Practical n8n pipelines and prompt kits for turning ideas into finished AI video without the manual grind.
The problem: Every new brief means re-building the same pipeline by hand. Idea gen, prompt crafting for the model du jour, waiting for renders, downloading, uploading to Drive, logging. Hours disappear before you even start directing.
This repo gives you the ready pieces I now use for client explainers, spec ads, and batch content. Import the workflows, drop in your credentials, tweak the prompts, and the loop runs.
Freelance AI video creators and small studios shipping regular client work or content series. If you find yourself copying prompts between Gemini, Kling, and n8n every week, this is for you.
- One n8n workflow that takes a schedule or manual trigger, generates ideas with Gemini, turns them into model prompts, calls Vertex or Kling via HTTP, waits, decodes the video, uploads to Drive, and logs the link back to Sheets.
- Copy-paste prompt libraries for idea generation, cinematic Veo3, Kling Omni reference sheets, camera and motion control, negatives for drift.
- Decision guide: when to use the full auto vs manual review vs hybrid.
From recent tutorials and community threads:
-
Building the n8n + Veo3 pipeline is fiddly. No native node, so HTTP, token from cloud shell, wait, base64 decode, drive + sheet. The tutorial shows the vision but you still have to wire every node and find the exact phrasing for the agents.
-
Character and motion consistency still requires specific reference sheet structures and prompt tricks even with Omni. 8-shot sheets (4 full body + closeups), labeled refs, correction loops, and "force end on last frame" phrasing.
-
Generic prompts waste credits. You need separated style/character blocks, camera keywords, photorealistic DSLR cues, and model-specific negatives.
-
Scaling to batch or scheduled content means the manual steps kill the fun and the margins.
-
Have n8n running (self-hosted or cloud).
-
Import workflows/video-gen-pipeline.json
-
Set up credentials:
-
Google Gemini for agents
-
HTTP for Vertex AI (project ID, location, access token from cloud shell gcloud auth print-access-token)
-
Google Drive and Google Sheets
-
-
Create the sheet with columns from the JSON (idea, prompt, model, drive_link, etc.)
-
Run.
See docs/setup-credentials.md for exact fields and regional notes.
-
Full idea to video pipeline (schedule + Gemini idea + prompt agent + HTTP gen + wait + decode + drive + sheet)
-
Review loop extension (for when you want to pause for human pick before final render)
In prompts/ :
-
idea-agent-system.md : structured JSON output for chaining
-
veo3-cinematic-prompts.md : style + action separation for commercial
-
kling-omni-ref-prompts.md : how to label and call reference images, end scene trick
-
camera-motion-control.md : specific keywords that move the camera without drift
-
negative-prompts-drift.md : what to put in negative to lock face/clothing
-
reference-sheet-generation.md : how to generate the 8 angle sheets that actually work
Real briefs with the prompts used, what the output looked like, what still needed hand fixes.
-
brand-ad-pipeline.md (luxury product)
-
client-explainer-example.md
-
faceless-batch.md
-
before-after-notes.md
-
setup-credentials.md
-
decision-tree.md : n8n auto vs manual vs agent vs other tools
-
cost-and-access.md : free credits, Vertex in India, Kling pricing notes
For a recent explainer, I scheduled the idea agent to run overnight with 5 seeds. Next morning I had 5 options in Sheets with drive links. Picked two, tweaked the prompt agent output for the client's brand palette, ran the single model pass, client reviewed in Telegram, final high res render went straight to their folder.
The automation handles the waiting and the file dance. I spend the time on the creative choices and the one or two frames that still need a fix.
Kling 3.0 Native Audio: Now includes workflows for generating synchronized audio with video. Separate prompts for SFX, music, and dialogue.
Veo 3.1 120-Second Clips: Extended duration support. New "long-form" workflow for narrative content up to 2 minutes.
Runway Gen-4.5 API: Added HTTP nodes for Runway's new API. Character consistency workflow with built-in lip-sync for dialogue scenes.
Wan 2.1 Self-Hosted: New ComfyUI integration node for local Wan generation. Privacy-focused workflows for sensitive client projects.
LTX 2.3 ID-LoRA: Character LoRA training workflow — generate consistent characters without manual reference management.
Pika 2.5: Updated nodes for Pika's latest model with improved text rendering.
Higgsfield Canvas: Node-based multi-model workflow example (visual pipeline for team collaboration).
Complex choreography with multiple interacting characters can still drift. The new Runway Gen-4.5 and Veo 3.1 help significantly, but I still budget time for 1-2 correction passes on hero shots.
Vertex credits remain generous but Veo 3.1 Full tier is required for native audio — budget accordingly.
The n8n json here includes the August 2026 updates: better error handling, LoRA integration nodes, and multi-model fallbacks.
| Model | Seconds per credit | Cost per 10s clip | Best for |
|---|---|---|---|
| Omni Flash | ~8s | $0.08 | Budget motion graphics |
| Kling 3.0 | ~6s | $0.15 | High-end cinematic |
| Veo 3.1 Full | ~4s | $0.25 | Native audio + lip-sync |
| Runway Gen-4 | ~5s | $0.20 | Character consistency |
| Wan 2.1 (local) | Unlimited (GPU) | $0 | High-volume batch |
| LTX 2.3 | ~6s | $0.15 | LoRA consistency |
Prices approximate based on mid-tier subscriptions. See docs/pricing-august-2026.md for details.
-
n8n-ai-creator-pipelines for the intake and review side
-
CharacterConsistency for the reference sheet bible
-
cinematic-ai-prompts for the prompt library
MIT. Use in your studio or client projects. If it saves you a day, a star or a mention is appreciated but not required.
Built by Sagarika Sultana.
Last updated: July 2026