start image with native audio (seedance1_5: start_image, generate_audio; the best-lip-sync pick in higgsfield-models § Video Models — Comparison)

When to use each:
- You have existing video + want to add speech → Lipsync Studio
- You're generating new content + need audio → Kling 3.0 (generate with audio from start)
- You want a consistent talking head character → Seedance 1.5 Pro (start image + audio) or Kling Avatars 2.0 (verify in the UI)

Lipsync prompt: Not really a text prompt — upload the video clip and the audio. The system handles sync automatically.

Stage 6 — Vibe Motion Overlay (Optional)

Add motion graphic layers — titles, captions, lower thirds, CTAs — on top of your generated video clips. See higgsfield-vibe-motion skill for full detail.

Common additions at this stage:

• Scene title cards (can be cut in before each scene)
• Character name lower thirds
• Location/time stamp
• End card with CTA, handle, or credit

Stage 7 — Upscale

Run finished clips through Higgsfield's Topaz-integrated upscale before delivery:

• Upscale from 720p to 1080p or 4K
• Sharpens detail lost in generation
• Reduces generation artifacts
• Sora 2 Upscale is for Sora 2 outputs, and Sora 2 is retired (OpenAI shut the Sora 2 API down on 2026-09-24; Higgsfield UI availability unconfirmed), so don't plan new clips around either

Stage 8 — Assembly

Higgsfield doesn't have a native timeline editor — assemble in your editing tool.

Recommended assembly workflow:

1. Export all clips from Higgsfield
2. Bring into DaVinci Resolve / Premiere / CapCut
3. Import Vibe Motion title cards/overlays
4. Add music (Kling 3.0 clips already have audio — other models need music added)
5. Color grade if needed (minimal — Higgsfield output is already color-intentional)
6. Export for platform

Cut-friendly source material: if you're landing morph cuts or smooth cuts in the editor, generate the source clips with explicit 2-second still-or-near- still moments at start and end. See ../higgsfield-cinema/SKILL.md § Multi-Shot Manual → Morph-Cut and Smooth-Cut breathing room.

Pipeline B: Social Content Series (Streamlined Chain)

Goal: Consistent weekly social posts with same character and aesthetic Credits required: Medium (Pro plan) Time: 30–60 minutes per post once Soul ID + Moodboard are set up

[1] SOUL ID          → Character locked once, reused forever
[2] MOODBOARD        → Aesthetic locked once, appended to every prompt
[3] GENERATE IMAGE   → Soul 2.0 or Nano Banana 2 for each post
[4] ANIMATE          → Kling 2.6 I2V, simple motion per post
[5] VIBE MOTION      → Caption/CTA overlay per post format

Per-post prompt template (after Soul ID + Moodboard are established):

[Soul ID character] is [specific action] at [specific location].
[One specific visual detail that changes this post from the last.]
Camera: [simple control — Dolly In / Arc / Overhead].
Aspect: 9:16. Duration: 5s.
[Moodboard style modifier — same every post.]

Example series (3 posts, same character):

Post 1: [Soul ID] sips coffee at a sun-drenched café terrace. Reading a book.
         Camera: Dolly In. Aspect: 9:16.
         Style: warm amber, shallow DOF, golden hour.

Post 2: [Soul ID] walks through a quiet morning market, tote bag on shoulder.
         Camera: slight Arc. Aspect: 9:16.
         Style: warm amber, shallow DOF, golden hour.

Post 3: [Soul ID] sits at a desk by a tall window, writing in a journal.
         Camera: Dolly In. Aspect: 9:16.
         Style: warm amber, shallow DOF, golden hour.

The style modifier is identical every time. Only the scene changes.

Pipeline C: Product Campaign (Commercial Chain)

Goal: Multi-asset product ad campaign — hero video + variants + social cuts Credits required: Medium-High

[1] NANO BANANA PRO  → Perfect product hero image (4K sharp)
[2] MOODBOARD        → Brand visual direction locked
[3] I2V ANIMATION    → Product comes alive (Kling 2.6)
[4] APPS             → Variant generation (Click to Ad, Packshot, Giant Product)
[5] VIBE MOTION      → Feature callouts, lower thirds, price/CTA cards
[6] ASSEMBLE         → Hero cut + 3–5 social cuts

Product hero prompt:

[Product — describe precisely: material, color, form, no brand name].
[Surface/setting — clean, brand-appropriate].
Camera: [Robo Arm / Lazy Susan / Dolly In].
Lighting: [soft diffused / dramatic side-light / macro backlit].
Style: Commercial quality, [clean/warm/dramatic]. [Ratio].

Pipeline D: Fast Iteration / Social Speed Run

Goal: Test 5 creative directions in under an hour Credits required: Low (Basic/Pro)

[1] SEEDANCE 2.0 FAST → Generate 5 fast test clips (one prompt each)
[2] PICK BEST        → Select 1–2 that work
[3] KLING 2.6        → Upgrade the winners to premium quality
[4] VIBE MOTION      → Add captions/CTAs
[5] POST             → Skip the assembly step — single clips, direct export

Pipeline E: Multi-Style Short Film (Soul Cinema + Seedance 2.0 Chain)

Goal: A narrative short where every scene is a completely different animation style (paper cutout, French graphic novel, chibi, stop motion, manga, claymation, 2D-on-live-action, live action) but the hero character and key props stay consistent across all of them.

Credits required: Medium-High (Pro or Business plan). Soul Cinema is cheap (~0.5 credits per 4-image batch); Seedance 2.0 is where the credits go. Time: 2–4 hours for an 8-scene short.

The core trick: feed the previous scene's video + prompt into the next scene's planning step. This is how style can change radically scene-to-scene while character, props, and story continuity hold.

[1] SOUL CINEMA      → Style-first keyframe for this scene (minimal prompt + enhancer)
[2] NANO BANANA PRO  → Targeted edits to match hero's appearance to previous scenes
[3] PROP SHEET       → (Once, at project start) multi-angle reference for recurring objects
[4] CLAUDE           → Write Seedance prompt given keyframe + PREVIOUS scene's prompt
[5] SEEDANCE 2.0     → Animate with keyframe (style ref) + previous video (continuity ref)
[6] REPEAT per scene → Each new scene feeds the previous video back into Claude
[7] ASSEMBLE         → Cut in DaVinci / Premiere / CapCut

Stage 1 — Soul Cinema keyframe (style-first, enhancer ON)

For each new scene, generate a keyframe that defines the style. Prompts here should be deliberately short — 5–15 words — with the Soul Cinema enhancer toggle ON. Short prompts + enhancer explore styles you would never think to ask for explicitly.

Example per-scene keyframe prompts:
- "cartoon highly stylized man waking up in the bedroom"
- "2D French graphic novel style, man in a suit on a horse, cowboy chasing him in a futuristic environment"
- "chibi gladiator toy in a low poly arena"
- "stop motion style man in a black suit, futuristic helicopter behind him"
- "claymation style man in a suit over a black cauldron carried by a red demon"

Style keyword locators: Identify the 1–2 words that control the entire aesthetic — these are the only words you change between scene variants. In the examples above: French graphic novel style + futuristic environment, chibi + low poly, stop motion + futuristic helicopter, claymation + red demon. Swap those and the whole world changes.

Run 2–4 batches per scene, pick the strongest frame.

Stage 2 — Nano Banana Pro: edit, don't regenerate

If Soul Cinema nails the style but the hero drifts (wrong hair color, wrong outfit, face too different from scene 1), do not re-roll. Take the image into Nano Banana Pro and swap only the mismatched elements:

Nano Banana Pro edit prompts:
- "swap hair to black"
- "put him in a black suit"
- "match this character to the reference image, keep the art style"

This is how you get cross-style character consistency — Soul Cinema owns the look, Nano Banana Pro owns the identity.